Initial release of Luciole-8B-Base

Co-authored-by: Jérôme Louradour <Jeronymous@users.noreply.huggingface.co>
Co-authored-by: Olivier Gouvert <oligou@users.noreply.huggingface.co>
Co-authored-by: Julie Hunter <juliehunter@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,41 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+metadata/evaluation/en_details.png filter=lfs diff=lfs merge=lfs -text
+metadata/evaluation/fr_details.png filter=lfs diff=lfs merge=lfs -text
+metadata/evaluation/multilingual_details.png filter=lfs diff=lfs merge=lfs -text
+metadata/evaluation/ruler_details.png filter=lfs diff=lfs merge=lfs -text
+metadata/evaluation/translation_details.png filter=lfs diff=lfs merge=lfs -text
+metadata/training_logs/ConvergenceCurve.png filter=lfs diff=lfs merge=lfs -text

LICENSE.md

@@ -0,0 +1,237 @@
+---
+title: Apache License 2.0
+spdx-id: Apache-2.0
+redirect_from: /licenses/apache/
+featured: true
+hidden: false
+
+description: A permissive license whose main conditions require preservation of copyright and license notices. Contributors provide an express grant of patent rights. Licensed works, modifications, and larger works may be distributed under different terms and without source code.
+
+how: Create a text file (typically named LICENSE or LICENSE.txt) in the root of your source code and copy the text of the license into the file.
+
+note: The Apache Software Foundation <a href="https://apache.org/foundation/license-faq.html#Apply-My-Software">recommends</a> taking the additional step of adding a boilerplate notice to the header of each source file. You can find the notice in the appendix at the very end of the license text.
+
+using:
+  Kubernetes: https://github.com/kubernetes/kubernetes/blob/master/LICENSE
+  PDF.js: https://github.com/mozilla/pdf.js/blob/master/LICENSE
+  Swift: https://github.com/apple/swift/blob/main/LICENSE.txt
+
+permissions:
+  - commercial-use
+  - modifications
+  - distribution
+  - patent-use
+  - private-use
+
+conditions:
+  - include-copyright
+  - document-changes
+
+limitations:
+  - trademark-use
+  - liability
+  - warranty
+
+---
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -0,0 +1,391 @@
+---
+license: apache-2.0
+pipeline_tag: text-generation
+language:
+- fr
+- en
+- it
+- de
+- es
+- pt
+- nl
+- ar
+tags:
+- pretrained
+- openllm-france
+datasets:
+  - OpenLLM-France/Luciole-Training-Dataset
+training_progress:
+  num_steps: 1216796
+  num_tokens: 5103612329984
+  context_length: 131072
+widget:
+  - text: |-
+      Quelle est la capitale de l'Espagne ? Madrid.
+      Quelle est la capitale de la France ?
+    example_title: Capital cities in French
+    group: 1-shot Question Answering
+---
+
+
+# Model Card for Luciole-8B-Base
+
+<!-- inspired from the following template:
+https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md?plain=1
+-->
+
+![luciole_logo.png](luciole_logo.png)
+
+
+**Table of Contents**
+
+- [Model Card for Luciole-8B-Base](#model-card-for-luciole-8b-base)
+  - [Model Description](#model-description)
+  - [Uses](#uses)
+    - [Direct use](#direct-use)
+    - [Downstream use](#downstream-use)
+    - [Out-of-Scope Use](#out-of-scope-use)
+  - [Bias, Risks, and Limitations](#bias-risks-and-limitations)
+    - [Recommendations](#recommendations)
+  - [Example Code in Python](#example-code-in-python)
+    - [Load the model](#load-the-model)
+    - [Sentence completion](#sentence-completion)
+  - [Loading Intermediate Checkpoints](#loading-intermediate-checkpoints)
+  - [Training Details](#training-details)
+    - [Training Data](#training-data)
+    - [Training Procedure](#training-procedure)
+      - [Neural Network Architecture](#neural-network-architecture)
+      - [Training Hyperparameters](#training-hyperparameters)
+    - [Training Convergence and Evaluation](#training-convergence-and-evaluation)
+      - [Training loss](#training-loss)
+      - [Evaluation](#evaluation)
+  - [Citation](#citation)
+  - [Acknowledgements](#acknowledgements)
+  - [Contact](#contact)
+
+## Model Description
+
+Luciole-8B-Base is a pretrained 8B parameter causal language model with a hybrid Mamba-transformer architecture. It was developed by [LINAGORA](https://labs.linagora.com/) and the [OpenLLM-France](https://openllm-france.fr/) consortium as a part of the OpenLLM France project, funded by [BPI France](https://www.bpifrance.fr/) through the [France 2030](https://www.info.gouv.fr/grand-dossier/france-2030) program.
+It was trained on the [GENCI](https://www.genci.fr/) supercomputer Jean Zay, managed by [IDRIS](http://www.idris.fr/docs/idris/missions).
+
+Luciole-8B-Base was trained on around 5 trillion tokens of multilingual data, including English (41.9%), French (30.4%), German (3.8%), Spanish (3.5%), Italian (1.9%), Portuguese (1.3%), Dutch (1.0%), Arabic (0.5%), and a small subset of regional languages including regional languages of the French metropolitan area, French variants, and French creoles from around the world (0.4%). 
+
+The latter were selected from the [FineWeb 2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2) dataset and include Basque, Breton, Catalan, Corsican, Franco-Provençal, Guadeloupean Creole French, Guianese Creole French, Occitan, Picard, Réunion Creole French, Saint Lucian Creole French, Seselwa Creole French, Tahitian, and Walloon.
+
+Training data also include parallel data from a selection of languages (1.7%), as well as several programming languages (9.2%) and English mathematical data (3.5%).
+
+* License: [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
+* Training repository: [Luciole-Training](https://github.com/OpenLLM-France/Luciole-Training)
+* Technical report: coming soon.
+
+
+
+## Uses
+
+### Direct use
+Luciole-8B-Base is a foundation language model trained solely to predict the most probable next word in a sequence. It is designed as the first brick in a more complex training pipeline that would include multitask training on diverse instructions or focused fine-tuning on select downstream tasks, as well as possible alignment for human preferences.
+
+### Downstream use 
+Due to its multilingual training, Luciole-8B-Base can be fine-tuned for downstream tasks centered on the generation of multilingual text, with a special focus on French and English. 
+
+### Out-of-Scope Use
+Luciole-8B-Base is not intended to generate text directly for end use cases. It must be fine-tuned first. Its pretraining is optimized for multilingual performance, especially in French and English, and might perform less well on other languages without additional training. While trained on code data, it is not optimized for code generation tasks.
+
+## Bias, Risks, and Limitations
+
+Like other foundation models, Luciole-8B-Base is trained on large amounts of web data. Additionally, due to the scarcity of French textual non-web data published under open licenses, much of our French data comes from older works in the public domain that carry biases from other time periods. While we made efforts to reduce toxic and offensive content in the [Luciole Training Dataset](https://huggingface.co/datasets/OpenLLM-France/Luciole-Training-Dataset), Luciole-8B-Base may still generate such content. Filtering of the Luciole Training Dataset is an ongoing project to which we welcome contributions. 
+
+### Recommendations
+To limit the generation of undesirable content, it is advised to fine-tune Luciole-8B-Base through instruction and preference tuning (DPO, RLHF, etc.).
+
+## Example Code in Python
+
+### Load the model
+
+Load the model (quantized version on GPU if possible, for efficient inference):
+```python
+import transformers
+
+model_name = "OpenLLM-France/Luciole-8B-Base"
+
+tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
+model = transformers.AutoModelForCausalLM.from_pretrained(model_name,
+    device_map="auto",
+    load_in_4bit=True       # For efficient inference, if quantization is supported by the GPU card
+)
+```
+### Sentence completion
+
+Wrap the model in a text generation pipeline, and specify some generation parameters:
+```
+pipeline = transformers.pipeline("text-generation", model=model, tokenizer=tokenizer)
+
+generation_kwargs = dict(
+    num_return_sequences=1,               # Number of variants to generate.
+    return_full_text= False,              # Do not include the prompt in the generated text.
+    do_sample=True,
+    temperature=1.0, top_p=1, top_k=None, # Sampling parameters.
+    max_new_tokens=200,                   # Maximum length for the output text (in number of tokens).
+)
+```
+
+Try 1-shot question answering:
+```python
+prompt = """\
+Quelle est la capitale de l'Espagne ? Madrid\n\
+Quelle est la capitale de la France ?\
+"""
+completions = pipeline(prompt, **generation_kwargs)
+for completion in completions:
+    print(prompt + "[…]" + completion['generated_text'])
+```
+This will print something like:
+```
+Quelle est la capitale de l'Espagne ? Madrid
+Quelle est la capitale de la France ?[…] Paris
+Quelle est la capitale du Brésil ? Brasilia
+Quelle est la capitale de la Belgique ? Bruxelles
+Quelle est la capitale de l'Italie ? Rome
+...
+```
+
+
+## Loading Intermediate Checkpoints
+
+Intermediate checkpoints are released under dedicated revision tags at regular intervals throughout training:
+- Every **1,000 steps** during the first **5,000 steps**  
+- Then every **5,000 steps** up to **30,000 steps**  
+- Then every **10,000 steps** beyond that  
+- In addition, a checkpoint is provided at the **end of each training phase**
+
+All checkpoints are available at:  
+https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/
+
+They are organized into the following subfolders:
+* **Phase 1 – Initial pretraining (context length: 4,096)**  
+  From [phase1-step0001000](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase1-step0001000) to [phase1-step0715787](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase1-step0715787)
+* **Phase 2 – Continued pretraining**  
+  From [phase2-step0010000](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase2-step0010000) to [phase2-step0358930](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase2-step0358930)
+* **Phase 3 – Annealing phase**  
+  From [phase3-annealing-step0010000](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase3-annealing-step0010000) to [phase3-annealing-step0118238](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase3-annealing-step0118238)
+* **Phase 4 – Context extension to 131k tokens**  
+  [phase4-context-extension-131k-step0003000](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase4-context-extension-131k-step0003000) to [phase4-context-extension-131k-step0023841](https://dl.labs.linagora.com/files/models/OpenLLM-France/Luciole-8B-Base/phase4-context-extension-131k-step0023841)
+
+The total cumulative number of training steps and training tokens for each checkpoint is specified in
+the YAML header of each `README.md` file, and in
+the `config.json` file (under the keys `"training_steps"` and `"training_tokens"`).
+
+## Training Details
+
+### Training Data
+
+The training dataset used for the pretraining of Luciole-8B-Base is available
+at [OpenLLM-France/Luciole-Training-Dataset](https://huggingface.co/datasets/OpenLLM-France/Luciole-Training-Dataset). Information on data preprocessing can be found on the data card or in the [Luciole-Training](https://github.com/OpenLLM-France/Luciole-Training) repository.
+
+Pretraining consisted of three principal phases of training with a context length of 4,096 tokens. The token breakdowns for the three phases are as follows:
+
+1. Initial pretraining: 3.5 trillion tokens of diverse data
+2. Continued pretraining: 1 trillion tokens introducing higher quality data and increasing math and code proportions
+3. Annealing phase: 0.5 trillion tokens introducing some instruction-style and reasoning data
+
+Pretraining was followed by one short mid-training phases to extend the context length to 131,072 tokens:
+
+4. Context extension: 100 billion tokens to extend context length from 4,096 to 131,072 tokens
+
+
+### Training Procedure 
+
+Luciole-8B-Base is a causal decoder-only model trained on a causal language modeling task (i.e., predict the next token).
+
+It was pre-trained on 128 - 256 H100 80GB GPUs (32 - 64 nodes) for about 237,036 GPU hours (870 hours) on the [Jean Zay supercomputer](http://www.idris.fr/docs/category/jean-zay).
+
+The training code is available at [https://github.com/OpenLLM-France/Luciole-Training](https://github.com/OpenLLM-France/Luciole-Training). Training used version 2.3.1 of NVIDIA's [NeMo framework](https://github.com/NVIDIA-NeMo/NeMo) (now transferred to [NeMo Automodel or NeMo Megatron-Bridge](https://github.com/NVIDIA-NeMo)).
+
+
+
+#### Neural Network Architecture
+
+The architecture of Luciole-8B-Base is a custom adaptation of the [NemotronH-8B](https://github.com/NVIDIA-NeMo/NeMo/blob/v2.3.1/nemo/collections/llm/recipes/nemotronh_8b.py) recipe.
+It has exactly 8.08 billion free parameters,
+with the following hyperparameters:
+| **Hyperparameter**        | **Value** |
+|---------------------------|---------|
+| Vocabulary size (\# tokens)| 128,000 |
+| \# blocks                 |      52 |
+| \# attention heads        |      32 |
+| \# key-value heads        |       8 |
+| Hidden size               |   4096 |
+| Intermediate size         |  21504 |
+| SSM State size            |  128 |
+| Mamba \# heads            |  128 |
+| MLP Activation            |  `relu2`|
+| Mamba Activation          |  `silu`|
+
+
+#### Training Hyperparameters
+
+The details of the intitial pretraining phase are listed below. For each subsequent phase, only the values that differ from the intitial pretraining phase are listed. 
+
+**1. Initial pretraining**
+
+| **Hyperparameter**     | **Value**  |
+|------------------------|------------|
+| Total \# samples| 732,965,888 (3T tokens) |
+| Total \# steps  | 715,787  |
+| Context length         | 4,096      |
+| Batch size       | 1,024      |
+| Learning rate schedule | Warmup (2M samples) + Constant  |
+| Learning rate  | 3e-4       |
+| Weight decay           | 0.1        |
+| Dropout                | _          |
+| Gradient clipping      | 1          |
+| Initializer range      | 0.009        |
+| Optimizer              | `AdamW` (β₁=0.9, β₂=0.95, ε=1e-5)    |
+| Precision              | `bfloat16` |
+| Tensor Parallelism (with 256 GPUs)   | 2           |
+| Pipeline Parallelism (with 256 GPUs) | 1           |
+| Data Parallelism (with 256 GPUs)     | 128         |
+
+**2. Continued Pretraining**
+
+| **Hyperparameter**     | **Value**  |
+|------------------------|------------|
+| Total \# samples| 367,545,344 (1.5T tokens) |
+| Total \# steps  | 358,931      |
+| Learning rate schedule | Cosine annealing  |
+| Maximum Learning rate  | 3e-4 |
+| Final Learning rate  | 6.87e-5 |
+
+
+**3. Annealing**
+
+| **Hyperparameter**     | **Value**  |
+|------------------------|------------|
+| Total \# samples | 121,075,712  (0.5T tokens) |
+| Total \# steps  | 118,238     |
+| Learning rate schedule | Linear annealing |
+| Maximum Learning rate  | 6.87e-5    |
+| Final Learning rate    | 0          |
+
+**4. Context extension to 131K**
+
+| **Hyperparameter**     | **Value**  |
+|------------------------|------------|
+| Total \# samples| 762,912 (100B tokens) |
+| Total \# steps  | 23,841|
+| Context length         | 131,072 |
+| Batch size             | 32 |
+| Context Parallelism (with 128 GPUs)  | 8          |
+| Tensor Parallelism (with 128 GPUs)   | 2          |
+| Data Parallelism (with 128 GPUs)     | 8          |
+
+
+### Training Convergence and Evaluation
+
+#### Training loss
+
+Information on training loss curves and training stability is available in the training logs, which are released at<br>
+[metadata/training_logs](https://huggingface.co/OpenLLM-France/Luciole-8B-Base/tree/main/metadata/training_logs)<br>
+├── [ConvergenceCurve_phase1.csv](metadata/training_logs/ConvergenceCurve_phase1.csv) -- training logs for phase 1 (initial pretraining) <br>
+├── [ConvergenceCurve_phase2.csv](metadata/training_logs/ConvergenceCurve_phase2.csv) -- training logs for phase 2 (continued pretraining) <br>
+├── [ConvergenceCurve_phase3-annealing.csv](metadata/training_logs/ConvergenceCurve_phase3-annealing.csv) -- training logs for phase 3 (annealing) <br>
+└── [ConvergenceCurve_phase4-context-extension-131k.csv](metadata/training_logs/ConvergenceCurve_phase4-context-extension-131k.csv) -- training logs for phase 4 (context extension to 131k tokens) <br>
+
+The following figure shows the training loss curve for the different training phases:
+![metadata/training_logs/ConvergenceCurve.png](metadata/training_logs/ConvergenceCurve.png)
+
+#### Evaluation
+
+During the training of Luciole-8B-Base, we conducted multiple evaluations to assess performance on standard benchmarks.
+The primary evaluation languages were French and English, with additional evaluations in German, Spanish, Italian, Portuguese, Dutch, and Arabic.
+These evaluations were performed on intermediate checkpoints throughout training and on the final checkpoint.
+We also evaluated the final checkpoint using the RULER benchmark to measure long-context performance.
+
+For comparison, we evaluated the model against the following models, which have between 8B and 2B parameters and were trained on multilingual data, including French:
+- [Gaperon-1125-8B](https://huggingface.co/almanach/Gaperon-1125-8B)
+- [EuroLLM-9B](https://huggingface.co/utter-project/EuroLLM-9B)
+- [salamandra-7b](https://huggingface.co/BSC-LT/salamandra-7b)
+- [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B)
+- [Apertus-8B-2509](https://huggingface.co/swiss-ai/Apertus-8B-2509)
+- [Qwen3-8B-Base](https://huggingface.co/Qwen/Qwen3-8B-Base)
+- [Ministral-3-8B-Base-2512](https://huggingface.co/mistralai/Ministral-3-8B-Base-2512)
+
+The main results are summarized in the figures below.
+**Click on a figure to view the complete evaluation results with detailed metrics.**
+The figures illustrate the evolution of evaluation performance across training checkpoints.
+The x-axis corresponds to the cumulative number of training tokens, while the y-axis reports the average performance across tasks within each category.
+The RULER figure differs in that it reports average performance across different context lengths.
+
+<table>
+<tr>
+<td><a href="https://huggingface.co/OpenLLM-France/Luciole-8B-Base/blob/main/metadata/evaluation/fr_details.png"><img src="metadata/evaluation/fr_average.png" alt="French average results"></a></td>
+<td><img src="metadata/evaluation/legend.png" alt="Legend"></td>
+</tr>
+<tr>
+<td><a href="https://huggingface.co/OpenLLM-France/Luciole-8B-Base/blob/main/metadata/evaluation/en_details.png"><img src="metadata/evaluation/en_average.png" alt="English average results"></a></td>
+<td><a href="https://huggingface.co/OpenLLM-France/Luciole-8B-Base/blob/main/metadata/evaluation/multilingual_details.png"><img src="metadata/evaluation/multilingual_average.png" alt="Multilingual average results"></a></td>
+</tr>
+<tr>
+<td><a href="https://huggingface.co/OpenLLM-France/Luciole-8B-Base/blob/main/metadata/evaluation/translation_details.png"><img src="metadata/evaluation/translation_average.png" alt="Translation average results"></a></td>
+<td><a href="https://huggingface.co/OpenLLM-France/Luciole-8B-Base/blob/main/metadata/evaluation/ruler_details.png"><img src="metadata/evaluation/ruler_average.png" alt="RULER average results"></a></td>
+</tr>
+</table>
+
+All figures can be found in the [metadata/evaluation](https://huggingface.co/OpenLLM-France/Luciole-8B-Base/tree/main/metadata/evaluation) folder.
+
+## Citation
+
+✍ Paper coming soon!
+
+
+## Acknowledgements
+
+We gratefully acknowledge BPI France for funding the OpenLLM France project under the [call](https://www.bpifrance.fr/nos-appels-a-projets-concours/appel-a-projets-communs-numeriques-pour-lintelligence-artificielle-generative) "Communs numériques pour l’intelligence artificielle générative" ("Digital commons for generative artificial intelligence") as a part of the [France 2030](https://www.info.gouv.fr/grand-dossier/france-2030) program.
+
+Training of Luciole-8B-Base was made possible by computing AI and storage resources by GENCI at IDRIS thanks to the grant 2025-AS011016445 on the supercomputer Jean Zay’s H100 partition. We gratefully acknowledge support from GENCI and IDRIS and from Stephane Requena (GENCI) and Pierre-François Lavallée (IDRIS) in particular. 
+
+Luciole-8B-Base was created by members of [LINAGORA](https://labs.linagora.com/) for the OpenLLM-France project, including in alphabetical order:  
+
+Audran Bert  
+Akshay Chaturvedi  
+Olivier Gouvert  
+Julie Hunter  
+Jean-Pierre Lorré  
+Jérôme Louradour  
+Charlotte Noel   
+Kate Thompson   
+
+We thank the support team from IDRIS for technical guidance throughout the project, especially:  
+
+
+Martin Comminges (IDRIS)  
+Rémi Lacroix (IDRIS)     
+Myriam Peyrounette (IDRIS)  
+
+as well as the support team from NVIDIA, especially:
+
+Meriem Bendris (NVIDIA)   
+Hayk Shoukourian (NVIDIA)    
+Oleg Sudakov (NVIDIA)
+
+We are also greatful to the partners of the [OpenLLM-France](https://www.openllm-france.fr/) consortium for their valuable input, with particular thanks to (in alphabetical order):  
+  
+Clément Bénesse (Opsci)    
+Bertrand Cabot (IDRIS)  
+Christophe Cerisara (LORIA)    
+Liam Duignan (CEA)   
+Olivier Ferret (CEA)    
+Emile Hazard (OpSci)  
+Léo Hunout (IDRIS)  
+Gabriel Lauzzana (LORIA)      
+Michel-Marie Maudet (LINAGORA)  
+
+
+We would also like to thank members of the [Gaperon](https://huggingface.co/collections/almanach/gaperon), [Salamandra](https://huggingface.co/collections/BSC-LT/salamandra) and [Apertus](https://huggingface.co/collections/swiss-ai/apertus-llm) projects for sharing their insights with us. We also acknowledge the numerous open source actors whose resources have guided us throughout the training process, with particular thanks to [Nvidia](https://www.nvidia.com/en-eu/), [Hugging Face](https://huggingface.co/) and [Allen AI](https://allenai.org/). 
+
+
+Finally, we thank the entire OpenLLM-France community, whose members have helped in diverse ways. 
+
+## Contact
+
+contact@openllm-france.fr

config.json

@@ -0,0 +1,55 @@
+{
+  "architectures": [
+    "NemotronHForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attention_head_dim": 128,
+  "bos_token_id": 0,
+  "chunk_size": 256,
+  "conv_kernel": 4,
+  "eos_token_id": 1,
+  "expand": 2,
+  "hidden_dropout": 0.0,
+  "hidden_size": 4096,
+  "hybrid_override_pattern": "M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M-",
+  "initializer_range": 0.02,
+  "intermediate_size": 21504,
+  "layer_norm_epsilon": 1e-05,
+  "mamba_head_dim": 64,
+  "mamba_hidden_act": "silu",
+  "mamba_num_heads": 128,
+  "mamba_proj_bias": false,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "mlp_hidden_act": "relu2",
+  "model_type": "nemotron_h",
+  "n_groups": 8,
+  "num_attention_heads": 32,
+  "num_hidden_layers": 52,
+  "num_key_value_heads": 8,
+  "num_logits_to_keep": 1,
+  "pad_token_id": 0,
+  "rescale_prenorm_residual": true,
+  "residual_in_fp32": false,
+  "sliding_window": null,
+  "ssm_state_size": 128,
+  "tie_word_embeddings": false,
+  "time_step_floor": 0.0001,
+  "time_step_max": 0.1,
+  "time_step_min": 0.001,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_bias": false,
+  "use_cache": true,
+  "use_conv_bias": true,
+  "use_mamba_kernels": true,
+  "vocab_size": 128000,
+  "auto_map": {
+    "AutoConfig": "configuration_nemotron_h.NemotronHConfig",
+    "AutoModelForCausalLM": "modeling_nemotron_h.NemotronHForCausalLM"
+  },
+  "rms_norm_eps": 1e-05,
+  "training_steps": 1216796,
+  "training_tokens": 5103612329984
+}

configuration_nemotron_h.py

@@ -0,0 +1,251 @@
+# coding=utf-8
+# Copyright 2024 AI21 Labs Ltd. and the HuggingFace Inc. team. All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""NemotronH model configuration"""
+
+import re
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class NemotronHConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`NemotronHModel`]. It is used to instantiate a
+    NemotronH model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of the NemotronH-v0.1 model.
+
+    [todo](todo)
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 131072):
+            Vocabulary size of the NemotronH model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`NemotronHModel`]
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied. Note that this is only relevant if the
+            model has a output word embedding layer.
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 21504):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 52):
+            Number of hidden layers in the Transformer encoder.
+        hybrid_override_pattern (`str`, *optional*, defaults to `"M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M-"`):
+            The pattern of the hybrid model. The pattern is a string of characters where each character represents M: Mamba2, *: Attention, -: MLP
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        attention_head_dim (`int`, *optional*, defaults to 128):
+            Dimension of each attention head.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used.
+        mlp_hidden_act (`str`, *optional*, defaults to "relu2"):
+            The non-linear activation function in the MLP layers.
+        attention_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in attention layers.
+        mlp_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in MLP layers.
+        use_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in the model.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        residual_in_fp32 (`bool`, *optional*, defaults to `False`):
+            Whether or not residuals should be in `float32`. If set to `False` residuals will keep the same `dtype` as the rest of the model.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        num_logits_to_keep (`int` or `None`, *optional*, defaults to 1):
+            Number of prompt logits to calculate during generation. If `None`, all logits will be calculated. If an
+            integer value, only last `num_logits_to_keep` logits will be calculated.
+        pad_token_id (`int`, *optional*, defaults to 0):
+            The id of the padding token.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            The id of the "end-of-sequence" token.
+        sliding_window (`int`, *optional*, defaults to None):
+            Sliding window attention window size.
+        max_position_embeddings (`int`, *optional*, defaults to 4096):
+            The maximum sequence length that this model might ever be used with.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        hidden_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the hidden states.
+        use_mamba_kernels (`bool`, *optional*, defaults to `True`):
+            Flag indicating whether or not to use the fast mamba kernels. These are available only if `mamba-ssm` and
+            `causal-conv1d` are installed, and the mamba modules are running on a CUDA device.
+        ssm_state_size (`int`, *optional*, defaults to 128):
+            The dimension of the mamba state space latents.
+        mamba_num_heads (`int`, *optional*, defaults to 128):
+            Number of heads in Mamba layers.
+        mamba_n_groups (`int`, *optional*, defaults to 8):
+            Number of groups in Mamba layers.
+        mamba_head_dim (`int`, *optional*, defaults to 64):
+            Dimension of each Mamba head.
+        mamba_d_conv (`int`, *optional*, defaults to 4):
+            The size of the mamba convolution kernel.
+        mamba_expand (`int`, *optional*, defaults to 2):
+            Expanding factor used to determine the mamba intermediate size.
+        mamba_hidden_act (`str`, *optional*, defaults to "silu"):
+            The non-linear activation function in the Mamba layers.
+        mamba_dt_min (`float`, *optional*, defaults to 0.001):
+            Minimum value for the time step in Mamba.
+        mamba_dt_max (`float`, *optional*, defaults to 0.1):
+            Maximum value for the time step in Mamba.
+        mamba_dt_limit (`tuple`, *optional*, defaults to (0.0, float("inf"))):
+            Limits for the time step in Mamba.
+        mamba_dt_init_floor (`float`, *optional*, defaults to 1e-4):
+            Floor value for time step initialization in Mamba.
+        mamba_conv_bias (`bool`, *optional*, defaults to `True`):
+            Whether to use bias in the convolution layer of the mamba mixer block.
+        mamba_proj_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in the input and output projections of the mamba mixer block.
+        mamba_chunk_size (`int`, *optional*, defaults to 256):
+            Size of chunks for Mamba processing.
+        rescale_prenorm_residual (`bool`, *optional*, defaults to `True`):
+            Whether to rescale the pre-normalization residual connections.
+    """
+
+    model_type = "nemotron_h"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=131072,
+        tie_word_embeddings=False,
+        hidden_size=4096,
+        intermediate_size=21504,
+        num_hidden_layers=52,
+        hybrid_override_pattern="M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M-",
+        num_attention_heads=32,
+        attention_head_dim=128,
+        num_key_value_heads=8,  # nemo: num_query_groups
+        mlp_hidden_act="relu2",
+        attention_bias=False,
+        mlp_bias=False,
+        use_bias=False,
+        initializer_range=0.02,  # nemo: init_method_std
+        layer_norm_epsilon=1e-5,  # nemo: layernorm_epsilon
+        residual_in_fp32=False,  #  Megatron Core default value
+        use_cache=True,
+        num_logits_to_keep=1,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        sliding_window=None,
+        max_position_embeddings=4096,
+        attention_dropout=0.0,
+        hidden_dropout=0.0,  # * ADDED
+        use_mamba_kernels=True,
+        ssm_state_size=128,  # mamba_state_size
+        mamba_num_heads=128,
+        mamba_n_groups=8,  # nemo: mamba_ssm_ngroups = num_heads
+        mamba_head_dim=64,
+        mamba_d_conv=4,
+        mamba_expand=2,
+        mamba_hidden_act="silu",
+        mamba_dt_min=0.001,
+        mamba_dt_max=0.1,
+        mamba_dt_limit=(0.0, float("inf")),
+        mamba_dt_init_floor=1e-4,
+        mamba_conv_bias=True,
+        mamba_proj_bias=False,
+        mamba_chunk_size=256,
+        rescale_prenorm_residual=True,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.tie_word_embeddings = tie_word_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.hybrid_override_pattern = hybrid_override_pattern
+        self.num_attention_heads = num_attention_heads
+        self.attention_head_dim = attention_head_dim
+        self.sliding_window = sliding_window
+        self.max_position_embeddings = max_position_embeddings
+        self.attention_dropout = attention_dropout
+        self.hidden_dropout = hidden_dropout
+
+        # Validate hybrid_override_pattern
+        # M: Mamba2, *: Attention, -: MLP
+        assert (
+            len(self.hybrid_override_pattern) == self.num_hidden_layers
+        ), "hybrid_override_pattern must have the same length as num_hidden_layers"
+        assert re.match(
+            r"^[*-M]+$", self.hybrid_override_pattern
+        ), "hybrid_override_pattern must only contain characters 'M', '*', or '-'"
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.mlp_hidden_act = mlp_hidden_act
+        self.attention_bias = attention_bias
+        self.mlp_bias = mlp_bias
+        self.use_bias = use_bias
+        self.initializer_range = initializer_range
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.residual_in_fp32 = residual_in_fp32
+
+        self.use_cache = use_cache
+        self.num_logits_to_keep = num_logits_to_keep
+
+        self.use_mamba_kernels = use_mamba_kernels
+        self.n_groups = mamba_n_groups
+        self.mamba_head_dim = mamba_head_dim
+        self.ssm_state_size = ssm_state_size
+        self.mamba_num_heads = mamba_num_heads
+        self.conv_kernel = mamba_d_conv
+        self.expand = mamba_expand
+        self.mamba_hidden_act = mamba_hidden_act
+        self.time_step_min = mamba_dt_min
+        self.time_step_max = mamba_dt_max
+        self.time_step_limit = mamba_dt_limit
+        self.time_step_floor = mamba_dt_init_floor
+        self.use_conv_bias = mamba_conv_bias
+        self.mamba_proj_bias = mamba_proj_bias
+        self.chunk_size = mamba_chunk_size
+        self.rescale_prenorm_residual = rescale_prenorm_residual
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    @property
+    def layers_block_type(self):
+        return [
+            "mamba"
+            if self.hybrid_override_pattern[i] == "M"
+            else "attention"
+            if self.hybrid_override_pattern[i] == "*"
+            else "mlp"
+            for i in range(self.num_hidden_layers)
+        ]

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "do_sample": true,
+  "temperature": 0.7,
+  "transformers_version": "4.51.3"
+}

metadata/training_logs/ConvergenceCurve_phase1.csv

@@ -0,0 +1,370 @@
+training_steps,training_tokens,training_loss,walltime,gputime,learning_rate
+1290,5410652160,5.970873527526855,1916.6399681568146,490659.83184814453,0.00019340329163242131
+3265,13694402560,3.2604511642456053,4848.888065814972,1241315.3448486328,0.0003000000142492354
+4575,19188940800,2.888548288345337,6652.545394897461,1703051.62109375,0.0003000000142492354
+7055,29590814720,2.7221426296234132,10465.404177904129,2679143.469543457,0.0003000000142492354
+8895,37308334080,2.616291103363037,13140.052797794342,3363853.5162353516,0.0003000000142492354
+11670,48947527680,2.536383981704712,17838.381856679916,4566625.755310059,0.0003000000142492354
+13505,56644075520,2.4767827796936035,20365.318793058395,5213521.611022949,0.0003000000142492354
+15805,66290974720,2.435423879623413,24113.398501873016,6173030.016479492,0.0003000000142492354
+17700,74239180800,2.3964507389068603,26849.107813835144,6873371.600341797,0.0003000000142492354
+19050,79901491200,2.3772277641296387,28767.1561896801,7364391.9845581055,0.0003000000142492354
+20940,87828725760,2.3991868019104006,31490.177367925644,8061485.406188965,0.0003000000142492354
+22690,95168757760,2.34321439743042,34120.41601896286,8734826.500854492,0.0003000000142492354
+24610,103221821440,2.317070655822754,36950.211074352264,9459254.03503418,0.0003000000142492354
+26810,112449290240,2.3007624912261964,39958.97602677345,10229497.862854004,0.0003000000142492354
+28705,120397496320,2.2849823093414305,42676.601992845535,10925210.110168457,0.0003000000142492354
+30270,126961582080,2.2731912517547608,45248.07564377785,11583507.364807129,0.0003000000142492354
+31601,132544200704,2.2603963216145835,47325.226069927216,12115257.873901367,0.0003000000142492354
+33941,142358872064,2.2521857452392577,50523.22919034958,12933946.672729492,0.0003000000142492354
+36466,152949489664,2.243403959274292,54359.25885820389,13915970.267700195,0.0003000000142492354
+38421,161149353984,2.2240604305267335,57226.93031692505,14650094.161132812,0.0003000000142492354
+40661,170544594944,2.2141646480560304,60267.03142309189,15428360.044311523,0.0003000000142492354
+42436,177989484544,2.2029472064971922,62833.42337155342,16085356.383117676,0.0003000000142492354
+44391,186189348864,2.1936937808990478,65558.18872356415,16782896.313232422,0.0003000000142492354
+47231,198101172224,2.1810549545288085,69742.60226631165,17854106.18017578,0.0003000000142492354
+49186,206301036544,2.1736184978485107,72398.7076356411,18534069.15472412,0.0003000000142492354
+51271,215046160384,2.253606328964233,75214.97789525986,19255034.341186523,0.0003000000142492354
+53241,223308939264,2.1548781967163086,78210.45499587059,20021876.47894287,0.0003000000142492354
+56012,234931355648,2.144113004207611,83794.61253786087,21451420.809692383,0.0003000000142492354
+57617,241663213568,2.142407789230347,86206.46052145958,22068853.893493652,0.0003000000142492354
+59247,248499929088,2.1375789260864257,88556.97177791595,22670584.775146484,0.0003000000142492354
+61052,256070647808,2.135520267486572,91073.18482899666,23314735.316223145,0.0003000000142492354
+63172,264962572288,2.135193681716919,94287.14429044724,24137508.938354492,0.0003000000142492354
+65447,274504613888,2.1228764247894287,97476.99922704697,24954111.802124023,0.0003000000142492354
+67567,283396538368,2.119365301132202,100516.82721877098,25732307.76800537,0.0003000000142492354
+69477,291407659008,2.123221168518066,103194.17188191414,26417708.00177002,0.0003000000142492354
+71137,298370203648,2.1131165409088135,105529.8620595932,27015644.68725586,0.0003000000142492354
+73537,308436533248,2.108151330947876,109247.23060750961,27967291.03552246,0.0003000000142492354
+75797,317915660288,2.104380235671997,112709.61656951904,28853661.841796875,0.0003000000142492354
+77482,324983062528,2.0942549324035644,114965.09275817871,29431063.74609375,0.0003000000142492354
+79682,334210531328,2.095156202316284,118112.65136384964,30236838.749145508,0.0003000000142492354
+81092,340124499968,2.0888681697845457,120136.9697060585,30755064.244750977,0.0003000000142492354
+83072,348429221888,2.089090986251831,123003.04564857483,31488779.686035156,0.0003000000142492354
+85122,357027545088,2.0807684326171874,126046.48082518578,32267899.09124756,0.0003000000142492354
+87342,366338899968,2.077386999130249,129345.90752029419,33112552.325195312,0.0003000000142492354
+89132,373846704128,2.076601963043213,131821.41572141647,33746282.42468262,0.0003000000142492354
+91507,383808176128,2.0750280094146727,135420.06685876846,34667537.11584473,0.0003000000142492354
+94137,394839195648,2.0730009651184083,139305.00298047066,35662080.76300049,0.0003000000142492354
+95297,399704588288,2.0671301078796387,140983.04415607452,36091659.30395508,0.0003000000142492354
+96782,405933129728,2.0642451190948488,143185.1160826683,36655389.717163086,0.0003000000142492354
+98422,412811788288,2.063115482330322,145511.34327673912,37250903.878845215,0.0003000000142492354
+100512,421577883648,2.0628562641143797,148511.48833179474,38018941.01293945,0.0003000000142492354
+102607,430364950528,2.0578442001342774,151656.45213866234,38824051.74749756,0.0003000000142492354
+104452,438103441408,2.0522845554351807,154306.36120462418,39502428.46838379,0.0003000000142492354
+106042,444772384768,2.063025493621826,156530.71333670616,40071862.61419678,0.0003000000142492354
+107807,452175331328,2.051406993865967,159062.34849119186,40719961.21374512,0.0003000000142492354
+109292,458403872768,2.050866708755493,161215.54089927673,41271178.470214844,0.0003000000142492354
+110058,461616709632,2.0372196435928345,162695.5256602764,41650054.56903076,0.0003000000142492354
+110858,464972152832,2.0528381304307417,164292.882314682,42058977.872558594,0.0003000000142492354
+112668,472563843072,2.048712902069092,168021.21762132645,43013431.71105957,0.0003000000142492354
+114208,479023071232,2.0424085903167724,171323.53970980644,43858826.16571045,0.0003000000142492354
+116483,488565112832,2.0410716342926025,176055.6183180809,45070238.28942871,0.0003000000142492354
+118228,495884173312,2.041899814605713,179628.25620746613,45984833.58911133,0.0003000000142492354
+120588,505782730752,2.0400297927856443,184538.86824464798,47241950.27062988,0.0003000000142492354
+123358,517400952832,2.0406958198547365,190125.91711258888,48672234.780822754,0.0003000000142492354
+125728,527341453312,2.036259994506836,195263.58475661278,49987477.69769287,0.0003000000142492354
+127663,535457431552,2.031679029464722,199415.867323637,51050462.034851074,0.0003000000142492354
+129318,542399004672,2.0375357055664063,203105.31869864464,51994961.58685303,0.0003000000142492354
+131178,550200410112,2.031361198425293,207202.1809539795,53043758.32421875,0.0003000000142492354
+133333,559239135232,2.030757417678833,212193.66782402992,54321578.96295166,0.0003000000142492354
+135333,567627743232,2.0281761837005616,216530.70245361328,55431859.828125,0.0003000000142492354
+136513,572577021952,2.0286090087890627,219083.33765363693,56085334.439331055,0.0003000000142492354
+138828,582286835712,2.026455774307251,223925.60991740227,57324956.13885498,0.0003000000142492354
+140213,588095946752,2.025322980880737,226956.98628759384,58100988.48962402,0.0003000000142492354
+141973,595477921792,2.024100871086121,230672.4495754242,59052147.091308594,0.0003000000142492354
+143943,603740700672,2.0255939197540282,235109.97332429886,60188153.17102051,0.0003000000142492354
+145393,609822441472,2.0222111415863036,238374.98892879486,61023997.165771484,0.0003000000142492354
+146808,615757381632,2.021069436073303,241464.79900813103,61814988.54608154,0.0003000000142492354
+148688,623642673152,2.0275738620758057,245573.41091251373,62866793.193603516,0.0003000000142492354
+150078,629472755712,2.0161587953567506,248615.07202363014,63645458.43804932,0.0003000000142492354
+152518,639706857472,2.0174728870391845,253896.8513557911,64997593.94708252,0.0003000000142492354
+155333,651513823232,2.0140232133865354,259972.51769518852,66552964.52996826,0.0003000000142492354
+156748,657448763392,2.0141440296173094,263131.802372694,67361741.40740967,0.0003000000142492354
+158518,664872681472,2.012250723838806,267109.7465071678,68380095.10583496,0.0003000000142492354
+160458,673009631232,2.010021390914917,271607.1601662636,69531433.00256348,0.0003000000142492354
+162013,679531773952,2.010418858528137,275090.4128572941,70423145.69146729,0.0003000000142492354
+163968,687731638272,2.0054669094085695,279294.41458678246,71499370.13421631,0.0003000000142492354
+165383,693666578432,2.007085976600647,282943.83094644547,72433620.72229004,0.0003000000142492354
+166938,700188721152,2.007470417022705,286987.54871463776,73468812.47094727,0.0003000000142492354
+169253,709898534912,2.006239137649536,292243.9255628586,74814444.9440918,0.0003000000142492354
+170968,717091766272,2.006234574317932,295954.49125385284,75764349.76098633,0.0003000000142492354
+172743,724536655872,2.00496572971344,300047.9483101368,76812274.76739502,0.0003000000142492354
+174568,732191260672,2.011391215324402,304318.44550824165,77905522.05010986,0.0003000000142492354
+176443,740055580672,2.010074944496155,308309.44185471535,78927217.11480713,0.0003000000142492354
+178318,747919900672,2.0036903190612794,312677.4088833332,80045416.6741333,0.0003000000142492354
+180388,756602109952,2.00131413936615,317592.41147732735,81303657.3381958,0.0003000000142492354
+182488,765410148352,2.0021464109420775,322315.0511870384,82512653.10388184,0.0003000000142492354
+184163,772435607552,2.0196765947341917,326161.1454627514,83497253.23846436,0.0003000000142492354
+186658,782900396032,2.000793972015381,332129.8559639454,85025243.12677002,0.0003000000142492354
+188778,791792320512,1.996970248222351,336970.2900085449,86264394.2421875,0.0003000000142492354
+190673,799740526592,1.99858877658844,341798.2998714447,87500364.76708984,0.0003000000142492354
+192538,807562903552,1.9963369178771972,346169.3793079853,88619361.10284424,0.0003000000142492354
+194168,814399619072,1.9985038137435913,350088.6111769676,89622684.46130371,0.0003000000142492354
+195483,819915128832,1.9990806484222412,353536.11716365814,90505245.99389648,0.0003000000142492354
+197943,830233116672,1.9929803466796876,359786.9988594055,92105471.70800781,0.0003000000142492354
+199888,838391037952,1.9950042867660522,364684.3672618866,93359198.01904297,0.0003000000142492354
+201788,846360215552,2.0014796590805055,369731.45154333115,94651251.59509277,0.0003000000142492354
+203888,855168253952,1.998976125717163,374842.93220973015,95959790.64569092,0.0003000000142492354
+205198,860662792192,1.9944438552856445,378113.42284178734,96797036.24749756,0.0003000000142492354
+206808,867415621632,1.9948081970214844,382128.2782063484,97824839.2208252,0.0003000000142492354
+208818,875846172672,1.9920325231552125,387198.35644340515,99122779.24951172,0.0003000000142492354
+210378,882389286912,1.9906885814666748,391134.9718568325,100130552.79534912,0.0003000000142492354
+212448,891071496192,1.9891257762908936,396670.63805675507,101547683.3425293,0.0003000000142492354
+214158,898243756032,1.9914959716796874,401418.5783264637,102763156.0515747,0.0003000000142492354
+215728,904828813312,1.9855500841140747,405583.5513267517,103829389.13964844,0.0003000000142492354
+217883,913867538432,1.9875614070892333,411368.37800836563,105310304.7701416,0.0003000000142492354
+220178,923493466112,1.9950485038757324,417221.3559541702,106808667.12426758,0.0003000000142492354
+222698,934063112192,1.992979145050049,423431.0784766674,108398356.09002686,0.0003000000142492354
+224348,940983713792,1.9899400997161865,427691.99754333496,109489151.37109375,0.0003000000142492354
+226988,952056676352,1.9866112327575685,434665.77219700813,111274437.68243408,0.0003000000142492354
+229223,961430945792,1.9877167844772339,440215.0782737732,112695060.03808594,0.0003000000142492354
+231223,969819553792,1.982833480834961,445192.8117439747,113969359.80645752,0.0003000000142492354
+232898,976845012992,1.9834683179855346,449444.1582725048,115057704.51776123,0.0003000000142492354
+234578,983891443712,1.980756688117981,453948.37761735916,116210784.67004395,0.0003000000142492354
+236398,991525076992,1.9824994945526122,458921.87237000465,117483999.32672119,0.0003000000142492354
+238478,1000249229312,1.9863429641723633,464148.6053750515,118822042.97601318,0.0003000000142492354
+240853,1010210701312,1.9868581438064574,470267.46969914436,120388472.24298096,0.0003000000142492354
+243668,1022017667072,1.9785122394561767,477878.8736114502,122336991.64453125,0.0003000000142492354
+245728,1030657933312,1.9805147171020507,483333.8747975826,123733471.94818115,0.0003000000142492354
+248088,1040556490752,1.98367919921875,489776.30638217926,125382734.43383789,0.0003000000142492354
+250128,1049112870912,1.9791311168670653,495235.54420924187,126780299.31756592,0.0003000000142492354
+252403,1058654912512,1.9796437311172486,501276.3134906292,128326736.25360107,0.0003000000142492354
+254138,1065932029952,1.9786444664001466,505866.6337928772,129501858.25097656,0.0003000000142492354
+255593,1072034742272,1.9749948978424072,510116.7923388481,130589898.83874512,0.0003000000142492354
+257163,1078619799552,1.9805126142501832,514394.63652586937,131685026.95062256,0.0003000000142492354
+259293,1087553667072,1.9805561542510985,519616.2602994442,133021762.63665771,0.0003000000142492354
+260693,1093425692672,1.982892918586731,522455.496768713,133748607.17279053,0.0003000000142492354
+262388,1100535037952,1.975461072921753,526073.201719284,134674739.64013672,0.0003000000142492354
+264618,1109888335872,1.9781722927093506,530781.3980076313,135880037.8899536,0.0003000000142492354
+265963,1115529674752,1.9748257446289061,533636.5440557003,136610955.27825928,0.0003000000142492354
+267888,1123603709952,1.9752889156341553,537880.8643171787,137697501.26519775,0.0003000000142492354
+269843,1131803574272,1.9739223384857179,542046.5531096458,138763917.59606934,0.0003000000142492354
+272388,1142478077952,1.9724583768844604,547377.2023067474,140128563.79052734,0.0003000000142492354
+274263,1150342397952,1.9735263442993165,551280.9313902855,141127918.4359131,0.0003000000142492354
+276518,1159800553472,1.9759536600112915,556069.8272666931,142353875.78027344,0.0003000000142492354
+278088,1166385610752,1.9720831680297852,559384.3518004417,143202394.0609131,0.0003000000142492354
+280248,1175445307392,1.9682595920562744,563717.7035460472,144311732.1077881,0.0003000000142492354
+282888,1186518269952,1.9799007892608642,569097.7370188236,145689020.67681885,0.0003000000142492354
+285588,1197842890752,1.972299075126648,574613.3726415634,147101023.39624023,0.0003000000142492354
+287458,1205686239232,1.9757914209365846,578463.5975465775,148086680.97192383,0.0003000000142492354
+289718,1215165366272,1.9713094091415406,583165.9108099937,149290473.1673584,0.0003000000142492354
+291768,1223763689472,1.968831171989441,587359.6732494831,150364076.35186768,0.0003000000142492354
+293488,1230977892352,1.9687632942199706,590978.6720416546,151290540.04266357,0.0003000000142492354
+296243,1242533199872,1.9675247526168824,596745.1840589046,152766767.1190796,0.0003000000142492354
+298333,1251299295232,1.9659273052215576,601140.3058359623,153891918.29400635,0.0003000000142492354
+300143,1258890985472,1.966808819770813,604976.6470561028,154874021.6463623,0.0003000000142492354
+301943,1266440732672,1.9661268281936646,608761.162166357,155842857.5145874,0.0003000000142492354
+304838,1278583242752,1.9702806615829467,614507.9426524639,157314033.31903076,0.0003000000142492354
+306473,1285440929792,1.9643911743164062,617861.4496645927,158172531.11413574,0.0003000000142492354
+308428,1293640794112,1.9633469486236572,621982.4583539963,159227509.33862305,0.0003000000142492354
+310853,1303811981312,1.9701481246948243,627004.7200045586,160513208.321167,0.0003000000142492354
+312728,1311676301312,1.9669458293914794,630913.0796144009,161513748.38128662,0.0003000000142492354
+314203,1317862899712,1.9643836832046508,633971.249876976,162296639.96850586,0.0003000000142492354
+315978,1325307789312,1.9652533531188965,637595.4410812855,163224432.91680908,0.0003000000142492354
+317268,1330718441472,1.9652329683303833,640352.2525105476,163930176.6427002,0.0003000000142492354
+319243,1339002191872,1.9665932512283326,644526.4115104675,164998761.3466797,0.0003000000142492354
+320903,1345964736512,1.965696063041687,647834.7610650063,165845698.8326416,0.0003000000142492354
+323268,1355884265472,1.9655056619644165,653158.2559595108,167208513.52563477,0.0003000000142492354
+325693,1366055452672,1.969166202545166,658094.7498605251,168472255.96429443,0.0003000000142492354
+328088,1376100810752,1.959834599494934,663075.8995423317,169747430.2828369,0.0003000000142492354
+329573,1382329352192,1.9652193641662599,666247.1070299149,170559259.3996582,0.0003000000142492354
+332383,1394115346432,1.9600944375991822,672137.7539036274,172067264.9993286,0.0003000000142492354
+334588,1403363786752,1.96407151222229,676819.1959176064,173265714.15490723,0.0003000000142492354
+336708,1412255711232,1.963568124771118,681316.4080533981,174417000.46166992,0.0003000000142492354
+338348,1419134369792,1.9614058494567872,684835.5685112476,175317905.5388794,0.0003000000142492354
+340543,1428340867072,1.9604207324981688,689632.6337156296,176545954.23120117,0.0003000000142492354
+342283,1435638956032,1.9608780097961427,693278.7204716206,177479352.44073486,0.0003000000142492354
+344258,1443922706432,1.9617660331726074,697295.1231441498,178507551.52490234,0.0003000000142492354
+346478,1453234061312,1.9609308624267578,701910.6264374256,179689120.36798096,0.0003000000142492354
+348378,1461203238912,1.9573345804214477,705759.8878202438,180674531.28198242,0.0003000000142492354
+349778,1467075264512,1.9614266681671142,708690.9684870243,181424887.93267822,0.0003000000142492354
+351773,1475442900992,1.9579530572891235,712835.2185485363,182485815.9484253,0.0003000000142492354
+353813,1483999281152,1.964455862045288,717132.7136621475,183585974.69750977,0.0003000000142492354
+355633,1491632914432,1.9603030967712403,720853.2569811344,184538433.7871704,0.0003000000142492354
+357368,1498910031872,1.9643068838119506,724534.5202586651,185480837.18621826,0.0003000000142492354
+359153,1506396864512,1.9605508756637573,728431.4290521145,186478445.8373413,0.0003000000142492354
+360998,1514135355392,1.9579992580413819,732383.5597906113,187490191.30639648,0.0003000000142492354
+363123,1523048251392,1.9576819372177123,736770.1569578648,188613160.18121338,0.0003000000142492354
+365603,1533450125312,1.959110941886902,742102.3170361519,189978193.16125488,0.0003000000142492354
+367503,1541419302912,1.955609769821167,746344.5685229301,191064209.54187012,0.0003000000142492354
+369203,1548549619712,1.9582535791397095,749822.9885122776,191954685.05914307,0.0003000000142492354
+371333,1557483487232,1.9579929542541503,754185.5399475098,193071498.2265625,0.0003000000142492354
+373388,1566102781952,1.957696418762207,758274.8679385185,194118366.19226074,0.0003000000142492354
+376168,1577762947072,1.9575824451446533,764248.9875659943,195647740.81689453,0.0003000000142492354
+377753,1584410918912,1.9581837558746338,767788.1166574955,196553757.86431885,0.0003000000142492354
+379323,1590995976192,1.9561321783065795,770935.5572018623,197359502.64367676,0.0003000000142492354
+380913,1597664919552,1.9606944847106933,774087.8183994293,198166481.5102539,0.0003000000142492354
+383008,1606451986432,1.9556411361694337,778503.4085962772,199296872.60064697,0.0003000000142492354
+385328,1616182771712,1.9551253461837768,783158.1024012566,200488474.21472168,0.0003000000142492354
+386958,1623019487232,1.9550975131988526,786599.5460090637,201369483.7783203,0.0003000000142492354
+388728,1630443405312,1.954553213119507,790218.0025575161,202295808.65472412,0.0003000000142492354
+390703,1638727155712,1.9554255104064941,794259.4525990486,203330419.86535645,0.0003000000142492354
+393058,1648604741632,1.9582529544830323,799226.6983923912,204602034.78845215,0.0003000000142492354
+395333,1658146783232,1.9579640626907349,803935.4146561623,205807466.15197754,0.0003000000142492354
+397283,1666325676032,1.9572097873687744,808031.7134544849,206856118.64434814,0.0003000000142492354
+399063,1673791537152,1.9551084566116332,811768.8775155544,207812832.64398193,0.0003000000142492354
+400948,1681697800192,1.9539657783508302,815695.0820395947,208817941.00213623,0.0003000000142492354
+402748,1689247547392,1.9534498262405395,819313.22417593,209744185.3890381,0.0003000000142492354
+404838,1698013642752,1.9571835231781005,823827.6516792774,210899878.82989502,0.0003000000142492354
+406978,1706989453312,1.9534950685501098,828320.0534090996,212049933.6727295,0.0003000000142492354
+408823,1714727944192,1.9579477405548096,832137.5924572945,213027223.66906738,0.0003000000142492354
+410858,1723263352832,1.9587352895736694,836624.1051552296,214175770.91973877,0.0003000000142492354
+413173,1732973166592,1.94909686088562,841489.677161932,215421357.3534546,0.0003000000142492354
+414963,1740480970752,1.9494858360290528,845319.1821036339,216401710.61853027,0.0003000000142492354
+416438,1746667569152,1.9514140510559081,848435.1806650162,217199406.25024414,0.0003000000142492354
+418218,1754133430272,1.953859715461731,852178.8924751282,218157796.4736328,0.0003000000142492354
+419538,1759669911552,1.9514509153366089,854988.7884757519,218877129.84979248,0.0003000000142492354
+421903,1769589440512,1.9497252225875854,859916.5196237564,220138629.02368164,0.0003000000142492354
+423563,1776551985152,1.950548481941223,863238.5992774963,220989081.41503906,0.0003000000142492354
+425998,1786765115392,1.9461521196365357,868412.1394166946,222313507.69067383,0.0003000000142492354
+428113,1795636068352,1.9487574100494385,872916.5063915253,223466625.63623047,0.0003000000142492354
+430068,1803835932672,1.9520085430145264,877108.2609584332,224539714.8053589,0.0003000000142492354
+431843,1811280822272,1.945929660797119,880759.6764304638,225474477.16619873,0.0003000000142492354
+433533,1818369196032,1.956407871246338,884301.2288644314,226381114.58929443,0.0003000000142492354
+435433,1826338373632,1.9516588306427003,888296.9268109798,227404013.26361084,0.0003000000142492354
+437573,1835314184192,1.954493727684021,892747.6124324799,228543388.78271484,0.0003000000142492354
+439148,1841920212992,1.9496833419799804,895965.43851161,229367152.25897217,0.0003000000142492354
+441068,1849973276672,1.9504491090774536,899925.4805030823,230380923.00878906,0.0003000000142492354
+442638,1856558333952,1.9469935894012451,903213.0597736835,231222543.302063,0.0003000000142492354
+443888,1861801213952,1.9494563484191894,905801.6484167576,231885221.99468994,0.0003000000142492354
+445928,1870357594112,1.9473052310943604,909896.5962805748,232933528.64782715,0.0003000000142492354
+448303,1880319066112,1.9493485498428345,914883.1788625717,234210093.78881836,0.0003000000142492354
+449618,1885834575872,1.9427200174331665,917573.1076028347,234898715.54632568,0.0003000000142492354
+451443,1893489180672,1.9486754369735717,921398.3529331684,235877978.3508911,0.0003000000142492354
+453178,1900766298112,1.9485606861114502,925094.3730487823,236824159.50048828,0.0003000000142492354
+455008,1908441874432,1.9489810371398926,928975.8425166607,237817815.68426514,0.0003000000142492354
+456268,1913726697472,1.9550775146484376,931600.2976133823,238489676.18902588,0.0003000000142492354
+458418,1922744451072,1.9456060075759887,935905.3285422325,239591764.10681152,0.0003000000142492354
+460183,1930147397632,1.947961401939392,939539.942980051,240522225.40289307,0.0003000000142492354
+461788,1936879255552,1.9488294267654418,942792.8803005219,241354977.3569336,0.0003000000142492354
+463823,1945414664192,1.947607946395874,946912.1931769848,242409521.4533081,0.0003000000142492354
+465948,1954327560192,1.9467726850509643,951234.2053678036,243515956.5741577,0.0003000000142492354
+468153,1963576000512,1.9497644996643066,955770.4690947533,244677240.08825684,0.0003000000142492354
+470463,1973264842752,1.9484503936767579,960534.4983091354,245896831.56713867,0.0003000000142492354
+472618,1982303567872,1.9491539764404298,965153.5004379749,247079296.11212158,0.0003000000142492354
+474008,1988133650432,1.9477979612350464,968103.7006523609,247834547.3670044,0.0003000000142492354
+475528,1994508992512,1.9452753591537475,971213.4961261749,248630655.00830078,0.0003000000142492354
+476808,1999877701632,1.9501544761657714,973806.6988768578,249294514.9124756,0.0003000000142492354
+479038,2009230999552,1.9513997220993042,978369.7799589634,250462663.66949463,0.0003000000142492354
+480708,2016235487232,1.9453705215454102,981754.654002428,251329191.42462158,0.0003000000142492354
+482438,2023491633152,1.9476304054260254,985335.1197412014,252245790.65374756,0.0003000000142492354
+484258,2031125266432,1.946502251625061,989192.7164342403,253233335.40716553,0.0003000000142492354
+486648,2041149652992,1.9477696752548217,994216.994954586,254519550.70837402,0.0003000000142492354
+488488,2048867172352,1.9466051578521728,998147.0724525452,255525650.54785156,0.0003000000142492354
+490083,2055557087232,1.9430851173400878,1001431.4499261379,256366451.1810913,0.0003000000142492354
+492583,2066042847232,1.9475297498703004,1006415.7696654797,257642437.0343628,0.0003000000142492354
+494498,2074074939392,1.9443493700027465,1010333.6361887455,258645410.86431885,0.0003000000142492354
+496963,2084413898752,1.9396081161499024,1015506.1208918095,259969566.94830322,0.0003000000142492354
+499103,2093389709312,1.9419655513763427,1019873.1831290722,261087534.88104248,0.0003000000142492354
+501178,2102092890112,1.9465229797363282,1024130.1322581768,262177313.85809326,0.0003000000142492354
+502393,2107188969472,1.9423305225372314,1026613.9520468712,262813171.72399902,0.0003000000142492354
+504313,2115242033152,1.9462104892730714,1030607.0607998371,263835407.5647583,0.0003000000142492354
+506513,2124469501952,1.9463778829574585,1036038.6564817429,265225896.05932617,0.0003000000142492354
+508283,2131893420032,1.94479172706604,1039523.2592954636,266117954.37963867,0.0003000000142492354
+510513,2141246717952,1.9500395441055298,1044186.520763874,267311749.31555176,0.0003000000142492354
+513238,2152676196352,1.9431734418869018,1050192.2955384254,268849227.6578369,0.0003000000142492354
+515463,2162008522752,1.940658025741577,1054769.0741755962,270020882.98895264,0.0003000000142492354
+516908,2168069292032,1.94367760181427,1057759.1069140434,270786331.3699951,0.0003000000142492354
+518913,2176478871552,1.9397131967544556,1061762.4483671188,271811186.7819824,0.0003000000142492354
+520548,2183336558592,1.9405004644393922,1065249.6324219704,272703905.9000244,0.0003000000142492354
+522343,2190865334272,1.9442265224456787,1068883.7777450085,273634247.10272217,0.0003000000142492354
+523948,2197597192192,1.9434235429763793,1072159.4496753216,274472819.1168823,0.0003000000142492354
+526028,2206321344512,1.944770050048828,1076342.403468132,275543655.2878418,0.0003000000142492354
+528138,2215171325952,1.9507891464233398,1080682.7243127823,276654777.42407227,0.0003000000142492354
+529448,2220665864192,1.9469588804244995,1083393.0977442265,277348633.022522,0.0003000000142492354
+531078,2227502579712,1.9411223173141479,1086933.0393912792,278254858.0841675,0.0003000000142492354
+533533,2237799596032,1.945275192260742,1092342.3397948742,279639638.9874878,0.0003000000142492354
+535718,2246964150272,1.9388134288787842,1096843.8852527142,280792034.6246948,0.0003000000142492354
+538088,2256904650752,1.9404128360748292,1101629.817631483,282017233.31365967,0.0003000000142492354
+539908,2264538284032,1.9479225158691407,1105403.6813309193,282983342.42071533,0.0003000000142492354
+541689,2272008339456,1.94042934690203,1109678.1022558212,284077594.17749023,0.0003000000142492354
+543469,2279474200576,1.947353720664978,1113744.8308417797,285118676.6954956,0.0003000000142492354
+545714,2288890413056,1.9407488679885865,1118675.616765976,286380957.89208984,0.0003000000142492354
+547824,2297740394496,1.943647451400757,1123465.322243929,287607122.4944458,0.0003000000142492354
+550129,2307408265216,1.936250605583191,1128619.5835754871,288926613.3953247,0.0003000000142492354
+552129,2315796873216,1.9332826375961303,1133199.2653152943,290099011.92071533,0.0003000000142492354
+553359,2320955867136,1.9386159372329712,1135856.7320129871,290779323.3953247,0.0003000000142492354
+555114,2328316870656,1.9416591119766236,1139984.8873410225,291836131.15930176,0.0003000000142492354
+556399,2333706551296,1.938959150314331,1142788.1939873695,292553777.6607666,0.0003000000142492354
+559029,2344737570816,1.939472279548645,1148689.0084109306,294064386.15319824,0.0003000000142492354
+560504,2350924169216,1.9378677558898927,1151954.130159378,294900257.3208008,0.0003000000142492354
+562404,2358893346816,1.93844388961792,1156252.3606095314,296000604.31604004,0.0003000000142492354
+563969,2365457432576,1.9385316944122315,1159754.4510278702,296897139.46313477,0.0003000000142492354
+565439,2371623059456,1.940172085762024,1163023.9423487186,297734129.241272,0.0003000000142492354
+567349,2379634180096,1.9446619653701782,1167051.341565609,298765143.4407959,0.0003000000142492354
+569419,2388316389376,1.9477505493164062,1171646.2587795258,299941442.2475586,0.0003000000142492354
+571179,2395698364416,1.940185685157776,1175623.4341797829,300959599.1500244,0.0003000000142492354
+572794,2402472165376,1.9385631704330444,1179060.1584682465,301839400.5678711,0.0003000000142492354
+575114,2412202950656,1.9357694053649903,1184413.2855353355,303209801.0970459,0.0003000000142492354
+577274,2421262647296,1.9404512023925782,1189386.485613346,304482940.3170166,0.0003000000142492354
+578794,2427637989376,1.938502631187439,1192710.6702563763,305333931.5856323,0.0003000000142492354
+580859,2436299227136,1.9388711261749267,1197265.731966257,306500027.3833618,0.0003000000142492354
+582864,2444708806656,1.9373184585571288,1201876.9888648987,307680509.14941406,0.0003000000142492354
+584749,2452615069696,1.93695809841156,1205988.6900863647,308733104.6621094,0.0003000000142492354
+586419,2459619557376,1.9383648490905763,1209720.9189519882,309688555.251709,0.0003000000142492354
+588349,2467714564096,1.939493908882141,1213947.5269494057,310770566.89904785,0.0003000000142492354
+589844,2473985048576,1.9410696029663086,1217308.0315742493,311630856.0830078,0.0003000000142492354
+591809,2482226855936,1.9402881526947022,1221565.3580245972,312720731.6542969,0.0003000000142492354
+593214,2488119853056,1.9392866373062134,1224773.7630198002,313542083.33306885,0.0003000000142492354
+594889,2495145312256,1.9365543270111083,1228733.0416152477,314555658.6535034,0.0003000000142492354
+597994,2508168626176,1.9365768003463746,1236302.7512440681,316493504.31848145,0.0003000000142492354
+600119,2517081522176,1.9411273717880249,1241421.3318464756,317803860.95269775,0.0003000000142492354
+602079,2525302358016,1.9350768852233886,1245993.7221405506,318974392.86798096,0.0003000000142492354
+604194,2534173310976,1.932094326019287,1250783.082157135,320200469.03222656,0.0003000000142492354
+606389,2543379808256,1.9376727628707886,1255871.4827952385,321503099.59558105,0.0003000000142492354
+608579,2552565334016,1.939154667854309,1261080.8259210587,322836691.435791,0.0003000000142492354
+610349,2559989252096,1.936561999320984,1265108.8604876995,323867868.2848511,0.0003000000142492354
+612709,2569887809536,1.937281560897827,1270597.0959351063,325272856.5593872,0.0003000000142492354
+615014,2579555680256,1.9350225877761842,1276151.3414213657,326694743.4038696,0.0003000000142492354
+616689,2586581139456,1.9427925062179565,1280048.6847863197,327692463.30529785,0.0003000000142492354
+619404,2597968674816,1.9364050149917602,1286357.8372046947,329307606.32440186,0.0003000000142492354
+622299,2610111184896,1.93864896774292,1293083.5514781475,331029389.17840576,0.0003000000142492354
+623764,2616255840256,1.9380575942993163,1296736.4023706913,331964519.006897,0.0003000000142492354
+625744,2624560562176,1.9355155229568481,1301489.093458414,333181207.925354,0.0003000000142492354
+627114,2630306758656,1.9352296638488768,1304957.2336623669,334069051.8175659,0.0003000000142492354
+628734,2637101531136,1.9340949487686157,1308999.6953866482,335103922.01898193,0.0003000000142492354
+630624,2645028765696,1.9315508031845092,1313462.8685352802,336246494.34503174,0.0003000000142492354
+632439,2652641427456,1.9320917558670043,1317971.702303648,337400755.7897339,0.0003000000142492354
+634469,2661155864576,1.9341316890716553,1323121.1190936565,338719006.4879761,0.0003000000142492354
+636004,2667594121216,1.93304123878479,1326822.0704636574,339666450.0386963,0.0003000000142492354
+637769,2674997067776,1.9356712484359742,1331277.904321909,340807143.5064087,0.0003000000142492354
+639954,2684161622016,1.9327107000350952,1336432.0497164726,342126604.727417,0.0003000000142492354
+642319,2694081150976,1.9354500675201416,1342074.1212949753,343570975.0515137,0.0003000000142492354
+644199,2701966442496,1.939101390838623,1346593.1013784409,344727833.95288086,0.0003000000142492354
+646039,2709683961856,1.9382933473587036,1351035.1993951797,345865011.045166,0.0003000000142492354
+647584,2716164161536,1.9352576589584352,1354715.6578097343,346807208.399292,0.0003000000142492354
+650044,2726482149376,1.9363348960876465,1361083.3268795013,348437331.68115234,0.0003000000142492354
+651704,2733444694016,1.941686954498291,1364981.465630293,349435255.201355,0.0003000000142492354
+653074,2739190890496,1.938481993675232,1368312.636607647,350288034.9715576,0.0003000000142492354
+655199,2748103786496,1.9358009958267213,1373476.0453662872,351609867.61376953,0.0003000000142492354
+657414,2757394169856,1.9342416048049926,1378714.6798548698,352950958.0428467,0.0003000000142492354
+659149,2764671287296,1.931928095817566,1383080.1730082035,354068524.2901001,0.0003000000142492354
+660985,2772372029440,1.935685157775879,1387002.5008759499,355072640.22424316,0.0003000000142492354
+663140,2781410754560,1.9334363698959351,1391618.811923027,356254415.8522949,0.0003000000142492354
+665290,2790428508160,1.9354631328582763,1396122.1793529987,357407277.9143677,0.0003000000142492354
+667650,2800327065600,1.931658730506897,1400966.8387339115,358647510.71588135,0.0003000000142492354
+669220,2806912122880,1.9370522975921631,1404169.4934687614,359467390.3280029,0.0003000000142492354
+671200,2815216844800,1.9337073278427124,1408546.461587906,360587894.1665039,0.0003000000142492354
+672570,2820963041280,1.9374622011184692,1411312.4806556702,361295995.04785156,0.0003000000142492354
+675160,2831826288640,1.9349578189849854,1416801.3328347206,362701141.2056885,0.0003000000142492354
+677100,2839963238400,1.9355492782592774,1420722.4307272434,363704942.2661743,0.0003000000142492354
+679220,2848855162880,1.9345752000808716,1425289.8431024551,364874199.8342285,0.0003000000142492354
+681030,2856446853120,1.9323256921768188,1429037.8618321419,365833692.6290283,0.0003000000142492354
+682755,2863682027520,1.9327951765060425,1432567.0128557682,366737155.29107666,0.0003000000142492354
+684625,2871525376000,1.9301513814926148,1436393.6035106182,367716762.49871826,0.0003000000142492354
+686665,2880081756160,1.9331067037582397,1440623.0019733906,368799488.505188,0.0003000000142492354
+688505,2887799275520,1.9330306482315063,1444489.784328699,369789384.788147,0.0003000000142492354
+690905,2897865605120,1.9323818731307982,1449590.7331762314,371095227.69311523,0.0003000000142492354
+692410,2904178032640,1.9332720708847047,1452694.2836449146,371889736.61309814,0.0003000000142492354
+694725,2913887846400,1.9385543632507325,1457739.7935256958,373181387.1425781,0.0003000000142492354
+696550,2921542451200,1.9355508661270142,1461589.2234847546,374166841.21209717,0.0003000000142492354
+698595,2930119802880,1.9339664602279663,1465841.0182807446,375255300.6798706,0.0003000000142492354
+701155,2940857221120,1.930873293876648,1471270.0683758259,376645137.5042114,0.0003000000142492354
+703215,2949497487360,1.9291921377182006,1475620.3646681309,377758813.3550415,0.0003000000142492354
+705305,2958263582720,1.9304812574386596,1479840.668790102,378839211.2102661,0.0003000000142492354
+707870,2969021972480,1.930038604736328,1484955.5733499527,380148626.7775879,0.0003000000142492354
+710170,2978668871680,1.932656879425049,1489657.720097065,381352376.34484863,0.0003000000142492354
+711900,2985925017600,1.9304987859725953,1493193.3803639412,382257505.37316895,0.0003000000142492354
+713940,2994481397760,1.9356035280227661,1497471.367437601,383352670.0640259,0.0003000000142492354
+715785,3002219888640,1.9410306358337401,1501453.1358907223,384372002.7880249,0.0003000000142492354

metadata/training_logs/ConvergenceCurve_phase2.csv

@@ -0,0 +1,386 @@
+training_steps,training_tokens,training_loss,walltime,gputime,learning_rate
+610,2558525440,1.8591474676132203,1577.0631277561188,403728.1607055664,0.00029999890830367804
+1305,5473566720,1.8590973711013794,3366.526930332184,861830.8941650391,0.00029999500839039683
+2285,9583984640,1.8500521612167358,5721.44074678421,1464688.8311767578,0.00029998470563441515
+3220,13505658880,1.844328899383545,7789.655754566193,1994151.8731689453,0.00029996962985023856
+4090,17154703360,1.8464346218109131,9910.505961179733,2537089.5260620117,0.00029995100339874625
+5000,20971520000,1.8453072452545165,11916.204143762589,3050548.2608032227,0.00029992678901180625
+6095,25564282880,1.8352785205841065,14300.995573997498,3661054.8669433594,0.00029989119502715766
+7065,29632757760,1.8348756313323975,16557.20342874527,4238644.077758789,0.0002998538257088512
+8065,33827061760,1.8344827222824096,19036.596146821976,4873368.613586426,0.0002998095005750656
+8945,37518049280,1.828942255973816,20997.795709371567,5375435.701599121,0.00029976567020639777
+9820,41188065280,1.8338001585006714,23086.91358447075,5910249.877624512,0.0002997176197823137
+10620,44543508480,1.8306499671936036,24985.23111462593,6396219.165344238,0.0002996697439812124
+12125,50855936000,1.830906572341919,28553.66275191307,7309737.664489746,0.00029956953949294984
+12855,53917777920,1.8340681982040405,30271.014125347137,7749379.616088867,0.00029951619217172265
+14065,58992885760,1.8334971237182618,32994.268872737885,8446532.831420898,0.00029942087712697685
+15045,63103303680,1.8326430654525756,35287.43020415306,9033582.132263184,0.0002993374364450574
+16145,67717038080,1.8295758533477784,37766.7780315876,9668295.176086426,0.0002992370864376426
+17180,72058142720,1.8289901733398437,40321.1701130867,10322219.548950195,0.00029913624166511
+17990,75455528960,1.8290645360946656,42194.40278887749,10801767.113952637,0.0002990529465023428
+18615,78076968960,1.8249164342880249,43660.11423587799,11176989.244384766,0.0002989860950037837
+19680,82543902720,1.8303661298751832,46085.95684170723,11798004.95147705,0.0002988669148180634
+20770,87115694080,1.830328001976013,48745.56445598602,12478864.500732422,0.00029873813036829233
+21640,90764738560,1.8286544513702392,50702.001802921295,12979712.461547852,0.00029863035888411105
+22640,94959042560,1.904537606239319,53069.18496012688,13585711.34979248,0.0002985011087730527
+23520,98650030080,1.8263292407989502,55266.38353705406,14148194.18548584,0.00029838253976777196
+24390,102299074560,1.8283975696563721,57249.15934920311,14655784.793395996,0.0002982609439641237
+25230,105822289920,1.8290383148193359,59200.21234869957,15155254.36126709,0.0002981393481604755
+26190,109848821760,1.8255306959152222,61412.546890974045,15721612.004089355,0.00029799542971886694
+27140,113833410560,1.836202530860901,63649.32084918022,16294226.137390137,0.0002978477568831295
+28160,118111600640,1.834192099571228,66052.28078484535,16909383.88092041,0.0002976834075525403
+29005,121655787520,1.8257641887664795,68040.92909216881,17418477.847595215,0.000297542690532282
+29670,124444999680,1.8256645202636719,69607.03523039818,17819401.018981934,0.00029742909828200936
+30470,127800442880,1.826603798866272,71474.74717855453,18297535.27770996,0.00029728905064985156
+31540,132288348160,1.828463740348816,73985.46221852303,18940278.327941895,0.00029709600494243205
+32370,135769620480,1.8263410711288453,75954.74554181099,19444414.858703613,0.0002969417255371809
+33020,138495918080,1.8224460744857789,77557.6713681221,19854763.870239258,0.0002968181506730616
+34010,142648279040,1.8231004333496095,79926.30450844765,20461133.954162598,0.0002966253086924553
+35155,147450757120,1.8239600515365602,82639.50283169746,21155712.72491455,0.00029639527201652527
+36105,151435345920,1.8210523796081544,84922.23833799362,21740093.014526367,0.0002961987629532814
+37005,155210219520,1.824894781112671,87062.47863411903,22287994.530334473,0.00029600784182548523
+37705,158146232320,1.821996898651123,88712.92949080467,22710509.949645996,0.00029585615266114473
+38625,162004992000,1.8280460500717164,90831.21213316917,23252790.30609131,0.0002956525713670999
+39510,165716951040,1.834221510887146,92997.24526286125,23807294.78729248,0.00029545219149440527
+40440,169617653760,1.8383075904846191,95242.45730638504,24382069.07043457,0.0002952368522528559
+41330,173350584320,1.830979480743408,97276.91481351852,24902890.192260742,0.00029502619872801006
+42160,176831856640,1.8327462816238402,99362.13406801224,25436706.321411133,0.0002948257024399936
+43330,181739192320,1.830559525489807,102191.28093719482,26160967.919921875,0.0002945365267805755
+44290,185765724160,1.828639841079712,104538.39753627777,26761829.76928711,0.00029429348069243133
+45485,190777917440,1.8290658330917358,107089.43394422531,27414895.08972168,0.00029398369952104986
+46220,193860730880,1.8209784078598021,108737.09371089935,27836695.989990234,0.00029378922772593796
+47515,199292354560,1.8233415842056275,111748.68362903595,28607663.009033203,0.0002934391959570348
+48345,202773626880,1.823132553100586,113723.25442433357,29113153.132629395,0.00029320994508452713
+49165,206212956160,1.8234637832641603,115590.78886032104,29591241.948242188,0.00029297967557795346
+50100,210134630400,1.821461820602417,117961.46292924881,30198134.509887695,0.00029271256062202156
+51100,214328934400,1.822403244972229,120286.80166840553,30793421.227111816,0.0002924215223174542
+51960,217936035840,1.8264352703094482,122454.73382234573,31348411.858520508,0.00029216677648946643
+53060,222549770240,1.8246722745895385,125172.27768063545,32044103.086242676,0.0002918350219260901
+53870,225947156480,1.820338830947876,127063.45998644829,32528245.75653076,0.0002915864752139896
+54905,230288261120,1.8193842124938966,129660.89175105095,33193188.288269043,0.00029126362642273307
+55730,233748561920,1.8256155633926392,131626.18517160416,33696303.403930664,0.00029100209940224886
+56550,237187891200,1.8185726022720337,133608.28788876534,34203721.699523926,0.0002907384478021413
+57680,241927454720,1.813451075553894,136299.1924214363,34892593.259887695,0.00029036912019364536
+58740,246373416960,1.8195957946777344,138947.14162111282,35570468.25500488,0.0002900163526646793
+59630,250106347520,1.8213485813140868,141040.1522409916,36106278.97369385,0.0002897154481615871
+60340,253084303360,1.8193719291687012,142725.4915714264,36537725.842285156,0.00028947231476195157
+61260,256943063040,1.8167563629150392,144952.09844470024,37107737.20184326,0.0002891532494686544
+62180,260801822720,1.821616997718811,147097.62022733688,37656990.77819824,0.00028882958577014506
+62935,263968522240,1.8151561307907105,149051.16348266602,38157097.8515625,0.0002885606081690639
+63710,267219107840,1.8175263404846191,151017.6004076004,38660505.7043457,0.00028828129870817065
+64735,271518269440,1.8181425285339357,153595.16024708748,39320361.023254395,0.00028790696524083614
+65980,276740177920,1.8494785690307618,156706.34971380234,40116825.5267334,0.0002874447382055223
+67065,281290997760,1.8292945432662964,159288.90483617783,40777959.63806152,0.0002870352182071656
+67970,285086842880,1.8285573816299439,161516.45609498024,41348212.76031494,0.0002866888535209
+69020,289490862080,1.8231487369537354,163975.17231583595,41977644.112854004,0.0002862815745174885
+70145,294209454080,1.8179030656814574,166737.71277189255,42684854.46960449,0.0002858387597370893
+71020,297879470080,1.82222318649292,168920.65007448196,43243686.41906738,0.0002854897466022521
+71900,301570457600,1.8190499830245972,171103.7938094139,43802571.21520996,0.00028513470897451043
+72730,305051729920,1.8164095449447633,173253.16641569138,44352810.60241699,0.00028479614411480725
+73500,308281344000,1.822648320198059,175279.33874964714,44871510.71990967,0.00028447885415516794
+74105,310818897920,1.8248628902435302,176744.5105755329,45246594.707336426,0.00028422739706002176
+75240,315579432960,1.8264307117462157,179418.98047208786,45931259.00085449,0.00028375055990181863
+76350,320235110400,1.8188332080841065,181883.70473980904,46562228.41339111,0.00028327779727987945
+77395,324618158080,1.8180358505249024,184523.65441918373,47238055.531311035,0.00028282692073844373
+78435,328980234240,1.8177628469467164,187024.7860479355,47878345.228271484,0.00028237266815267503
+79440,333195509760,1.8160990762710572,189346.2539935112,48472641.02233887,0.00028192848549224436
+80275,336697753600,1.8202384662628175,191327.86167621613,48979932.58911133,0.0002815555199049413
+81150,340367769600,1.8183535814285279,193326.0135269165,49491459.462890625,0.00028116090106777847
+82065,344205557760,1.8186530447006226,195503.11517953873,50048797.485961914,0.0002807441051118076
+82880,347623915520,1.8188217210769653,197428.6881160736,50541744.157714844,0.0002803693350870162
+83860,351734333440,1.815137538909912,199687.94003486633,51120112.64892578,0.0002799142966978252
+84670,355131719680,1.826065616607666,201694.90508198738,51633895.70098877,0.00027953454991802573
+85575,358927564800,1.8222646951675414,203917.17332839966,52202796.37207031,0.0002791064325720072
+86690,363604213760,1.8206488370895386,206349.41662192345,52825450.6552124,0.00027857336681336164
+87850,368469606400,1.8193091201782225,209131.0117805004,53537539.015808105,0.00027801230316981673
+88770,372328366080,1.8248542833328247,211409.14634346962,54120741.46392822,0.0002775626489892602
+89610,375851581440,1.8217721843719483,213419.38222527504,54635361.84967041,0.00027714844327419996
+90430,379290910720,1.8160008716583251,215920.88285684586,55275746.01135254,0.00027674081502482295
+91205,382541496320,1.8158236122131348,217760.031799078,55746568.140563965,0.00027635254082269967
+92375,387448832000,1.8178369760513307,220478.92622113228,56442605.11260986,0.0002757608890533447
+93305,391349534720,1.815445532798767,222732.27391839027,57019462.12310791,0.00027528591454029083
+94105,394704977920,1.8197657203674316,225054.1094648838,57613852.023010254,0.0002748740080278367
+95046,398651817984,1.817927827835083,228682.3605451584,58542684.29956055,0.0002743855875451118
+96146,403265552384,1.8115335369110108,231807.71195149422,59342774.25958252,0.00027380933170206845
+97451,408739119104,1.8139947843551636,235055.7580397129,60174274.058166504,0.0002731182612478733
+98176,411779989504,1.8124797201156617,236880.4615137577,60641398.14752197,0.0002727308892644942
+99011,415282233344,1.8136535596847534,240303.2180700302,61517623.825927734,0.00027228170074522495
+99926,419120021504,1.8121351099014282,242852.75603866577,62170305.54589844,0.0002717857132665813
+100606,421972148224,1.810523238182068,244572.99504256248,62610686.730895996,0.0002714146103244275
+101501,425726050304,1.8080514574050903,246895.61300730705,63205276.929870605,0.0002709229302126914
+102501,429920354304,1.8128723192214966,249515.85827207565,63876059.71765137,0.0002703692007344216
+103261,433108025344,1.8141535377502442,251568.564111948,64401552.41265869,0.0002699453034438193
+104066,436484440064,1.8122142505645753,253572.77815914154,64914631.208740234,0.00026949343737214804
+105421,442167721984,1.8123528671264648,257055.2549932003,65806145.27825928,0.0002687262312974781
+106391,446236196864,1.8125610208511354,259383.88051605225,66402273.412109375,0.00026817191974259913
+107301,450053013504,1.810970335006714,261630.19577002525,66977330.117126465,0.0002676480507943779
+108131,453534285824,1.8094647312164307,263655.1596364975,67495720.86694336,0.0002671670517884195
+108926,456868757504,1.8094174718856813,265667.19418144226,68010801.71044922,0.0002667034568730742
+109641,459867684864,1.8133175992965698,267440.3481326103,68464729.12194824,0.0002662840997800231
+110701,464313647104,1.8144058895111084,270033.11794018745,69128478.19268799,0.0002656583092175424
+111481,467585204224,1.8157741832733154,272057.5650296211,69646736.64758301,0.00026519468519836664
+112356,471255220224,1.811423945426941,274298.8903687,70220515.9343872,0.000264671427430585
+113021,474044432384,1.813492259979248,276090.6790766716,70679213.84362793,0.00026427156990393996
+113731,477022388224,1.8101095819473267,277969.1963288784,71160114.26019287,0.0002638425212353468
+114646,480860176384,1.803206763267517,280198.889534235,71730915.72076416,0.0002632864052429795
+115606,484886708224,1.8086436986923218,282700.1856443882,72371247.52496338,0.00026269908994436264
+116671,489353641984,1.8087895679473878,285431.34041285515,73070423.14569092,0.00026204297319054604
+117696,493652803584,1.8093397521972656,288079.37056040764,73748318.86346436,0.0002614069962874055
+119021,499210256384,1.8060203695297241,291618.83398628235,74654421.50048828,0.00026057835202664137
+120376,504893538304,1.8072790145874023,295250.15199518204,75584038.9107666,0.0002597234270069748
+121671,510325161984,1.8061434650421142,298584.27656030655,76437574.79943848,0.00025889932294376194
+122741,514813067264,1.8125104475021363,301429.6234998703,77165983.6159668,0.00025821320014074445
+123981,520014004224,1.8782437419891358,305031.46193122864,78088054.25439453,0.00025741232093423605
+124911,523914706944,1.8073894453048707,307510.0171813965,78722564.3984375,0.00025680760154500604
+125676,527123349504,1.8105022525787353,309504.09009194374,79233047.0635376,0.0002563075686339289
+126606,531024052224,1.8082721614837647,312166.23198723793,79914555.38873291,0.0002556965919211507
+127521,534861840384,1.8071044921875,314556.3413772583,80526423.39257812,0.0002550921344663948
+128381,538468941824,1.8064283752441406,317071.62604546547,81170336.26763916,0.0002545210300013423
+129341,542495473664,1.8118570137023926,319792.94766163826,81866994.6013794,0.0002538801054470241
+129871,544718454784,1.8066246509552002,321223.2166991234,82233143.47497559,0.0002535247476771474
+130616,547843211264,1.805527606010437,323247.12055540085,82751262.86218262,0.0002530234050936997
+131616,552037515264,1.8055236339569092,326068.3877441883,83473507.2625122,0.00025234714848920703
+133056,558077313024,1.8115102243423462,330177.5115237236,84525442.95007324,0.00025136664044111967
+134396,563697680384,1.8067532444000245,334005.05911922455,85505295.13452148,0.0002504472795408219
+135281,567409639424,1.80417941570282,336263.9260504246,86083565.06890869,0.0002498364192433655
+136061,570681196544,1.8022291994094848,338469.36423516273,86648157.24420166,0.00024929564096964896
+137421,576385449984,1.8071019792556762,342423.4924445152,87660414.0657959,0.0002483474381733686
+138276,579971579904,1.8065129804611206,344758.2636282444,88258115.48883057,0.0002477478701621294
+138926,582697877504,1.8003232765197754,346453.48848080635,88692093.05108643,0.0002472903288435191
+139761,586200121344,1.8033902835845947,348963.66530394554,89334698.31781006,0.0002467003359925002
+140466,589157105664,1.8003369140625,350948.2806882858,89842759.85620117,0.0002462002739775926
+141156,592051175424,1.8041338872909547,353014.50160837173,90371712.41174316,0.0002457091468386352
+142386,597210169344,1.8076526641845703,356616.3554201126,91293786.98754883,0.00024482954177074134
+143036,599936466944,1.8042847633361816,358354.11049461365,91738652.2866211,0.0002443625999148935
+144081,604319514624,1.798984398841858,361473.3364651203,92537174.1350708,0.00024360882525797933
+144991,608136331264,1.8027111721038818,363746.11743426323,93119006.06317139,0.0002429494052194059
+145761,611365945344,1.8014432573318482,365923.27353334427,93676358.02453613,0.00024238924379460514
+146921,616231337984,1.802777690887451,369161.1822578907,94505262.65802002,0.00024154159473255277
+147606,619104436224,1.799904360771179,371180.13634061813,95022114.90319824,0.00024103895702864975
+148346,622208221184,1.796466965675354,373459.21619033813,95605559.34472656,0.00024049420608207583
+149326,626318639104,1.7952518892288207,376243.7543103695,96318401.10345459,0.00023977003002073616
+150426,630932373504,1.8000261306762695,379202.4017164707,97075814.8394165,0.00023895344929769635
+151466,635294449664,1.7987578439712524,382167.5741980076,97834898.99468994,0.00023817784676793963
+152481,639551668224,1.8038560628890992,385010.169318676,98562603.34558105,0.0002374175819568336
+153446,643599171584,1.8035392475128174,387753.55447626114,99264909.94592285,0.00023669176152907312
+154376,647499874304,1.8053948736190797,390543.30555176735,99979086.22125244,0.00023598952975589782
+155336,651526406144,1.8094010257720947,393225.47508335114,100665721.62133789,0.00023526183213107288
+156421,656077225984,1.808717474937439,396344.19108366966,101464112.91741943,0.00023443601094186306
+157101,658929352704,1.800858392715454,398201.42544722557,101939564.91448975,0.00023391663853544742
+157981,662620340224,1.7937493753433227,400643.78608202934,102564809.23699951,0.00023324244830291718
+158811,666101612544,1.8010107421875,403040.6518623829,103178406.87677002,0.00023260447778739035
+159611,669457055744,1.7987049579620362,405264.3903121948,103747683.91992188,0.0002319876366527751
+160336,672497926144,1.8018701553344727,407265.4658946991,104259959.26904297,0.00023142702411860228
+161641,677971492864,1.7971449041366576,410591.1344573498,105111330.42108154,0.00023041410895530134
+162541,681746366464,1.8039043569564819,413360.1294693947,105820193.14416504,0.0002297127211932093
+163591,686150385664,1.8012043523788452,416478.45546102524,106618484.59802246,0.00022889155661687255
+164281,689044455424,1.7961911725997926,418587.60516381264,107158426.92193604,0.000228350269026123
+165206,692924186624,1.794422287940979,421477.7675292492,107898308.4874878,0.00022762258595321327
+166126,696782946304,1.7978273344039917,424045.84924674034,108555737.40716553,0.0002268965181428939
+167366,701983883264,1.7981152725219727,427657.40657663345,109480296.08361816,0.00022591432207264006
+168326,706010415104,1.78720947265625,430287.94194436073,110153713.13775635,0.0002251511177746579
+169041,709009342464,1.7941506385803223,432258.2249650955,110658105.59106445,0.00022458111925516278
+169796,712176041984,1.796060562133789,434596.4668176174,111256695.50531006,0.00022397781140170991
+170946,716999491584,1.7885006952285767,438238.4229443073,112189036.27374268,0.00022305604943539947
+172116,721906827264,1.797088713645935,441693.4589443207,113073525.4897461,0.0002221148315584287
+173061,725870444544,1.7923605060577392,444468.4626171589,113783926.42999268,0.00022135215112939477
+173791,728932286464,1.7948624181747437,446910.59599018097,114409112.57348633,0.00022076147433836013
+174796,733147561984,1.7934260940551758,449828.64092206955,115156132.0760498,0.00021994620328769088
+175601,736523976704,1.793698182106018,452460.98252153397,115830011.5255127,0.00021929142531007528
+176591,740676337664,1.7940114498138429,455405.4120359421,116583785.48120117,0.00021848407050129026
+177601,744912584704,1.7944616556167603,458631.7381057739,117409724.95507812,0.0002176580746890977
+178766,749798948864,1.7966568183898926,462002.992303133,118272766.02960205,0.0002167024213122204
+179886,754496569344,1.7952658557891845,465408.72168397903,119144632.75109863,0.00021578080486506224
+180826,758439215104,1.7959301233291627,468060.1989803314,119823410.93896484,0.00021500517323147506
+181901,762948091904,1.7965751028060912,471336.89985489845,120662246.362854,0.00021411580382846296
+182711,766345478144,1.7932391119003297,473834.4921145439,121301629.98132324,0.00021344402921386063
+183546,769847721984,1.790315465927124,476549.98883342743,121996797.14135742,0.0002127500920323655
+184471,773727453184,1.7923971509933472,479378.33332276344,122720853.33062744,0.00021197966998443007
+185281,777124839424,1.7929616165161133,481879.62176418304,123361183.17163086,0.00021130358800292015
+186351,781612744704,1.794197874069214,485144.32627534866,124196947.52648926,0.00021040848514530808
+187561,786687852544,1.7894897413253785,489017.726013422,125188537.85943604,0.0002093935472657904
+188376,790106210304,1.7925705671310426,491489.75262260437,125821376.67138672,0.00020870832668151706
+189496,794803830784,1.798021569252014,494918.69303560257,126699185.41711426,0.00020776463497895747
+190446,798788419584,1.7914582395553589,497918.52365231514,127467142.05499268,0.00020696235878858715
+191431,802919809024,1.7888238143920898,500912.22327923775,128233529.15948486,0.00020612878142856061
+192231,806275252224,1.7891856002807618,503363.70205020905,128861107.72485352,0.0002054505021078512
+193001,809504866304,1.7909438037872314,505660.291626215,129449034.65631104,0.00020479658269323409
+193816,812923224064,1.7878537321090697,508161.10489702225,130089242.8536377,0.0002041033294517547
+194731,816761012224,1.7863385152816773,511053.355697155,130829659.05847168,0.00020332368148956448
+195686,820766572544,1.7863201522827148,514147.6957974434,131621810.12414551,0.00020250846864655614
+196596,824583389184,1.7843420505523682,516835.26493406296,132309827.82312012,0.00020173027587588876
+197316,827603288064,1.7888860940933227,519109.3273601532,132891987.80419922,0.00020111363846808672
+198091,830853873664,1.7861195135116577,521508.52252459526,133506181.76629639,0.00020044897974003106
+199061,834922348544,1.7838082551956176,524697.4795041084,134322554.75305176,0.00019961578072980046
+200081,839200538624,1.7849241256713868,527931.8647477627,135150557.37542725,0.00019873811106663197
+201001,843059298304,1.781349549293518,530598.5689041615,135833233.63946533,0.0001979451917577535
+201926,846939029504,1.7849823999404908,533479.1805455685,136570670.21966553,0.000197146728169173
+202601,849770184704,1.786809048652649,535447.9412600994,137074672.96258545,0.00019656331278383732
+203616,854027403264,1.7845071172714233,538755.1341161728,137921314.33374023,0.0001956848573172465
+204476,857634504704,1.7818892192840576,541460.8675048351,138613982.0812378,0.0001949394791154191
+205661,862604754944,1.780834264755249,544882.3630354404,139489884.93707275,0.00019391084788367152
+206426,865813397504,1.784772605895996,547292.2951827049,140106827.56677246,0.00019324585446156561
+207411,869944786944,1.781447787284851,550311.2109003067,140879669.99047852,0.00019238855747971684
+208386,874034233344,1.7845117664337158,553355.1096127033,141658908.06085205,0.00019153884204570204
+209316,877934936064,1.7804460096359254,556255.9352133274,142401519.41461182,0.00019072731083724648
+210246,881835638784,1.7838315677642822,559155.422185421,143143788.07946777,0.00018991481920238584
+211266,886113828864,1.7828630161285401,562134.9864623547,143906556.5343628,0.00018902262672781944
+212491,891251851264,1.7827301454544067,565977.1395728588,144890147.73065186,0.0001879496849142015
+213266,894502436864,1.7800197172164918,568480.5009863377,145531008.25250244,0.00018727011047303677
+214001,897585250304,1.781924958229065,570723.8563010693,146105307.21307373,0.00018662508227862418
+214816,901003608064,1.7810442113876344,573380.5803165436,146785428.56103516,0.00018590924446471035
+215776,905030139904,1.7792480564117432,576595.4381833076,147608432.17492676,0.00018506527703721076
+217116,910650507264,1.783895535469055,580763.5647056103,148675472.56463623,0.00018388590251561254
+218206,915222298624,1.7796743202209473,584161.3940868378,149545316.88623047,0.00018292547611054033
+218821,917801795584,1.7835624265670775,586042.7060410976,150026932.746521,0.00018238316988572478
+219936,922478444544,1.777990803718567,589289.7597429752,150858178.49420166,0.000181399256689474
+220771,925980688384,1.7799182176589965,591183.597155571,151343000.87182617,0.00018066182383336127
+221776,930195963904,1.7843797159194947,597498.8848052025,152959714.51013184,0.0001797736476873979
+222601,933656264704,1.7795804595947267,599244.4550690651,153406580.49768066,0.00017904405831359327
+223381,936927821824,1.7775488567352296,600881.6357765198,153825698.75878906,0.00017835387552622706
+224351,940996296704,1.7769137907028199,602931.2270174026,154350394.11645508,0.000177495094249025
+225276,944876027904,1.7799413394927979,604898.2233688831,154853945.18243408,0.00017667567590251565
+226281,949091303424,1.7780797576904297,608879.9292657375,155873261.8920288,0.0001757848949637264
+227396,953767952384,1.7739339399337768,611322.3980209827,156498533.89337158,0.00017479606322012842
+228351,957773512704,1.776587553024292,613331.262953043,157012803.315979,0.00017394870519638062
+229076,960814383104,1.7716466903686523,614939.4162056446,157424490.54864502,0.00017330517584923655
+229956,964505370624,1.7750402116775512,639042.4843976498,163594876.00579834,0.00017252382531296462
+230836,968196358144,1.773681025505066,674079.9672315121,172564471.6112671,0.00017174221284221858
+231576,971300143104,1.7731093502044677,675813.0097255707,173008130.4897461,0.0001710847718641162
+232616,975662219264,1.7770291709899901,678184.0271470547,173615110.949646,0.00017016056517604738
+233746,980401782784,1.7675523900985717,683098.4393451214,174873200.47235107,0.00016915611922740936
+234641,984155684864,1.7734109878540039,685042.5715277195,175370898.3110962,0.00016836037684697658
+235651,988391931904,1.773057723045349,687278.392385006,175943268.45056152,0.00016746226174291223
+236816,993278296064,1.7715669631958009,689921.1887881756,176619824.32977295,0.00016642617993056774
+237641,996738596864,1.7745240354537963,691761.8118855953,177091023.8427124,0.00016569241415709257
+238351,999716552704,1.7686384201049805,693362.7359149456,177500860.39422607,0.00016506091924384236
+239566,1004812632064,1.772002272605896,695954.8275504112,178164435.85290527,0.0001639802794670686
+240602,1009157931008,1.7767112471840598,698570.652377367,178834087.00860596,0.00016305888129863888
+241592,1013310291968,1.769608917236328,700894.995667696,179429118.89093018,0.00016217847587540746
+242257,1016099504128,1.7703381299972534,702365.3212473392,179805522.23931885,0.00016158715880010277
+242857,1018616086528,1.7689974737167358,703739.399348259,180157286.2331543,0.00016105370013974607
+243817,1022642618368,1.7718166542053222,705748.6755046844,180671660.92919922,0.00016020028851926327
+244677,1026249719808,1.7722198677062988,707609.003256321,181147904.83361816,0.0001594359491718933
+245822,1031052197888,1.7759829998016357,710156.6984853745,181800114.81225586,0.0001584185811225325
+246987,1035938562048,1.7718148088455201,712665.8724341393,182442463.34313965,0.00015738382353447378
+247827,1039461777408,1.768600239753723,714449.5088384151,182899074.26263428,0.00015663800877518952
+248497,1042271961088,1.7677575302124025,715969.9635241032,183288310.6621704,0.00015604331565555185
+249382,1045983920128,1.7679042387008668,717901.4534947872,183782772.09466553,0.0001552580506540835
+250387,1050199195648,1.7747929763793946,720140.9932420254,184356094.2699585,0.0001543667312944308
+251297,1054016012288,1.764917149543762,722106.4530000687,184859251.96801758,0.00015356004587374628
+252247,1058000601088,1.7661123704910278,724152.5236961842,185383046.06622314,0.00015271836309693754
+253387,1062782107648,1.7647111511230469,726702.471143961,186035832.612854,0.00015170896949712187
+254092,1065739091968,1.7652067375183105,728205.9307692051,186420718.2769165,0.00015108509978745133
+254812,1068758990848,1.7635297060012818,729782.4905691147,186824317.58569336,0.00015044827887322754
+255752,1072701636608,1.764196171760559,731793.3384439945,187339094.6416626,0.0001496173645136878
+256607,1076287766528,1.7638698720932007,733637.4365136623,187811183.74749756,0.00014886210556142032
+257687,1080817614848,1.768958978652954,736014.2164382935,188419639.40820312,0.00014790882414672524
+258502,1084235972608,1.7669451904296876,737910.9490272999,188905202.95098877,0.00014719001774210483
+259262,1087423643648,1.7625267791748047,739466.6771183014,189303469.34228516,0.00014652017853222787
+260217,1091429203968,1.7691507577896117,741740.4932889938,189885566.28198242,0.00014567913603968918
+261152,1095350878208,1.7591209745407104,743739.9462983608,190397426.25238037,0.00014485644351225346
+262292,1100132384768,1.764429850578308,746338.3023200035,191062605.3939209,0.00014385442773345858
+263162,1103781429248,1.7621108627319335,748322.664481163,191570602.10717773,0.00014309051039163023
+263867,1106738413568,1.7602579402923584,749833.4018764496,191957350.8803711,0.00014247202489059418
+264607,1109842198528,1.762261848449707,751597.109572649,192408860.05059814,0.00014182332961354405
+265337,1112904040448,1.759526162147522,753215.5254647732,192823174.51898193,0.0001411839621141553
+266597,1118188863488,1.7672685623168944,756125.3316853046,193568084.911438,0.00014008168363943696
+267532,1122110537728,1.7621809148788452,758342.7242014408,194135737.39556885,0.00013926482643000782
+268487,1126116098048,1.7622956466674804,760522.285752058,194693705.15252686,0.0001384315110044554
+269417,1130016800768,1.762388858795166,762551.8313989639,195213268.83813477,0.00013762101298198104
+270407,1134169161728,1.7595157241821289,764804.6590254307,195789992.71051025,0.00013675936497747898
+271522,1138845810688,1.757354588508606,767161.2214336395,196393272.68701172,0.00013579036749433726
+272377,1142431940608,1.75795738697052,769135.4329416752,196898670.83306885,0.00013504837988875806
+273117,1145535725568,1.7606944990158082,770828.8185880184,197332177.5585327,0.0001344069605693221
+273932,1148954083328,1.7549466705322265,772728.0841300488,197818389.53729248,0.00013370136730372906
+274807,1152624099328,1.754970712661743,774835.183011055,198357806.85083008,0.00013294484233483672
+275477,1155434283008,1.7563565397262573,776427.785389185,198765513.05963135,0.0001323662872891873
+276527,1159838302208,1.7561344528198242,778740.9613273144,199357686.09979248,0.00013146086712367833
+277247,1162858201088,1.7567948341369628,780449.0009460449,199794944.2421875,0.00013084092643111944
+278457,1167933308928,1.7568131160736085,783327.161365509,200531753.3095703,0.00012980082829017192
+279207,1171079036928,1.7493878316879272,785064.701248169,200976563.51953125,0.00012915725528728217
+280067,1174686138368,1.7545626974105835,787161.4134166241,201513321.83465576,0.00012842037540394813
+280847,1177957695488,1.7537751245498656,788885.1290473938,201954593.0361328,0.00012775305367540568
+281792,1181921312768,1.7535089874267578,791144.1982910633,202532914.7625122,0.00012694588804151863
+282527,1185004126208,1.7483990144729615,792868.6627147198,202974377.65496826,0.00012631912250071764
+283242,1188003053568,1.7518246269226074,794551.7392299175,203405245.2428589,0.00012571027036756277
+284042,1191358496768,1.7504371070861817,796430.7740638256,203886278.16033936,0.00012503007019404322
+284962,1195217256448,1.7529449987411498,798542.8514752388,204426969.97766113,0.00012424921442288905
+285932,1199285731328,1.7503743505477904,800732.0254025459,204987398.50305176,0.00012342755508143455
+286977,1203668779008,1.753233289718628,803180.8929703236,205614308.60040283,0.00012254425382707268
+287987,1207905026048,1.7477283954620362,805582.5547001362,206229134.00323486,0.00012169245019322261
+289097,1212560703488,1.750057682991028,808215.335089922,206903125.78302002,0.00012075852282578126
+290688,1219233841152,1.746289751108955,812524.6454071999,208006309.22424316,0.00011942403216380626
+291853,1224120205312,1.7490476608276366,815008.9563155174,208642292.81677246,0.00011845002882182598
+292978,1228838797312,1.7491591882705688,817575.6542432308,209299367.4862671,0.00011751207057386637
+293878,1232613670912,1.7509082746505737,819695.9801375866,209842170.91522217,0.00011676357826218009
+294713,1236115914752,1.7504260921478272,821648.4782016277,210342010.4196167,0.00011607065971475095
+295719,1240335384576,1.7448712022680986,823744.3839178085,210878562.28295898,0.00011523780995048583
+296789,1244823289856,1.750373787879944,825984.0210280418,211451909.3831787,0.0001143543777288869
+297474,1247696388096,1.7485026979446412,827417.0756909847,211818771.3768921,0.00011379012721590698
+298299,1251156688896,1.7513857746124268,829165.2048549652,212266292.4428711,0.00011311194975860417
+299279,1255267106816,1.7491823244094848,831246.0725734234,212798994.5787964,0.00011230834934394807
+300344,1259734040576,1.7427535009384156,833399.7559521198,213350337.52374268,0.00011143753363285214
+301039,1262649081856,1.7455463981628419,834808.7789354324,213711047.4074707,0.00011087067832704633
+302119,1267178930176,1.7416390991210937,837000.7716414928,214272197.54022217,0.000109992062789388
+302754,1269842313216,1.744261622428894,838301.4644382,214605174.8961792,0.00010947677219519392
+303674,1273701072896,1.7468348407745362,840212.588075161,215094422.5472412,0.00010873192513827235
+304569,1277454974976,1.7426301002502442,842025.3971657753,215558501.67443848,0.0001080093061318621
+305629,1281900937216,1.7432089567184448,844223.7876865864,216121289.6477661,0.00010715603275457397
+306539,1285717753856,1.7474224805831908,846121.174690485,216607020.72076416,0.000106425752164796
+307359,1289157083136,1.7466583347320557,847824.3574724197,217043035.51293945,0.00010576950444374233
+308279,1293015842816,1.7465320491790772,849770.5021283627,217541248.54486084,0.00010503528028493747
+309319,1297377918976,1.7457301092147828,851883.0546424389,218082061.98846436,0.00010420793842058629
+310224,1301173764096,1.7416068077087403,853690.4648611546,218544759.00445557,0.00010349031799705699
+310979,1304340463616,1.7464453649520875,855328.2401285172,218964029.4729004,0.00010289331112289801
+312049,1308828368896,1.7403749656677245,857482.9606339931,219515637.92230225,0.00010204986028838903
+312794,1311953125376,1.7413122367858886,859020.4251036644,219909228.8265381,0.00010146444401470944
+313669,1315623141376,1.7418276071548462,860802.2398767471,220365373.40844727,0.00010077882325276732
+314359,1318517211136,1.741833634376526,862207.261920929,220725059.0517578,0.00010023967479355633
+315254,1322271113216,1.7439262866973877,864031.612282753,221192092.74438477,9.954233246389776e-05
+316439,1327241363456,1.7401941537857055,866466.0625932217,221815312.02386475,9.862254228210077e-05
+317594,1332085784576,1.7393305683135987,868843.8917312622,222424036.28320312,9.772992780199274e-05
+318789,1337097977856,1.735496621131897,871252.1925876141,223040561.3024292,9.681050141807646e-05
+319509,1340117876736,1.7347302961349487,872755.9427747726,223425521.3503418,9.625857637729496e-05
+320265,1343288770560,1.7467496991157532,874580.2731509209,223892549.92663574,9.568071982357651e-05
+321180,1347126558720,1.7449059057235718,876753.4169347286,224448874.73529053,9.498363215243444e-05
+321800,1349727027200,1.7405218362808228,878178.1442410946,224813604.9257202,9.451273217564449e-05
+322715,1353564815360,1.7354085969924926,880189.6319646835,225328545.78295898,9.381993004353717e-05
+323880,1358451179520,1.7389711236953735,882752.0865805149,225984534.16461182,9.294157644035295e-05
+324695,1361869537280,1.7336279153823853,884600.4706878662,226457720.49609375,9.232963202521205e-05
+325655,1365896069120,1.7351105260848998,886714.7610874176,226998978.8383789,9.161149500869215e-05
+326905,1371138949120,1.7359595346450805,889553.2458705902,227725630.9428711,9.068083454621956e-05
+327810,1374934794240,1.7329195594787599,891570.9603435993,228242165.84796143,9.00101731531322e-05
+328840,1379254927360,1.7390318965911866,893916.9914896488,228842749.8213501,8.92501266207546e-05
+330425,1385902899200,1.7353940725326538,897621.8574626446,229791195.510437,8.808737038634717e-05
+331335,1389719715840,1.7316031885147094,899737.9237539768,230332908.48101807,8.74235774972476e-05
+332180,1393263902720,1.7348940515518187,901687.4612410069,230831990.07769775,8.680969040142372e-05
+333435,1398527754240,1.7322529554367065,904602.0812029839,231578132.78796387,8.590242214268073e-05
+334440,1402743029760,1.7298837900161743,906809.8383600712,232143318.62017822,8.517977403244004e-05
+335016,1405158948864,1.7256827354431152,908038.849224329,232457945.40142822,8.476718357997015e-05
+335931,1408996737024,1.7300648736953734,909922.6999692917,232940211.19213867,8.411413728026673e-05
+336606,1411827892224,1.7293842363357543,911300.830876112,233293012.70428467,8.363425877178088e-05
+337361,1414994591744,1.7288749599456787,912929.468590498,233709943.95916748,8.30994249554351e-05
+338311,1418979180544,1.7297750091552735,914871.997964859,234207231.4790039,8.242932381108403e-05
+339316,1423194456064,1.7286423635482788,917048.3813505173,234764385.62573242,8.172395610017702e-05
+340161,1426738642944,1.7277494335174561,918803.4824960232,235213691.51898193,8.113370859064162e-05
+341256,1431331405824,1.727659764289856,921161.3741235733,235817311.77563477,8.037270163185894e-05
+341956,1434267418624,1.726815447807312,922619.1768329144,236190509.26922607,7.98885157564655e-05
+342631,1437098573824,1.7283334350585937,924017.1923296452,236548401.23638916,7.942333468236029e-05
+343641,1441334820864,1.7262948656082153,926179.4275717735,237101933.45837402,7.873045979067683e-05
+344671,1445654953984,1.723881149291992,928277.1951835155,237638961.96697998,7.802779146004468e-05
+345736,1450121887744,1.7232005310058593,930507.7530829906,238209984.7892456,7.73054562159814e-05
+346886,1454945337344,1.7255478954315187,932878.4606127739,238816885.91687012,7.653032662346959e-05
+347916,1459265470464,1.7246453189849853,935008.2810490131,239362119.94854736,7.584038394270465e-05
+348811,1463019372544,1.726148428916931,936869.9410157204,239838704.9000244,7.524420652771369e-05
+349646,1466521616384,1.7213160181045533,938576.3196783066,240275537.83764648,7.469081174349412e-05
+350561,1470359404544,1.7229260206222534,940519.2642490864,240772931.6477661,7.408753299387172e-05
+351531,1474427879424,1.7213821077346803,942564.1647593975,241296426.17840576,7.345160702243447e-05
+352506,1478517325824,1.7217609167099,944661.2691597939,241833284.90490723,7.281616126419976e-05
+353286,1481788882944,1.7240455532073975,946261.5434937477,242242955.1343994,7.231054769363254e-05
+354296,1486025129984,1.721031813621521,948370.1250343323,242782752.00878906,7.165947317844257e-05
+355311,1490282348544,1.722795114517212,950490.4979889393,243325567.48516846,7.100934453774244e-05
+356241,1494183051264,1.7186929035186767,952315.2354710102,243792700.2805786,7.041735079837963e-05
+356996,1497349750784,1.7216988754272462,953859.0474669933,244187916.1515503,6.993936403887346e-05
+357981,1501481140224,1.7183150625228882,955918.729688406,244715194.80023193,6.931929965503514e-05
+358930,1505461534720,1.7198513889312743,958704.589379549,245428374.88116455,6.872571248095483e-05

metadata/training_logs/ConvergenceCurve_phase3-annealing.csv

@@ -0,0 +1,401 @@
+training_steps,training_tokens,training_loss,walltime,gputime,learning_rate
+340,1426063360,1.618240556716919,805.9963190555573,206335.05767822266,6.852795922895893e-05
+615,2579496960,1.6049977731704712,1480.500453710556,379008.11614990234,6.836811371613294e-05
+990,4152360960,1.594797191619873,2333.9633750915527,597494.6240234375,6.815014785388485e-05
+1255,5263851520,1.5912671709060668,2961.687971353531,758192.1206665039,6.79961231071502e-05
+1565,6564085760,1.5901156234741212,3699.9308836460114,947182.3062133789,6.781593401683494e-05
+1810,7591690240,1.586198697090149,4255.91148018837,1089513.3389282227,6.767352897441015e-05
+2150,9017753600,1.5839121961593627,5024.551464319229,1286285.1748657227,6.747590668965131e-05
+2500,10485760000,1.5820579528808594,5818.159787654877,1489448.9056396484,6.727247091475874e-05
+2765,11597250560,1.5825859642028808,6431.716231107712,1646519.3551635742,6.711844616802409e-05
+3160,13254000640,1.578509578704834,7345.451099872589,1880435.4815673828,6.688885332550853e-05
+3480,14596177920,1.5791614484786987,8092.022440433502,2071557.7447509766,6.670285802101716e-05
+3775,15833497600,1.5744779205322266,8767.39313364029,2244452.642211914,6.653139280388132e-05
+4060,17028874240,1.5753804922103882,9447.192639112473,2418481.315612793,6.636573380092159e-05
+4310,18077450240,1.5748329401016234,10027.022123336792,2566917.6635742188,6.622042565140873e-05
+4550,19084083200,1.5730882167816163,10576.379052400589,2707553.037414551,6.608092371607199e-05
+4805,20153630720,1.5715075159072875,11158.414714336395,2856554.166870117,6.593271245947108e-05
+5105,21411921920,1.569187421798706,11862.036151885986,3036681.2548828125,6.575833685928956e-05
+5425,22754099200,1.5699743795394898,12592.760034799576,3223746.5689086914,6.557234155479819e-05
+5650,23697817600,1.5652864265441895,13133.071051836014,3362066.1892700195,6.544155621668324e-05
+6035,25312624640,1.5638192701339722,14030.22176027298,3591736.770629883,6.521777686430141e-05
+6330,26549944320,1.5681873321533204,14711.906752586365,3766248.1286621094,6.504631164716557e-05
+6645,27871150080,1.5627027702331544,15456.196798086166,3956786.3803100586,6.486321944976225e-05
+6910,28982640640,1.5610841083526612,16066.988289356232,4113149.0020751953,6.47091947030276e-05
+7165,30052188160,1.5615871906280518,16654.42129969597,4263531.852722168,6.456097617046908e-05
+7425,31142707200,1.5645559978485108,17265.00631093979,4419841.615600586,6.440985453082249e-05
+7785,32652656640,1.5607586336135864,18104.521532058716,4634757.512207031,6.420060526579618e-05
+8070,33848033280,1.5604530096054077,18757.52961397171,4801927.581176758,6.403495353879407e-05
+8295,34791751680,1.5613171815872193,19274.202206134796,4934195.764770508,6.390416820067912e-05
+8535,35798384640,1.5578596353530885,19813.71232509613,5072310.355224609,6.37646735413e-05
+8795,36888903680,1.5551342582702636,20414.72296833992,5226169.0798950195,6.361355190165341e-05
+9025,37853593600,1.5575386190414429,20940.55059313774,5360780.951843262,6.34798634564504e-05
+9400,39426457600,1.5554211330413819,21812.109808921814,5583900.111083984,6.326189759420231e-05
+9715,40747663360,1.556917018890381,22542.035999298096,5770761.2158203125,6.3078805396799e-05
+10070,42236641280,1.5582432985305785,23352.388293743134,5978211.403198242,6.287246651481837e-05
+10295,43180359680,1.5568170738220215,23872.827875614166,6111443.936157227,6.274168845266104e-05
+10605,44480593920,1.5549594163894653,24589.534301757812,6294920.78125,6.256149936234578e-05
+10930,45843742720,1.5545053482055664,25358.808262348175,6491854.915161133,6.237259367480874e-05
+11215,47039119360,1.550887818336487,26025.84851026535,6662617.21862793,6.220694194780663e-05
+11555,48465182720,1.5518133783340453,26791.96462750435,6858742.944641113,6.200931966304779e-05
+11890,49870274560,1.5535324907302857,27571.72452569008,7058361.47857666,6.181460048537701e-05
+12175,51065651200,1.550883536338806,28212.412959337234,7222377.717590332,6.16489487583749e-05
+12425,52114227200,1.5485265445709229,28782.546991109848,7368332.029724121,6.150364060886204e-05
+12730,53393489920,1.5510961532592773,29507.32570052147,7553875.379333496,6.132636190159246e-05
+13015,54588866560,1.549284954071045,30158.78279209137,7720648.394775391,6.116070289863274e-05
+13415,56266588160,1.552486686706543,31101.609414815903,7962012.010192871,6.092820694902912e-05
+13805,57902366720,1.552199697494507,31992.497676610947,8190079.405212402,6.070152448955923e-05
+14180,59475230720,1.5470599460601806,32886.17706298828,8418861.328125,6.048355498933233e-05
+14450,60607692800,1.5489162635803222,33514.932136535645,8579822.626953125,6.032661985955201e-05
+14695,61635297280,1.5477478408813476,34075.356810092926,8723291.343383789,6.018421481712721e-05
+15000,62914560000,1.543528437614441,34807.71875858307,8910776.002197266,6.0006936109857634e-05
+15305,64193822720,1.5468650436401368,35514.63099241257,9091745.534057617,5.9829657402588055e-05
+15680,65766686720,1.540996379852295,36391.95647358894,9316340.85723877,5.961169154033996e-05
+15965,66962063360,1.5399002981185914,37070.284098386765,9489992.729187012,5.944603981333785e-05
+16300,68367155200,1.5444617700576782,37853.563311100006,9690512.207641602,5.925132063566707e-05
+16670,69919047680,1.5386622858047485,38695.66263437271,9906089.634399414,5.9036261518485844e-05
+16910,70925680640,1.5419238090515137,39262.56024956703,10051215.42388916,5.889676322112791e-05
+17165,71995228160,1.5399140405654907,39862.66890287399,10204843.239135742,5.874854832654819e-05
+17425,73085747200,1.5421667003631592,40465.11421442032,10359069.238891602,5.859742304892279e-05
+17690,74197237760,1.5381574249267578,41070.201156139374,10513971.49597168,5.8443394664209336e-05
+17945,75266785280,1.5403591775894165,41660.75303196907,10665152.776184082,5.8295176131650805e-05
+18275,76650905600,1.5373179054260253,42444.894955158234,10865893.108520508,5.81033673370257e-05
+18510,77636567040,1.5400733470916748,42997.210401296616,11007285.862731934,5.7966775784734637e-05
+18750,78643200000,1.5406368732452393,43574.14296269417,11154980.598449707,5.7827277487376705e-05
+19035,79838576640,1.5359001493453979,44288.694413900375,11337905.769958496,5.7661622122395784e-05
+19325,81054924800,1.5356152725219727,44993.05548071861,11518222.203063965,5.74930636503268e-05
+19595,82187386880,1.537904191017151,45617.38076686859,11678049.47631836,5.7336124882567674e-05
+19770,82921390080,1.534038815498352,46021.857204437256,11781595.444335938,5.723440699512139e-05
+20040,84053852160,1.5376662874221803,46687.448335886,11951986.773986816,5.707747186534107e-05
+20280,85060485120,1.5351179456710815,47294.76152420044,12107458.950195312,5.6937973567983136e-05
+20535,86130032640,1.5340012741088866,47921.1394572258,12267811.701049805,5.678975867340341e-05
+20860,87493181440,1.5338728380203248,48772.13713765144,12485667.10723877,5.660085298586637e-05
+21095,88478842880,1.5385234260559082,49363.669286966324,12637099.337463379,5.64642614335753e-05
+21370,89632276480,1.532801251411438,50073.3399438858,12818775.025634766,5.6304419558728114e-05
+21645,90785710080,1.5361132955551147,50762.75538563728,12995265.378723145,5.6144577683880925e-05
+22020,92358574080,1.531789288520813,51738.95875930786,13245173.442382812,5.592661182163283e-05
+22385,93889495040,1.5336677026748657,52670.29964208603,13483596.708374023,5.571445944951847e-05
+22695,95189729280,1.5282061100006104,53473.869824409485,13689310.675048828,5.5534273997182027e-05
+23015,96531906560,1.5331871032714843,54302.397018671036,13901413.636779785,5.534827505471185e-05
+23305,97748254720,1.5322805833816528,55052.88818240166,14093539.374694824,5.517971658264287e-05
+23605,99006545920,1.5294403648376464,55848.05881381035,14297103.05633545,5.500534098246135e-05
+23870,100118036480,1.531037163734436,56520.91497302055,14469354.233093262,5.485131259774789e-05
+24135,101229527040,1.5261157083511352,57187.208659648895,14639925.416870117,5.4697284213034436e-05
+24385,102278103040,1.5261563014984132,57812.448143959045,14799986.724853516,5.455197242554277e-05
+24705,103620280320,1.5302351379394532,58628.562000989914,15008911.872253418,5.4365973483072594e-05
+24955,104668856320,1.529397587776184,59267.39612221718,15172453.407287598,5.4220665333559737e-05
+25220,105780346880,1.5290697193145752,59945.24328708649,15345982.28149414,5.4066633310867473e-05
+25580,107290296320,1.526914210319519,60849.63791370392,15577507.305908203,5.385738768381998e-05
+25895,108611502080,1.524157953262329,61658.11358332634,15784477.077331543,5.367429548641667e-05
+26235,110037565440,1.5237834024429322,62533.954092264175,16008692.247619629,5.347667320165783e-05
+26555,111379742720,1.5251441431045532,63354.29300546646,16218699.009399414,5.3290674259187654e-05
+26905,112847749120,1.5265145874023438,64249.2104947567,16447797.886657715,5.308724212227389e-05
+27185,114022154240,1.5199985694885254,64974.279589653015,16633415.574951172,5.292449350235984e-05
+27475,115238502400,1.527224998474121,65715.40770769119,16823144.373168945,5.275593139231205e-05
+27760,116433879040,1.5214627742767335,66476.47451162338,17017977.474975586,5.2590279665309936e-05
+28075,117755084800,1.525670132637024,67291.38590097427,17226594.790649414,5.2407187467906624e-05
+28370,118992404480,1.5220741653442382,68019.16993141174,17412907.502441406,5.223571861279197e-05
+28775,120691097600,1.5192509555816651,69026.66728496552,17670826.824951172,5.200031591812149e-05
+29050,121844531200,1.5173545169830323,69711.20947599411,17846069.625854492,5.18404740432743e-05
+29335,123039907840,1.5229442358016967,70404.63313817978,18023586.083374023,5.167481867829338e-05
+29630,124277227520,1.5194816541671754,71168.1383061409,18219043.40637207,5.150335346115753e-05
+29885,125346775040,1.5225177240371703,71798.30358982086,18380365.71899414,5.1355134928599e-05
+30210,126709923840,1.521646456718445,72616.72042250633,18589880.42816162,5.116623287904076e-05
+30465,127779471360,1.5177456331253052,73263.86911273003,18755550.492858887,5.101801434648223e-05
+30735,128911933440,1.5173807764053344,73955.50183176994,18932608.468933105,5.086107921670191e-05
+31050,130233139200,1.5206219005584716,74767.35306620598,19140442.38494873,5.06779870192986e-05
+31415,131764060160,1.5155780935287475,75706.64231777191,19380900.43334961,5.046583464718424e-05
+31665,132812636160,1.5219171285629272,76349.90894556046,19545576.690063477,5.0320522859692574e-05
+32045,134406471680,1.5170358562469481,77302.66745686531,19789482.86895752,5.0099650252377614e-05
+32300,135476019200,1.5156394338607788,77972.0146214962,19960835.743103027,4.995143535779789e-05
+32615,136797224960,1.5168628215789794,78756.26611709595,20161604.125976562,4.976834316039458e-05
+32940,138160373760,1.5172155475616456,79585.85103392601,20373977.86468506,4.957944111083634e-05
+33240,139418664960,1.5123344993591308,80379.44974303246,20577139.13421631,4.940506551065482e-05
+33580,140844728320,1.5176030588150025,81217.35864281654,20791643.812561035,4.9207443225895986e-05
+33890,142144962560,1.5181758880615235,82003.55335474014,20992909.658813477,4.902725777355954e-05
+34165,143298396160,1.5155006074905395,82727.45544552803,21178228.594055176,4.886741589871235e-05
+34475,144598630400,1.5112392854690553,83544.65241909027,21387431.01928711,4.8687230446375906e-05
+34820,146045665280,1.5120015907287598,84422.05125689507,21612045.121765137,4.8486701416550204e-05
+35125,147324928000,1.511203556060791,85196.82748556137,21810387.83630371,4.8309422709280625e-05
+35400,148478361600,1.511547703742981,85872.22816109657,21983290.409240723,4.8149580834433436e-05
+35700,149736652800,1.513823585510254,86634.15090918541,22178342.632751465,4.7975208872230724e-05
+36005,151015915520,1.5114430284500122,87417.37316012383,22378847.5289917,4.7797930164961144e-05
+36270,152127406080,1.5147535657882691,88070.50267338753,22546048.684387207,4.764390178024769e-05
+36615,153574440960,1.5105689668655395,88966.28969430923,22775370.161743164,4.7443372750421986e-05
+36900,154769817600,1.51490469455719,89686.37483000755,22959711.956481934,4.7277717385441065e-05
+37200,156028108800,1.5133596897125243,90450.27981996536,23155271.633911133,4.710334542323835e-05
+37465,157139599360,1.5079569578170777,91108.37686300278,23323744.47692871,4.6949317038524896e-05
+37780,158460805120,1.5125086879730225,91913.1192946434,23529758.53942871,4.6766224841121584e-05
+38110,159844925440,1.50558678150177,92761.00912809372,23746818.336791992,4.6574412408517674e-05
+38385,160998359040,1.51372416973114,93450.92206478119,23923436.048583984,4.6414570533670485e-05
+38700,162319564800,1.5078394937515258,94224.86657977104,24121565.844421387,4.623148197424598e-05
+38970,163452026880,1.506821026802063,94925.99536538124,24301054.813537598,4.607454320648685e-05
+39270,164710318080,1.510329613685608,95708.0318236351,24501256.146850586,4.590017124428414e-05
+39580,166010552320,1.5082668447494507,96510.14937281609,24706598.239440918,4.571998579194769e-05
+39820,167017185280,1.5053207445144654,97089.1982126236,24854834.74243164,4.558048749458976e-05
+40085,168128675840,1.5021627807617188,97724.10911917686,25017371.934509277,4.5426459109876305e-05
+40395,169428910080,1.5078684282302857,98467.72787332535,25207738.33557129,4.524627365753986e-05
+40650,170498457600,1.5049097061157226,99065.08206963539,25360661.00982666,4.509805512498133e-05
+40990,171924520960,1.504285216331482,99840.21076202393,25559093.955078125,4.490043284022249e-05
+41225,172910182400,1.506738648414612,100397.13245034218,25701665.907287598,4.476384128793143e-05
+41555,174294302720,1.5037948179244995,101157.30819225311,25896270.897216797,4.4572032493306324e-05
+41810,175363850240,1.507575240135193,101747.4647963047,26047350.987854004,4.442381396074779e-05
+42115,176643112960,1.5017842960357666,102464.62705016136,26230944.52484131,4.4246535253478214e-05
+42395,177817518080,1.5062606859207153,103136.38542175293,26402914.66796875,4.408378663356416e-05
+42680,179012894720,1.5022754383087158,103820.6661593914,26578090.5368042,4.3918134906562045e-05
+43035,180501872640,1.5056888008117675,104653.6878824234,26791344.09790039,4.371179238660261e-05
+43375,181927936000,1.5049325942993164,105455.63975095749,26996643.776245117,4.3514170101843774e-05
+43705,183312056320,1.5028763580322266,106236.37662887573,27196512.416992188,4.332236130721867e-05
+43965,184402575360,1.5004971170425414,106841.67947530746,27351469.94567871,4.3171236029593274e-05
+44300,185807667200,1.5000527667999268,107599.03004646301,27545351.69189453,4.2976520489901304e-05
+44635,187212759040,1.5022150135040284,108396.13827347755,27749411.398010254,4.2781804950209334e-05
+44880,188240363520,1.5062885856628419,108970.39048075676,27896419.96307373,4.2639399907784536e-05
+45170,189456711680,1.5018709230422973,109648.80755186081,28070094.733276367,4.247083779773675e-05
+45580,191176376320,1.4989476108551025,110604.36952710152,28314718.59893799,4.22325283579994e-05
+45850,192308838400,1.493067693710327,111237.1611392498,28476713.25164795,4.207559322821908e-05
+46165,193630044160,1.4996389389038085,111963.59861969948,28662681.246643066,4.1892501030815765e-05
+46435,194762506240,1.500433030128479,112603.90734314919,28826600.27984619,4.173556590103544e-05
+46650,195664281600,1.4946138525009156,113108.95254850388,28955891.852416992,4.1610597691033036e-05
+46960,196964515840,1.4971131420135497,113825.1804459095,29139246.194152832,4.143041223869659e-05
+47215,198034063360,1.4937334489822387,114435.3586423397,29295451.812438965,4.1282197344116867e-05
+47470,199103610880,1.4962233591079712,115072.24508047104,29458494.740600586,4.1133978811558336e-05
+47740,200236072960,1.4953879356384276,115708.65568161011,29621415.854492188,4.0977043681778014e-05
+48035,201473392640,1.4945304012298584,116384.64575529099,29794469.313354492,4.080557846464217e-05
+48260,202417111040,1.4929394960403441,116917.8391110897,29930966.812438965,4.067479676450603e-05
+48540,203591516160,1.495974998474121,117569.86111283302,30097884.444885254,4.051205178257078e-05
+48865,204954664960,1.4918129301071168,118329.11504745483,30292253.452148438,4.0323146095033735e-05
+49080,205856440320,1.4982340002059937,118833.6501467228,30421414.437561035,4.019817788503133e-05
+49370,207072788480,1.5003968715667724,119512.44303226471,30595185.416259766,4.002961941296235e-05
+49690,208414965760,1.4929927015304565,120253.42932701111,30784877.907714844,3.984362047049217e-05
+50005,209736171520,1.4936941194534301,120972.4400138855,30968944.643554688,3.966052827308886e-05
+50300,210973491200,1.4882380294799804,121647.13416337967,31141666.345825195,3.948906305595301e-05
+50550,212022067200,1.486804165840149,122218.50893497467,31287938.287353516,3.934375126846135e-05
+50810,213112586240,1.4881820821762084,122820.96964716911,31442168.229675293,3.919262962881476e-05
+51100,214328934400,1.4913074493408203,123471.17196083069,31608620.021972656,3.902406751876697e-05
+51335,215314595840,1.498164072036743,124019.37543749809,31748960.11199951,3.8887475966475904e-05
+51665,216698716160,1.4910090827941895,124821.46028637886,31954293.83331299,3.86956671718508e-05
+51950,217894092800,1.4903889608383178,125483.5426557064,32123786.91986084,3.853001180686988e-05
+52240,219110440960,1.496640019416809,126171.77869081497,32299975.344848633,3.836144969682209e-05
+52575,220515532800,1.494093623161316,126951.69672012329,32499634.360351562,3.816673415713012e-05
+52870,221752852480,1.4892810773849487,127615.13600826263,32669474.818115234,3.7995268939994276e-05
+53160,222969200640,1.4859894609451294,128297.73133397102,32844219.221496582,3.782670682994649e-05
+53375,223870976000,1.4920428323745727,128778.67433786392,32967340.630493164,3.770174225792289e-05
+53640,224982466560,1.490599317550659,129393.81948709488,33124817.78869629,3.7547710235230625e-05
+53960,226324643840,1.4861323928833008,130156.15509748459,33319975.704956055,3.7361714930739254e-05
+54300,227750707200,1.4913449001312256,130951.43420100212,33523567.155456543,3.716409264598042e-05
+54595,228988026880,1.4833179426193237,131635.60251832008,33698714.24468994,3.6992623790865764e-05
+54850,230057574400,1.487194871902466,132223.70856165886,33849269.39178467,3.684440889628604e-05
+55185,231462666240,1.491267008781433,133005.1627588272,34049321.666259766,3.6649689718615264e-05
+55465,232637071360,1.4909626865386962,133658.57569098473,34216595.37689209,3.648694109870121e-05
+55750,233832448000,1.4853796434402466,134339.60628509521,34390939.208984375,3.6321289371699095e-05
+56005,234901995520,1.4857091856002809,134944.86699581146,34545885.950927734,3.6173070839140564e-05
+56260,235971543040,1.483737931251526,135523.6685371399,34694059.14550781,3.602485594456084e-05
+56515,237041090560,1.4861033582687377,136111.21580958366,34844471.24725342,3.587663741200231e-05
+56845,238425210880,1.4830087089538575,136868.66960287094,35038379.41833496,3.568482861737721e-05
+57135,239641559040,1.4836278200149535,137566.16293907166,35216937.712402344,3.551626650732942e-05
+57470,241046650880,1.485690279006958,138331.76188111305,35412931.04156494,3.532155096763745e-05
+57760,242262999040,1.4853268766403198,138988.45338225365,35581044.06585693,3.515298885758966e-05
+58205,244129464320,1.4833063077926636,140067.45882749557,35857269.45983887,3.4894335840363055e-05
+58445,245136097280,1.4823443174362183,140625.04994630814,36000012.78625488,3.4754837543005124e-05
+58750,246415360000,1.4824813842773437,141325.74165940285,36179389.86480713,3.4577558835735545e-05
+59070,247757537280,1.4827831983566284,142058.24861884117,36366911.64642334,3.4391563531244174e-05
+59365,248994856960,1.4806864595413207,142746.6637802124,36543145.927734375,3.422009467612952e-05
+59645,250169262080,1.4780278778076172,143398.50452661514,36710017.15881348,3.4057346056215465e-05
+59870,251112980480,1.4825284004211425,143913.37279200554,36841823.43475342,3.392656799405813e-05
+60195,252476129280,1.4798415756225587,144675.52619338036,37036934.70550537,3.373766230652109e-05
+60485,253692477440,1.4777485179901122,145348.05514454842,37209102.117004395,3.356910383445211e-05
+60780,254929797120,1.4796476650238037,146030.58737277985,37383830.36743164,3.3397634979337454e-05
+61155,256502661120,1.480228796005249,146908.87734293938,37608672.59979248,3.317966911708936e-05
+61455,257760952320,1.4823005390167237,147605.4583489895,37786997.33734131,3.300529715488665e-05
+61695,258767585280,1.4817911291122436,148163.06012272835,37929743.39141846,3.2865798857528716e-05
+61990,260004904960,1.4828647470474243,148848.27455759048,38105158.286743164,3.269433364039287e-05
+62255,261116395520,1.4760107707977295,149463.70156526566,38262707.60070801,3.2540305255679414e-05
+62505,262164971520,1.4802721977233886,150039.92844581604,38410221.682128906,3.239499346818775e-05
+62750,263192576000,1.4781893348693849,150612.27514147758,38556742.43621826,3.225258842576295e-05
+63040,264408924160,1.4819868993759155,151304.97952342033,38734074.757995605,3.2084026315715164e-05
+63440,266086645760,1.4778487586975098,152211.1390864849,38966051.60614014,3.185153036611155e-05
+63675,267072307200,1.4757684564590454,152758.44339370728,39106161.50878906,3.171493881382048e-05
+64055,268666142720,1.4742608451843262,153643.5551381111,39332750.115356445,3.149406620650552e-05
+64430,270239006720,1.4781750535964966,154527.9770743847,39559162.13104248,3.127610034425743e-05
+64750,271581184000,1.4749154949188232,155266.63118600845,39748257.583618164,3.109010140178725e-05
+65105,273070161920,1.4730145406723023,156085.60028910637,39957913.67401123,3.0883762519806623e-05
+65445,274496225280,1.4764969491958617,156891.59435510635,40164248.15490723,3.068614023504779e-05
+65695,275544801280,1.4745435762405394,157459.66235876083,40309673.56384277,3.054082844755612e-05
+66015,276886978560,1.4767353296279908,158198.5747241974,40498835.12939453,3.0354829505085945e-05
+66315,278145269760,1.4752719068527222,158901.02135300636,40678661.46636963,3.0180457542883232e-05
+66710,279802019840,1.473205132484436,159820.03469586372,40913928.88214111,2.9950866519357078e-05
+66995,280997396480,1.4723931550979614,160476.88212037086,41082081.82281494,2.9785211154376157e-05
+67295,282255687680,1.47007230758667,161191.35992765427,41264988.14147949,2.9610839192173444e-05
+67540,283283292160,1.4693548011779785,161760.2142548561,41410614.849243164,2.9468434149748646e-05
+67955,285023928320,1.4752857398986816,162760.50145626068,41666688.372802734,2.922721796494443e-05
+68270,286345134080,1.4734775400161744,163467.50890016556,41847682.27844238,2.904412758653052e-05
+68515,287372738560,1.4698681640625,164039.26407814026,41994051.604003906,2.8901722544105724e-05
+68795,288547143680,1.470857663154602,164711.34090185165,42166103.27087402,2.873897392419167e-05
+69075,289721548800,1.4690834808349609,165361.93133115768,42332654.42077637,2.8576227123267017e-05
+69370,290958868480,1.4690316343307495,166044.55415010452,42507405.86242676,2.8404760087141767e-05
+69680,292259102720,1.4698637628555298,166765.3598856926,42691932.130737305,2.822457463480532e-05
+70015,293664194560,1.4656069946289063,167535.41150689125,42889065.34576416,2.8029859095113352e-05
+70295,294838599680,1.4658060693740844,168203.70962262154,43060149.66339111,2.7867110475199297e-05
+70615,296180776960,1.4710540628433229,168979.6358883381,43258786.78741455,2.7681113351718523e-05
+70955,297606840320,1.469628415107727,169777.67864346504,43463085.73272705,2.7483489247970283e-05
+71240,298802216960,1.4723303127288818,170448.73286485672,43634875.61340332,2.7317835701978765e-05
+71635,300458967040,1.4629393482208253,171377.95411467552,43872756.25335693,2.708824467845261e-05
+71850,301360742400,1.468685245513916,171882.100212574,44001817.654418945,2.6963278287439607e-05
+72165,302681948160,1.469798617362976,172615.52517652512,44189574.44519043,2.6780186090036295e-05
+72395,303646638080,1.4712815189361572,173153.7472462654,44327359.295043945,2.6646499463822693e-05
+72645,304695214080,1.464143590927124,173731.09820842743,44475161.14135742,2.6501189495320432e-05
+72915,305827676160,1.4653117704391478,174352.08979010582,44634134.98626709,2.634425436554011e-05
+73215,307085967360,1.4687677574157716,175055.78487992287,44814280.929260254,2.6169880584347993e-05
+73420,307945799680,1.4661717796325684,175527.70881414413,44935093.4564209,2.6050725864479318e-05
+73750,309329920000,1.4686805772781373,176280.46760320663,45127799.7064209,2.585891525086481e-05
+74060,310630154240,1.4662387561798096,176974.81370449066,45305552.30834961,2.567873161751777e-05
+74405,312077189120,1.4684185075759888,177774.08547639847,45510165.88195801,2.5478202587692067e-05
+74660,313146736640,1.470201678276062,178373.23720812798,45663548.72528076,2.532998587412294e-05
+74935,314300170240,1.4661515903472901,179018.95270824432,45828851.89331055,2.517014399927575e-05
+75245,315600404480,1.4673150062561036,179748.42839598656,46015597.66937256,2.4989958546939306e-05
+75580,317005496320,1.4635962295532225,180513.01770210266,46211332.53173828,2.4795241188257933e-05
+75860,318179901440,1.4616325998306274,181166.42831087112,46378605.64758301,2.463249438733328e-05
+76155,319417221120,1.4656786823272705,181845.31468749046,46552400.55999756,2.446102735120803e-05
+76410,320486768640,1.463839659690857,182431.91593766212,46702570.480041504,2.4312810637638904e-05
+76720,321787002880,1.4636628484725953,183150.79497933388,46886603.51470947,2.413262518530246e-05
+77035,323108208640,1.4582185935974121,183895.1369035244,47077155.047302246,2.3949532987899147e-05
+77340,324387471360,1.4637500190734862,184588.93916249275,47254768.425598145,2.3772254280629568e-05
+77630,325603819520,1.462220973968506,185265.27911663055,47427911.45385742,2.3603693989571184e-05
+77890,326694338560,1.4622275257110595,185858.72430443764,47579833.421936035,2.345257053093519e-05
+78300,328414003200,1.4603347873687744,186851.575268507,47834003.26873779,2.321426109119784e-05
+78640,329840066560,1.4623171043395997,187642.81193494797,48036559.85534668,2.3016638806439005e-05
+78945,331119329280,1.4616960954666138,188349.22605633736,48217401.87042236,2.2839358280180022e-05
+79280,332524421120,1.459079713821411,189150.4116165638,48422505.37384033,2.2644642740488052e-05
+79610,333908541440,1.460066180229187,189929.30950641632,48621903.23364258,2.2452832126873545e-05
+79940,335292661760,1.458433895111084,190710.2808189392,48821831.88964844,2.2261023332248442e-05
+80200,336383180800,1.4602740097045899,191315.44288516045,48976753.378601074,2.210989987361245e-05
+80535,337788272640,1.4585590028762818,192075.24276852608,49171262.148742676,2.1915182514931075e-05
+80745,338669076480,1.4575935792922974,192565.70645332336,49296820.85205078,2.1793121049995534e-05
+80995,339717652480,1.4569976568222045,193166.014264822,49450499.65179443,2.1647811081493273e-05
+81285,340934000640,1.4563476514816285,193841.25965118408,49623362.470703125,2.147925079043489e-05
+81690,342632693760,1.4594932317733764,194807.54062771797,49870730.4006958,2.1243848095764406e-05
+81990,343890984960,1.4542385864257812,195511.27005696297,50050885.13458252,2.106947431457229e-05
+82260,345023447040,1.4538869619369508,196146.32533311844,50213459.28527832,2.0912539184791967e-05
+82540,346197852160,1.4536242580413818,196820.66618442535,50386090.54321289,2.0749790564877912e-05
+82845,347477114880,1.4568992137908936,197527.9463839531,50567154.27429199,2.0572511857608333e-05
+83130,348672491520,1.461669783592224,198197.2515809536,50738496.40472412,2.0406858311616816e-05
+83380,349721067520,1.4520815563201905,198790.98207473755,50890491.41113281,2.026154652412515e-05
+83700,351063244800,1.4519271898269652,199516.74213290215,51076285.98602295,2.0075549400644377e-05
+84005,352342507520,1.4575488901138305,200226.67574048042,51258028.98956299,1.9898270693374798e-05
+84300,353579827200,1.4544986629486083,200895.65742373466,51429288.300476074,1.9726803657249548e-05
+84600,354838118400,1.45297345161438,201623.05980324745,51615503.30963135,1.955242987605743e-05
+84850,355886694400,1.4539190101623536,202193.24289917946,51761470.18218994,1.940711990755517e-05
+85090,356893327360,1.4516407442092896,202735.50936222076,51900290.396728516,1.926762161019724e-05
+85440,358361333760,1.4553143405914306,203517.53140807152,52100488.04046631,1.906418583530467e-05
+85775,359766425600,1.454245114326477,204292.7538573742,52298944.98748779,1.88694702956127e-05
+86045,360898887680,1.4581734180450439,204924.96027827263,52460789.83123779,1.8712535165832378e-05
+86370,362262036480,1.4495676803588866,205714.482524395,52662907.52624512,1.8523631297284737e-05
+86740,363813928960,1.4492269659042358,206592.34785556793,52887641.05102539,1.8308570361114107e-05
+87085,365260963840,1.4531880235671997,207404.72061872482,53095608.478393555,1.8108041331288405e-05
+87520,367085486080,1.4498338222503662,208419.38573479652,53355362.74810791,1.785520180419553e-05
+87775,368155033600,1.449286036491394,209015.15674829483,53507880.12756348,1.7706985090626404e-05
+88035,369245552640,1.445965623855591,209624.789083004,53663946.00524902,1.755586163199041e-05
+88260,370189271040,1.4476247262954711,210147.98101305962,53797883.13934326,1.7425081750843674e-05
+88515,371258818560,1.451582098007202,210752.92172312737,53952747.961120605,1.7276865037274547e-05
+88790,372412252160,1.4487579679489135,211389.40759897232,54115688.345336914,1.711702316242736e-05
+89110,373754429440,1.4427510118484497,212160.9178698063,54313194.97467041,1.693102421995718e-05
+89380,374886891520,1.4471882104873657,212785.03998875618,54472970.23712158,1.6774089090176858e-05
+89655,376040325120,1.447092981338501,213445.27336359024,54641989.9810791,1.661424721532967e-05
+89965,377340559360,1.4516633462905884,214169.31757092476,54827345.29815674,1.6434061762993224e-05
+90245,378514964480,1.4478979873657227,214824.39868569374,54995046.0635376,1.6271314962068573e-05
+90585,379941027840,1.4449454545974731,215616.49609565735,55197823.00048828,1.6073690858320333e-05
+90840,381010575360,1.4462517023086547,216213.5972893238,55350680.906066895,1.5925474144751206e-05
+91095,382080122880,1.4513818311691284,216805.89086055756,55502308.060302734,1.577725743118208e-05
+91345,383128698880,1.4503194522857665,217378.04878234863,55648780.48828125,1.5631947462679818e-05
+91675,384512819200,1.441401767730713,218140.78364372253,55844040.61279297,1.544013684906531e-05
+91995,385854996480,1.4448133182525635,218864.7469806671,56029375.22705078,1.5254138816089835e-05
+92280,387050373120,1.4494926691055299,219533.4955906868,56200574.87121582,1.5088485270098317e-05
+92635,388539351040,1.442607936859131,220371.54222488403,56415114.80957031,1.4882143659633584e-05
+92900,389650841600,1.4453481721878052,220998.7774169445,56575687.01873779,1.4728114365425427e-05
+93095,390468730880,1.4452113437652587,221446.39208316803,56690276.373291016,1.4614772226195782e-05
+93380,391664107520,1.4432289028167724,222113.27166557312,56860997.54638672,1.4449117770709563e-05
+93620,392670740480,1.4445120525360107,222671.36609077454,57003869.71923828,1.4309619473351631e-05
+93880,393761259520,1.443903260231018,223281.7130446434,57160118.53942871,1.4158496014715638e-05
+94205,395124408320,1.445061569213867,224047.59836912155,57356185.18249512,1.3969592146167997e-05
+94475,396256870400,1.4411702537536621,224676.60982489586,57517212.11517334,1.3812657016387675e-05
+94765,397473218560,1.442785415649414,225343.2132947445,57687862.60345459,1.364409581583459e-05
+95065,398731509760,1.4441711521148681,226041.96816682816,57866743.85070801,1.3469722944137175e-05
+95350,399926886400,1.4450367927551269,226705.79229354858,58036682.82714844,1.3304069398145657e-05
+95635,401122263040,1.4430587100982666,227366.83624863625,58205910.07965088,1.3138414942659438e-05
+96015,402716098560,1.442037172317505,228247.7006161213,58431411.35772705,1.2917542335344478e-05
+96355,404142161920,1.4418480587005615,229056.66600227356,58638506.49658203,1.2719920050585642e-05
+96720,405673082880,1.4404057931900025,229924.46289420128,58860662.50091553,1.250776676897658e-05
+96990,406805544960,1.4440303468704223,230542.48742699623,59018876.781311035,1.2350830729701556e-05
+97355,408336465920,1.438180251121521,231375.35725951195,59232091.45843506,1.2138677448092494e-05
+97575,409259212800,1.439077558517456,231882.92070150375,59362027.69958496,1.2010803402517922e-05
+97905,410643333120,1.438628978729248,232649.20167446136,59558195.62866211,1.1818993698398117e-05
+98225,411985510400,1.4397692346572877,233372.72338867188,59743417.1875,1.163299566542264e-05
+98525,413243801600,1.436825532913208,234065.885130167,59920866.593322754,1.1458622793725226e-05
+98810,414439178240,1.4378496599197388,234738.96312594414,60093174.5602417,1.1292968338239007e-05
+99075,415550668800,1.435820026397705,235371.6471362114,60255141.66687012,1.1138939044030849e-05
+99390,416871874560,1.4408235406875611,236099.70878958702,60441525.45013428,1.0955847756122239e-05
+99690,418130165760,1.4354854536056518,236815.7348573208,60624828.12347412,1.0781474884424824e-05
+100065,419703029760,1.4379744958877563,237705.01027584076,60852482.630615234,1.0563509022176731e-05
+100325,420793548800,1.4341013383865358,238303.8801074028,61005793.30749512,1.0412385563540738e-05
+100570,421821153280,1.4373250198364258,238865.27826070786,61149511.23474121,1.0269981430610642e-05
+100840,422953615360,1.4392485189437867,239486.5515244007,61308557.19024658,1.0113045391335618e-05
+101170,424337735680,1.438763918876648,240252.76848316193,61504708.73168945,9.921235687215813e-06
+101445,425491169280,1.439057412147522,240885.06843209267,61666577.51861572,9.761393812368624e-06
+101725,426665574400,1.4349830961227417,241555.94319343567,61838321.45751953,9.598645192454569e-06
+102025,427923865600,1.4397229623794556,242284.0287578106,62024711.36199951,9.424272320757154e-06
+102260,428909527040,1.4354581880569457,242823.4775042534,62162810.24108887,9.287680768466089e-06
+102520,430000046080,1.4331795167922974,243445.43759346008,62322032.02392578,9.136557309830096e-06
+102810,431216394240,1.4352128314971924,244119.4434416294,62494577.52105713,8.967997018771712e-06
+103060,432264970240,1.4369250345230102,244697.56983041763,62642577.876586914,8.82268614077475e-06
+103385,433628119040,1.4285495948791505,245475.68335533142,62841774.938964844,8.633782272227108e-06
+103655,434760581120,1.4313248443603515,246102.15421915054,63002151.48010254,8.476847142446786e-06
+103930,435914014720,1.434038095474243,246752.50554800034,63168641.420288086,8.317005267599598e-06
+104190,437004533760,1.431107611656189,247339.81801509857,63318993.411865234,8.165881808963604e-06
+104510,438346711040,1.4295264863967896,248075.56574225426,63507344.83001709,7.979883775988128e-06
+104800,439563059200,1.4344947957992553,248759.19261574745,63682353.30963135,7.811323484929744e-06
+105130,440947179520,1.433149938583374,249534.01866197586,63880708.77746582,7.619513326062588e-06
+105395,442058670080,1.429944167137146,250154.07029986382,64039441.99676514,7.46548403185443e-06
+105695,443316961280,1.4318817663192749,250851.6952586174,64218033.986206055,7.2911111601570155e-06
+105985,444533309440,1.4361343097686767,251532.3388671875,64392278.75,7.122550869098632e-06
+106210,445477027840,1.4285934686660766,252057.66741394997,64526762.85797119,6.991770987951895e-06
+106480,446609489920,1.4349623727798462,252694.37735319138,64689760.60241699,6.834835403424222e-06
+106765,447804866560,1.428853497505188,253356.4467072487,64859250.357055664,6.6691814026853535e-06
+107065,449063157760,1.428532705307007,254044.0347957611,65035272.907714844,6.494808530987939e-06
+107405,450489221120,1.4294294834136962,254823.04517149925,65234699.56390381,6.297185791481752e-06
+107715,451789455360,1.4271887636184692,255532.40568637848,65416295.85571289,6.1170003391453065e-06
+107995,452963860480,1.4311064767837525,256178.45214009285,65581683.74786377,5.954252628725953e-06
+108290,454201180160,1.4330047512054442,256872.0287501812,65759239.36004639,5.782785592600703e-06
+108560,455333642240,1.4316231393814087,257536.53442502022,65929352.812805176,5.62585000807303e-06
+108895,456738734080,1.4261334419250489,258315.80425143242,66128845.8883667,5.431134013633709e-06
+109160,457850224640,1.4281217861175537,258946.94263100624,66290417.3135376,5.277104264678201e-06
+109415,458919772160,1.4271953868865968,259539.20039772987,66442035.30181885,5.128887551109074e-06
+109665,459968348160,1.4293783569335938,260125.15892148018,66592040.683898926,4.983576673112111e-06
+109955,461184696320,1.4300671958923339,260828.90233635902,66772198.99810791,4.815016382053727e-06
+110260,462463959040,1.4307714796066284,261527.6113629341,66951068.50891113,4.637737220036797e-06
+110620,463973908480,1.4274176740646363,262364.33426618576,67165269.57214355,4.4284897739999e-06
+110885,465085399040,1.4223868942260742,262973.0377840996,67321097.67272949,4.274460479791742e-06
+111170,466280775680,1.4271921110153198,263636.1656498909,67490858.40637207,4.1088060243055224e-06
+111460,467497123840,1.4233088064193726,264358.1378698349,67675683.29467773,3.9402457332471386e-06
+111685,468440842240,1.4286174154281617,264894.9823355675,67813115.47790527,3.8094660794740776e-06
+111975,469657190400,1.4273700523376465,265550.1725387573,67980844.16992188,3.6409055610420182e-06
+112290,470978396160,1.4266534185409545,266293.6787331104,68171181.75567627,3.457814045759733e-06
+112540,472026972160,1.425380835533142,266881.8741571903,68321759.78424072,3.3125033951364458e-06
+112830,473243320320,1.42493492603302,267565.7266032696,68496826.01043701,3.1439428767043864e-06
+113080,474291896320,1.4286265516281127,268158.7574005127,68648641.89453125,2.9986322260810994e-06
+113370,475508244480,1.4271038579940796,268828.6733150482,68820140.36865234,2.83007170764904e-06
+113595,476451962880,1.4221965885162353,269348.790794611,68953290.44342041,2.699292053875979e-06
+113925,477836083200,1.4260382127761841,270119.6772670746,69150637.3803711,2.507481895008823e-06
+114315,479471861760,1.425297842025757,271028.3651256561,69383261.47216797,2.280797161802184e-06
+114565,480520437760,1.426431736946106,271618.3868908882,69534307.04406738,2.1354865111788968e-06
+114880,481841643520,1.4250082492828369,272354.43952679634,69722736.51885986,1.9523949958966114e-06
+115270,483477422080,1.4197765398025513,273313.7539386749,69968321.00830078,1.7257102626899723e-06
+115575,484756684800,1.4240530490875245,274083.2974374294,70165324.14398193,1.5484311006730422e-06
+115910,486161776640,1.4242891073226929,274924.7234261036,70380729.19708252,1.3537147651732084e-06
+116215,487441039360,1.4271942853927613,275708.40203762054,70581350.92163086,1.176435716843116e-06
+116495,488615444480,1.4240837621688842,276388.6677002907,70755498.93127441,1.0136876653632498e-06
+116760,489726935040,1.4214415645599365,277054.4333803654,70925934.94537354,8.596582574682543e-07
+117000,490733568000,1.4226779556274414,277679.9043869972,71086055.52307129,7.201599601103226e-07
+117260,491824087040,1.4236890649795533,278377.1483194828,71264549.9697876,5.690368425348424e-07
+117645,493438894080,1.4247114610671998,279393.5856318474,71524757.92175293,3.4525828596088104e-07
+117925,494613299200,1.4296443748474121,280118.82418990135,71710418.99261475,1.8251027711357892e-07
+118235,495913533440,1.4239346981048584,280927.33439207077,71917397.60437012,2.324971770661932e-09

metadata/training_logs/ConvergenceCurve_phase4-context-extension-131k.csv

@@ -0,0 +1,192 @@
+training_steps,training_tokens,training_loss,walltime,gputime,learning_rate
+125,524288000,2.554782357215881,984.4803285598755,126013.48205566406,4.273942613508552e-05
+250,1048576000,1.492537350654602,1976.7299485206604,253021.43341064453,6.872427184134722e-05
+375,1572864000,1.4975359773635863,2973.380742073059,380592.73498535156,6.871581717859954e-05
+500,2097152000,1.4889154148101806,3958.0899834632874,506635.5178833008,6.869788194308057e-05
+625,2621440000,1.4991708898544311,4951.81186914444,633831.9192504883,6.867046613479033e-05
+750,3145728000,1.507406897544861,5936.469957828522,759868.1546020508,6.863359158160165e-05
+875,3670016000,1.498574013710022,6925.770752429962,886498.6563110352,6.858726555947214e-05
+1000,4194304000,1.4995141220092774,7912.4334597587585,1012791.4828491211,6.853148806840181e-05
+1125,4718592000,1.497445855140686,8903.68198633194,1139671.2942504883,6.846628821222112e-05
+1250,5242880000,1.5074591159820556,9896.094131469727,1266700.048828125,6.839167326688766e-05
+1375,5767168000,1.5002111291885376,10887.623195648193,1393615.7690429688,6.830767233623192e-05
+1500,6291456000,1.4991462469100951,11878.205075263977,1520410.249633789,6.821430724812672e-05
+1625,6815744000,1.4921324682235717,12867.40041255951,1647027.2528076172,6.81115998304449e-05
+1750,7340032000,1.523764271736145,13853.18541765213,1773207.7334594727,6.799957918701693e-05
+1875,7864320000,1.515320167541504,14843.755986690521,1900000.7662963867,6.787827442167327e-05
+2000,8388608000,1.5183739233016968,15826.274373531342,2025763.1198120117,6.774772919015959e-05
+2125,8912896000,1.5051481771469115,16821.798691749573,2153190.2325439453,6.760796532034874e-05
+2250,9437184000,1.5188077306747436,17806.938366889954,2279288.110961914,6.745903374394402e-05
+2375,9961472000,1.495236349105835,18797.5475025177,2406086.0803222656,6.73009708407335e-05
+2500,10485760000,1.4941252279281616,19785.608899593353,2532557.939147949,6.713382026646286e-05
+2625,11010048000,1.50147403717041,20781.896586418152,2660082.7630615234,6.695761840092018e-05
+2750,11534336000,1.497354221343994,21773.439099788666,2787000.204772949,6.677243072772399e-05
+2875,12058624000,1.492594337463379,22754.230978488922,2912541.565246582,6.657830090261996e-05
+3000,12582912000,1.4926585578918456,23750.513849258423,3040065.772705078,6.63752798573114e-05
+3125,13107200000,1.4984318923950195,24744.183931350708,3167255.5432128906,6.616342579945922e-05
+3250,13631488000,1.514411540031433,25741.15688085556,3294868.0807495117,6.594280421268195e-05
+3375,14155776000,1.5011372566223145,26731.743879318237,3421663.2165527344,6.571345875272527e-05
+3500,14680064000,1.4847777557373047,27724.0012717247,3548672.1627807617,6.547546945512295e-05
+3625,15204352000,1.5044882965087891,28723.126101493835,3676560.140991211,6.522889452753589e-05
+3750,15728640000,1.4940216732025147,29723.706996440887,3804634.4955444336,6.497379945358261e-05
+3875,16252928000,1.4894179916381836,30711.214339733124,3931035.43548584,6.471026426879689e-05
+4000,16777216000,1.49800621509552,31701.409969329834,4057780.4760742188,6.443835445679724e-05
+4125,17301504000,1.484294719696045,32693.367190361023,4184751.000366211,6.415814277715981e-05
+4250,17825792000,1.5007028341293336,33689.12601232529,4312208.129577637,6.386970926541835e-05
+4375,18350080000,1.493506498336792,34682.32307434082,4439337.353515625,6.357313395710662e-05
+4500,18874368000,1.4821550846099854,35666.2157535553,4565275.616455078,6.326850416371599e-05
+4625,19398656000,1.4927472066879273,36650.89477300644,4691314.530944824,6.29558926448226e-05
+4750,19922944000,1.4983328580856323,37638.09194803238,4817675.7693481445,6.263539398787543e-05
+4875,20447232000,1.5008873271942138,38624.85774278641,4943981.79107666,6.230709550436586e-05
+5000,20971520000,1.4903150367736817,39610.86201429367,5070190.33782959,6.197109178174287e-05
+5125,21495808000,1.4989620018005372,40605.754997730255,5197536.639709473,6.162746285554022e-05
+5250,22020096000,1.5058591556549072,41593.60118150711,5323980.95123291,6.12763105891645e-05
+5375,22544384000,1.4910902786254883,42581.0441327095,5450373.648986816,6.091773684602231e-05
+5500,23068672000,1.4767752408981323,43574.69149827957,5577560.511779785,6.055183621356264e-05
+5625,23592960000,1.4914532852172853,44564.69118833542,5704280.472106934,6.0178706917213276e-05
+5750,24117248000,1.5105006313323974,45558.96450281143,5831547.456359863,5.979845445835963e-05
+5875,24641536000,1.4840446662902833,46552.533349990845,5958724.268798828,5.94111843383871e-05
+6000,25165824000,1.4915190172195434,47544.051904678345,6085638.643798828,5.90170020586811e-05
+6125,25690112000,1.4764682626724244,48542.83632040024,6213483.0490112305,5.8616020396584645e-05
+6250,26214400000,1.4947553205490112,49535.04325628281,6340485.536804199,5.820834121550433e-05
+6375,26738688000,1.4819254302978515,50523.48965167999,6467006.675415039,5.779408820671961e-05
+6500,27262976000,1.4844385862350464,51516.32220029831,6594089.241638184,5.737336687161587e-05
+6625,27787264000,1.4932914543151856,52510.58640241623,6721355.059509277,5.694629726349376e-05
+6750,28311552000,1.4815798425674438,53504.04151201248,6848517.313537598,5.65129957976751e-05
+6875,28835840000,1.4881651401519775,54497.4182844162,6975669.540405273,5.607358252746053e-05
+7000,29360128000,1.4796349620819091,55486.41178607941,7102260.708618164,5.5628181144129485e-05
+7125,29884416000,1.4790048694610596,56482.15041399002,7229715.252990723,5.51769117009826e-05
+7250,30408704000,1.4827857828140258,57463.426065444946,7355318.536376953,5.471989788929932e-05
+7375,30932992000,1.4792558193206786,58459.337282180786,7482795.172119141,5.4257267038337886e-05
+7500,31457280000,1.4781176567077636,59457.12294816971,7610511.737365723,5.378914647735655e-05
+7625,31981568000,1.4841596031188964,60450.812039375305,7737703.941040039,5.3315667173592374e-05
+7750,32505856000,1.4755689001083374,61440.799317359924,7864422.31262207,5.2836956456303596e-05
+7875,33030144000,1.4764146280288697,62426.203353405,7990554.02923584,5.2353148930706084e-05
+8000,33554432000,1.4695409059524536,63427.24039077759,8118686.770019531,5.18643792020157e-05
+8125,34078720000,1.4723982000350953,64413.376405239105,8244912.1798706055,5.1370778237469494e-05
+8250,34603008000,1.4811096334457396,65416.51983499527,8373314.5388793945,5.087248428026214e-05
+8375,35127296000,1.4618944311141968,66408.41525316238,8500277.152404785,5.036963921156712e-05
+8500,35651584000,1.4810321426391602,67396.42352819443,8626742.211608887,4.986237399862148e-05
+8625,36175872000,1.4964146041870117,68386.16661310196,8753429.32647705,4.93508305225987e-05
+8750,36700160000,1.4802560997009278,69378.4465432167,8880441.157531738,4.883515430265106e-05
+8875,37224448000,1.4757713842391968,70373.2158613205,9007771.630249023,4.831548358197324e-05
+9000,37748736000,1.4619942569732667,71361.52238368988,9134274.865112305,4.7791963879717514e-05
+9125,38273024000,1.4840023851394653,72348.26117753983,9260577.430725098,4.726473707705736e-05
+9250,38797312000,1.4723041582107543,73342.45064258575,9387833.682250977,4.6733952331123874e-05
+9375,39321600000,1.4644659996032714,74334.40888643265,9514804.337463379,4.6199755161069334e-05
+9500,39845888000,1.460612325668335,75321.32328510284,9641129.380493164,4.566228744806722e-05
+9625,40370176000,1.4779297924041748,76309.20647382736,9767578.428649902,4.512170562520623e-05
+9750,40894464000,1.4698872137069703,77308.77655506134,9895523.399047852,4.457815521163866e-05
+9875,41418752000,1.468229856491089,78304.61529016495,10022990.757141113,4.403178536449559e-05
+10000,41943040000,1.4836175775527953,79304.67171669006,10150997.979736328,4.3482745240908116e-05
+10125,42467328000,1.4738706302642823,80295.66708803177,10277845.387268066,4.293119127396494e-05
+10250,42991616000,1.473243055343628,81286.63889408112,10404689.778442383,4.237727262079716e-05
+10375,43515904000,1.4678482389450074,82276.86048984528,10531438.142700195,4.182114207651466e-05
+10500,44040192000,1.4657092475891114,83264.51493740082,10657857.911987305,4.126295607420616e-05
+10625,44564480000,1.4797403383255006,84252.94907569885,10784377.481689453,4.070286377100274e-05
+10750,45088768000,1.4585044288635254,85236.22227907181,10910236.451721191,4.014102159999311e-05
+10875,45613056000,1.4693016576766968,86232.91317224503,11037812.886047363,3.9577585994265974e-05
+11000,46137344000,1.4555245351791382,87218.0213546753,11163906.733398438,3.901270974893123e-05
+11125,46661632000,1.4605278205871581,88211.04061365128,11291013.198547363,3.844655293505639e-05
+11250,47185920000,1.4623273992538453,89203.6712551117,11418069.920654297,3.787926834775135e-05
+11375,47710208000,1.4501904296875,90203.31879377365,11546024.805603027,3.7311012420104817e-05
+11500,48234496000,1.4555465793609619,91199.40769195557,11673524.184570312,3.67419452231843e-05
+11625,48758784000,1.4745562267303467,92195.03290891647,11800964.212341309,3.6172219552099705e-05
+11750,49283072000,1.4669678735733032,93188.7359547615,11928158.202209473,3.5601995477918535e-05
+11875,49807360000,1.4451501083374023,94178.76350164413,12054881.72821045,3.503142579575069e-05
+12000,50331648000,1.4586689186096191,95164.90592002869,12181107.957763672,3.44606778526213e-05
+12125,50855936000,1.4642741441726685,96151.27073287964,12307362.653808594,3.388989716768265e-05
+12250,51380224000,1.4485573005676269,97140.30460119247,12433958.988952637,3.3319251087959856e-05
+12375,51904512000,1.452154188156128,98129.22229290009,12560540.453491211,3.274888877058402e-05
+12500,52428800000,1.4662384748458863,99116.62769794464,12686928.345336914,3.217897392460145e-05
+12625,52953088000,1.4696976613998414,100105.01482248306,12813441.897277832,3.160966298310086e-05
+12750,53477376000,1.4392393255233764,101090.24580478668,12939551.463012695,3.1041112379170954e-05
+12875,54001664000,1.4627568244934082,102077.42817878723,13065910.806884766,3.0473474907921627e-05
+13000,54525952000,1.4575440549850465,103067.62155056,13192655.55847168,2.9906910640420392e-05
+13125,55050240000,1.4550797605514527,104058.55529546738,13319495.077819824,2.9341577828745358e-05
+13250,55574528000,1.4468258619308472,105043.32339286804,13445545.39428711,2.877762926800642e-05
+13375,56098816000,1.4521300840377807,106031.43281936646,13572023.400878906,2.821522139129229e-05
+13500,56623104000,1.4469180059432984,107026.1884021759,13699352.115478516,2.7654510631691664e-05
+13625,57147392000,1.4474766778945922,108017.4569606781,13826234.490966797,2.7095649784314446e-05
+13750,57671680000,1.4578331899642945,109009.42044258118,13953205.81665039,2.6538793463259935e-05
+13875,58195968000,1.436700382232666,109997.29449272156,14079653.69506836,2.5984098101616837e-05
+14000,58720256000,1.452729549407959,110987.51145124435,14206401.465759277,2.5431712856516242e-05
+14125,59244544000,1.460614848136902,111977.30929374695,14333095.58959961,2.4881790523068048e-05
+14250,59768832000,1.4357179069519044,112969.85160589218,14460141.0055542,2.4334485715371557e-05
+14375,60293120000,1.4537902307510375,113955.03861904144,14586244.943237305,2.378994759055786e-05
+14500,60817408000,1.446098918914795,114945.9777879715,14713085.156860352,2.3248327124747448e-05
+14625,61341696000,1.4509311485290528,115940.73630571365,14840414.247131348,2.2709773475071415e-05
+14750,61865984000,1.4349867057800294,116929.04821395874,14966918.171386719,2.217443397967145e-05
+14875,62390272000,1.441828212738037,117920.14652967453,15093778.75579834,2.1642457795678638e-05
+15000,62914560000,1.4551848554611206,118914.3961763382,15221042.710571289,2.1113990442245267e-05
+15125,63438848000,1.4536888933181762,119912.7855849266,15348836.554870605,2.0589179257513024e-05
+15250,63963136000,1.4343218231201171,120902.67904758453,15475542.91809082,2.0068167941644788e-05
+15375,64487424000,1.434973349571228,121890.9710073471,15602044.28894043,1.9551102013792843e-05
+15500,65011712000,1.4304817056655883,122881.00431919098,15728768.552856445,1.903812153614126e-05
+15625,65536000000,1.4382383155822753,123871.80215597153,15855590.675964355,1.8529370208852924e-05
+15750,66060288000,1.4488927793502808,124860.45211076736,15982137.870178223,1.80249862751225e-05
+15875,66584576000,1.4491661930084228,125852.39286899567,16109106.287231445,1.7525111616123468e-05
+16000,67108864000,1.4428399467468263,126844.71413850784,16236123.409729004,1.7029880837071687e-05
+16125,67633152000,1.4600327444076537,127834.77737665176,16362851.504211426,1.6539434000151232e-05
+16250,68157440000,1.4474923944473266,128827.42879390717,16489910.885620117,1.605390389158856e-05
+16375,68681728000,1.4306964492797851,129818.6548280716,16616787.817993164,1.557342693558894e-05
+16500,69206016000,1.4370262432098388,130809.4248843193,16743606.385192871,1.5098133189894725e-05
+16625,69730304000,1.461853132247925,131801.40537023544,16870579.887390137,1.462815453123767e-05
+16750,70254592000,1.4365899896621703,132795.2848958969,16997796.466674805,1.4163621017360128e-05
+16875,70778880000,1.4278438949584962,133781.98801994324,17124094.466552734,1.3704660887015052e-05
+17000,71303168000,1.452727656364441,134781.81574344635,17252072.415161133,1.3251400559965987e-05
+17125,71827456000,1.4432354736328126,135770.59500694275,17378636.160888672,1.2803964636987075e-05
+17250,72351744000,1.4370049810409546,136763.4261727333,17505718.550109863,1.236247771885246e-05
+17375,72876032000,1.4415734529495239,137749.88814115524,17631985.68206787,1.1927060768357478e-05
+17500,73400320000,1.448450403213501,138734.07413721085,17757961.48956299,1.1497834748297464e-05
+17625,73924608000,1.4521562385559081,139720.14720916748,17884178.842773438,1.1074916983488947e-05
+17750,74448896000,1.4284476900100709,140720.67550897598,18012246.465148926,1.0658423889253754e-05
+17875,74973184000,1.4349009227752685,141708.66920232773,18138709.65789795,1.0248471880913712e-05
+18000,75497472000,1.4354215097427367,142702.52597332,18265923.32458496,9.84517373581184e-06
+18125,76021760000,1.4429131412506104,143692.95391321182,18392698.100891113,9.448639502807055e-06
+18250,76546048000,1.4341587257385253,144678.950984478,18518905.726013184,9.05898014025297e-06
+18375,77070336000,1.4438130140304566,145667.07129478455,18645385.125732422,8.676302059029695e-06
+18500,77594624000,1.4676322078704833,146658.7003850937,18772313.649291992,8.300710760522634e-06
+18625,78118912000,1.452341685295105,147646.1154794693,18898702.78137207,7.932311746117193e-06
+18750,78643200000,1.431972212791443,148639.95228767395,19025913.892822266,7.571204150735866e-06
+18875,79167488000,1.4444401311874389,149629.34332609177,19152555.945739746,7.21748892829055e-06
+19000,79691776000,1.4298672103881835,150616.8995141983,19278963.137817383,6.871263394714333e-06
+19125,80216064000,1.4420791864395142,151607.81851053238,19405800.769348145,6.532623046950903e-06
+19250,80740352000,1.4332106876373292,152591.41368865967,19531700.952148438,6.201661562954541e-06
+19375,81264640000,1.4423624134063722,153581.10126018524,19658380.96130371,5.878470346942777e-06
+19500,81788928000,1.4339903974533081,154569.26450490952,19784865.856628418,5.563138074649032e-06
+19625,82313216000,1.4349782180786133,155560.8947300911,19911794.52545166,5.2557525123120286e-06
+19750,82837504000,1.4391859865188599,156555.20820617676,20039066.650390625,4.956397333444329e-06
+19875,83361792000,1.4334424209594727,157543.97389650345,20165628.65875244,4.665156211558497e-06
+20000,83886080000,1.4429494667053222,158535.4375243187,20292536.003112793,4.382109182188287e-06
+20125,84410368000,1.4282504177093507,159526.97605133057,20419452.934570312,4.107334007130703e-06
+20250,84934656000,1.4241476202011107,160516.53708219528,20546116.746520996,3.840906629193341e-06
+20375,85458944000,1.4211939144134522,161512.73339509964,20673629.874572754,3.582900717447046e-06
+20500,85983232000,1.4452216482162477,162514.04677152634,20801797.98675537,3.3333876672259066e-06
+20625,86507520000,1.4331603813171387,163513.922894001,20929782.13043213,3.092435918006231e-06
+20750,87031808000,1.4439285373687745,164507.00857162476,21056897.09716797,2.8601120902749244e-06
+20875,87556096000,1.4297708702087402,165491.65361166,21182931.66229248,2.6364803034084616e-06
+21000,88080384000,1.4380652761459352,166482.06634283066,21309704.491882324,2.421602175672888e-06
+21125,88604672000,1.440790328979492,167471.32928848267,21436330.14892578,2.2155370515974937e-06
+21250,89128960000,1.4167712259292602,168465.6027150154,21563597.147521973,2.0183420019748155e-06
+21375,89653248000,1.4362161588668823,169454.4678258896,21690171.881713867,1.8300711417396087e-06
+21500,90177536000,1.4421902465820313,170449.9449276924,21817592.95074463,1.6507766531503876e-06
+21625,90701824000,1.4315487194061278,171439.339158535,21944235.41229248,1.4805077626078855e-06
+21750,91226112000,1.4244873237609863,172431.13037109375,22071184.6875,1.3193115364629193e-06
+21875,91750400000,1.4125955533981323,173419.4010901451,22197683.339538574,1.1672325399558758e-06
+22000,92274688000,1.4402506971359252,174406.1421227455,22323986.191711426,1.0243126098430366e-06
+22125,92798976000,1.4289155435562133,175393.06967496872,22450312.918395996,8.905913091439288e-07
+22250,93323264000,1.4243291425704956,176383.74596118927,22577119.483032227,7.661054723939742e-07
+22375,93847552000,1.436662836074829,177379.34653520584,22704556.356506348,6.508893761747458e-07
+22500,94371840000,1.4352865982055665,178370.07630109787,22831369.766540527,5.449749664876435e-07
+22625,94896128000,1.436001124382019,179361.4937210083,22958271.196289062,4.483913187414146e-07
+22750,95420416000,1.440036449432373,180355.07584095,23085449.7076416,3.6116514934292354e-07
+22875,95944704000,1.441613302230835,181344.45638895035,23212090.417785645,2.83320531480058e-07
+23000,96468992000,1.4311069631576538,182340.00187397003,23339520.239868164,2.1487893775429256e-07
+23125,96993280000,1.4302243566513062,183328.21281433105,23466011.240234375,1.5585924018068908e-07
+23250,97517568000,1.434647183418274,184314.68320846558,23592279.450683594,1.0627772439875116e-07
+23375,98041856000,1.423925666809082,185305.42181253433,23719093.992004395,6.614807546156953e-08
+23500,98566144000,1.4347464084625243,186310.39970636368,23847731.16241455,3.5481360072253665e-08
+23625,99090432000,1.4226790523529054,187302.3525762558,23974701.129760742,1.4286050564749075e-08
+23750,99614720000,1.431354284286499,188291.8809390068,24101360.76019287,2.5679884885221327e-09
+23840,99992207360,1.4358789854579501,189009.04103040695,24193157.25189209,1.2136202416287989e-12

model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29fdaae091083d6b6c63c6b52cfa9b404d3fbddb1ac89aefebe96dbac3924fd9
+size 4957588344

model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c633742b9e36a42ddf708c14401c6ad71185df89985b873d64312ac2b1933e51
+size 4916952064

model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1efb32b46e5d31b0433f983a82186d821b731e040854300dd560aa780edab266
+size 4875961752

model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1da5d16e6e186738874618de7e0d47eadbaaeb69dd7b85fe05c8f2dd384c5b44
+size 1400906184

model.safetensors.index.json

@@ -0,0 +1,318 @@
+{
+  "metadata": {
+    "total_size": 16151373824
+  },
+  "weight_map": {
+    "backbone.embeddings.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.0.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.1.mixer.down_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.1.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.1.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.10.mixer.down_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.10.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.10.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.11.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.12.mixer.down_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.12.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.12.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.13.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.14.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.14.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.14.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.15.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.15.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.16.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.16.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.16.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.17.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.18.mixer.k_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.18.mixer.o_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.18.mixer.q_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.18.mixer.v_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.18.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.19.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.19.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.19.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.2.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.2.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.20.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.20.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.21.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.21.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.21.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.22.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.23.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.23.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.23.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.24.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.25.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.25.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.25.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.26.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.27.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.27.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.27.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.28.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.29.mixer.k_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.29.mixer.o_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.29.mixer.q_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.29.mixer.v_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.29.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.3.mixer.down_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.3.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.3.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.30.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.30.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.30.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.in_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.mixer.out_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.31.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.32.mixer.down_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.32.mixer.up_proj.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.32.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.33.mixer.A_log": "model-00002-of-00004.safetensors",
+    "backbone.layers.33.mixer.D": "model-00002-of-00004.safetensors",
+    "backbone.layers.33.mixer.conv1d.bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.33.mixer.conv1d.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.33.mixer.dt_bias": "model-00002-of-00004.safetensors",
+    "backbone.layers.33.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.33.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.33.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.33.norm.weight": "model-00002-of-00004.safetensors",
+    "backbone.layers.34.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.34.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.34.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.35.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.36.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.36.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.36.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.37.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.38.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.38.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.38.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.39.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.4.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.4.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.40.mixer.k_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.40.mixer.o_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.40.mixer.q_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.40.mixer.v_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.40.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.41.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.41.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.41.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.42.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.43.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.43.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.43.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.44.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.45.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.45.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.45.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.46.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.47.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.47.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.47.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.48.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.49.mixer.down_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.49.mixer.up_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.49.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.5.mixer.down_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.5.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.5.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.50.mixer.A_log": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.D": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.conv1d.bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.conv1d.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.dt_bias": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.in_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.mixer.out_proj.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.50.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.51.mixer.down_proj.weight": "model-00004-of-00004.safetensors",
+    "backbone.layers.51.mixer.up_proj.weight": "model-00004-of-00004.safetensors",
+    "backbone.layers.51.norm.weight": "model-00003-of-00004.safetensors",
+    "backbone.layers.6.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.6.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.7.mixer.k_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.7.mixer.o_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.7.mixer.q_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.7.mixer.v_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.7.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.8.mixer.down_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.8.mixer.up_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.8.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.A_log": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.D": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.conv1d.bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.conv1d.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.dt_bias": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.in_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.mixer.out_proj.weight": "model-00001-of-00004.safetensors",
+    "backbone.layers.9.norm.weight": "model-00001-of-00004.safetensors",
+    "backbone.norm_f.weight": "model-00004-of-00004.safetensors",
+    "lm_head.weight": "model-00004-of-00004.safetensors"
+  }
+}

modeling_nemotron_h.py

@@ -0,0 +1,1887 @@
+# coding=utf-8
+# Copyright 2024 HuggingFace Inc. team.
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""PyTorch NemotronH model."""
+
+import math
+from dataclasses import dataclass
+from typing import Any, Dict, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import DynamicCache  # we need __iter__ and __len__ of pkv
+from transformers.generation import GenerationMixin
+from transformers.modeling_attn_mask_utils import (
+    AttentionMaskConverter,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    ModelOutput,
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+)
+from transformers.utils.import_utils import (
+    is_causal_conv1d_available,
+    is_flash_attn_2_available,
+    is_flash_attn_greater_or_equal_2_10,
+    is_mamba_2_ssm_available,
+)
+from .configuration_nemotron_h import NemotronHConfig
+
+logger = logging.get_logger(__name__)
+
+
+# Copied from transformers.models.mamba.modeling_mamba2.modeling_mamba2.py with MAMBA2->NEMOTRONH,Mamba2->NemotronH
+# For Mamba2 components Mamba2->NemotronHMamba2
+if is_mamba_2_ssm_available():
+    from mamba_ssm.ops.triton.selective_state_update import selective_state_update
+    from mamba_ssm.ops.triton.ssd_combined import (
+        mamba_chunk_scan_combined,
+        mamba_split_conv1d_scan_combined,
+    )
+else:
+    (
+        mamba_chunk_scan_combined,
+        mamba_split_conv1d_scan_combined,
+        selective_state_update,
+    ) = None, None, None
+
+try:
+    # from mamba_ssm.ops.triton.layernorm_gated import RMSNorm as RMSNormGated
+    from mamba_ssm.ops.triton.layernorm_gated import rmsnorm_fn
+except ImportError:
+    raise ImportError("mamba-ssm is required by the Mamba model but cannot be imported")
+
+if is_causal_conv1d_available():
+    from causal_conv1d import causal_conv1d_fn, causal_conv1d_update
+else:
+    causal_conv1d_update, causal_conv1d_fn = None, None
+
+if is_flash_attn_2_available():
+    from transformers.modeling_flash_attention_utils import _flash_attention_forward
+
+is_fast_path_available = all(
+    (
+        selective_state_update,
+        mamba_chunk_scan_combined,
+        mamba_split_conv1d_scan_combined,
+        causal_conv1d_fn,
+        causal_conv1d_update,
+    )
+)
+
+
+_CHECKPOINT_FOR_DOC = "nvidia/Nemotron-H-56B-Base-8K"
+_CONFIG_FOR_DOC = "NemotronHConfig"
+
+
+# Helper methods for segment sum computation
+
+
+def pad_tensor_by_size(input_tensor: torch.Tensor, pad_size: int):
+    """
+    Padding x tensor with `pad_size` on the seq_len dim (dim=1)
+
+    Assumes that we only have tensors of either size 4 or 3
+    """
+    pad_shape = (
+        (0, 0, 0, 0, 0, pad_size, 0, 0)
+        if len(input_tensor.shape) == 4
+        else (0, 0, 0, pad_size, 0, 0)
+    )
+
+    return torch.nn.functional.pad(input_tensor, pad_shape, mode="constant", value=0)
+
+
+def reshape_into_chunks(input_tensor, pad_size, chunk_size):
+    """
+    Padding input_tensor with `pad_size` on the seq_len dim (dim=1) and
+    simultaneously splitting it into chunk sequences.
+
+    Assumes that we only have tensors of either size 4 or 3
+    """
+    # [bsz, seq_len, ...] -> [bsz, seq_len multiple of chunk_size, ...]
+    input_tensor = pad_tensor_by_size(input_tensor, pad_size)
+
+    if len(input_tensor.shape) == 3:
+        # [bsz, seq_len multiple of chunk_size, num_heads] -> [bsz, -1, chunk_size, num_heads]
+        return input_tensor.reshape(
+            input_tensor.shape[0], -1, chunk_size, input_tensor.shape[2]
+        )
+    else:
+        # [bsz, seq_len multiple of chunk_size, num_heads, head_dim or state_size] -> [bsz, -1, chunk_size, num_heads, head_dim or state_size]
+        return input_tensor.reshape(
+            input_tensor.shape[0],
+            -1,
+            chunk_size,
+            input_tensor.shape[2],
+            input_tensor.shape[3],
+        )
+
+
+def segment_sum(input_tensor):
+    """
+    More stable segment sum calculation. Uses cumulative sums and masking instead of direct subtractions.
+    """
+    chunk_size = input_tensor.size(-1)
+    # 1. expand input tensor to have an additional dimension and repeat along that dimension
+    # [..., chunk_size] -> [..., chunk_size, chunk_size]
+    input_tensor = input_tensor[..., None].expand(*input_tensor.size(), chunk_size)
+    # 2. create a lower triangular mask with the diagonal set to 0 to 0 out elements above diag
+    mask = torch.tril(
+        torch.ones(
+            chunk_size, chunk_size, device=input_tensor.device, dtype=torch.bool
+        ),
+        diagonal=-1,
+    )
+    input_tensor = input_tensor.masked_fill(~mask, 0)
+    # 3. compute actual cumsum
+    tensor_segsum = torch.cumsum(input_tensor, dim=-2)
+
+    # 4. apply mask to keep only the lower triangular part of the cumulative sum result (incl diagonal this time)
+    mask = torch.tril(
+        torch.ones(
+            chunk_size, chunk_size, device=input_tensor.device, dtype=torch.bool
+        ),
+        diagonal=0,
+    )
+    tensor_segsum = tensor_segsum.masked_fill(~mask, -torch.inf)
+    return tensor_segsum
+
+
+def apply_mask_to_padding_states(hidden_states, attention_mask):
+    """
+    Tunes out the hidden states for padding tokens, see https://github.com/state-spaces/mamba/issues/66
+    """
+    if (
+        attention_mask is not None
+        and attention_mask.shape[1] > 1
+        and attention_mask.shape[0] > 1
+    ):
+        dtype = hidden_states.dtype
+        hidden_states = (hidden_states * attention_mask[:, :, None]).to(dtype)
+
+    return hidden_states
+
+
+# Copied from https://github.com/huggingface/transformers/blob/main/src/transformers/models/jamba/modeling_jamba.py
+class HybridMambaAttentionDynamicCache(DynamicCache):
+    """
+    A dynamic cache that can handle both the attention cache (which has a seq_len dimension) and the mamba cache
+    (which has a constant shape regardless of seq_len).
+
+    This cache has two sets of lists of tensors: `key_cache` and `value_cache` for attention cache and `conv_states`
+    and `ssm_states` for mamba cache. Each of these lists has `num_layers` tensors. The expected shape for each tensor
+    For attention layers, `key_cache` and `value_cache` have a shape of `(batch_size, num_heads, seq_len, head_dim)`,
+    while `conv_states` and `ssm_states` have a shape of `(batch_size, 0)` (empty tensors).
+    For mamba layers, `key_cache` and `value_cache` have a shape of `(batch_size, 0)` (empty tensors),
+    while `conv_states` represents the convolution state and has a shape of `(batch_size, d_inner, d_conv)`,
+    and `ssm_states` represents the ssm state and has a shape of `(batch_size, d_inner, d_state)`.
+    """
+
+    def __init__(self, config, batch_size, dtype=torch.float16, device=None):
+        super().__init__()
+        self.dtype = dtype
+        self.hybrid_override_pattern = config.hybrid_override_pattern
+        self.has_previous_state = False  # only used by mamba
+        intermediate_size = config.expand * config.hidden_size
+        ssm_state_size = config.ssm_state_size
+        conv_kernel_size = config.conv_kernel
+        self.conv_states = []
+        self.ssm_states = []
+        self.transformer_layers = []
+        for i in range(config.num_hidden_layers):
+            if self.hybrid_override_pattern[i] == "M":
+                # Mamba layer
+                self.conv_states += [
+                    torch.zeros(
+                        batch_size,
+                        intermediate_size,
+                        conv_kernel_size,
+                        device=device,
+                        dtype=dtype,
+                    )
+                ]
+                self.ssm_states += [
+                    torch.zeros(
+                        batch_size,
+                        intermediate_size,
+                        ssm_state_size,
+                        device=device,
+                        dtype=dtype,
+                    )
+                ]
+            else:
+                # Attention or MLP layer
+                self.conv_states += [torch.tensor([[]] * batch_size, device=device)]
+                self.ssm_states += [torch.tensor([[]] * batch_size, device=device)]
+                self.transformer_layers.append(i)
+
+        self.key_cache = [
+            torch.tensor([[]] * batch_size, device=device)
+            for _ in range(config.num_hidden_layers)
+        ]
+        self.value_cache = [
+            torch.tensor([[]] * batch_size, device=device)
+            for _ in range(config.num_hidden_layers)
+        ]
+
+    def update(
+        self,
+        key_states: torch.Tensor,
+        value_states: torch.Tensor,
+        layer_idx: int,
+        cache_kwargs: Optional[Dict[str, Any]] = None,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        # Update the cache
+        if self.key_cache[layer_idx].shape[-1] == 0:
+            self.key_cache[layer_idx] = key_states
+            self.value_cache[layer_idx] = value_states
+        else:
+            self.key_cache[layer_idx] = torch.cat(
+                [self.key_cache[layer_idx], key_states], dim=2
+            )
+            self.value_cache[layer_idx] = torch.cat(
+                [self.value_cache[layer_idx], value_states], dim=2
+            )
+
+        return self.key_cache[layer_idx], self.value_cache[layer_idx]
+
+    def reorder_cache(self, beam_idx: torch.LongTensor):
+        """Reorders the cache for beam search, given the selected beam indices."""
+        for layer_idx in range(len(self.key_cache)):
+            device = self.key_cache[layer_idx].device
+            self.key_cache[layer_idx] = self.key_cache[layer_idx].index_select(
+                0, beam_idx.to(device)
+            )
+            device = self.value_cache[layer_idx].device
+            self.value_cache[layer_idx] = self.value_cache[layer_idx].index_select(
+                0, beam_idx.to(device)
+            )
+
+            device = self.conv_states[layer_idx].device
+            self.conv_states[layer_idx] = self.conv_states[layer_idx].index_select(
+                0, beam_idx.to(device)
+            )
+            device = self.ssm_states[layer_idx].device
+            self.ssm_states[layer_idx] = self.ssm_states[layer_idx].index_select(
+                0, beam_idx.to(device)
+            )
+
+    def get_seq_length(self, layer_idx: Optional[int] = 0) -> int:
+        """Returns the sequence length of the cached states. A layer index can be optionally passed."""
+        # take any layer that contains cache and not empty tensor
+        layer_idx = (
+            self.transformer_layers[0]
+            if layer_idx not in self.transformer_layers
+            else layer_idx
+        )
+        if len(self.key_cache) <= layer_idx:
+            return 0
+        return self.key_cache[layer_idx].shape[-2]
+
+    def to_legacy_cache(self) -> Tuple[Tuple[torch.Tensor], Tuple[torch.Tensor]]:
+        raise NotImplementedError(
+            "HybridMambaAttentionDynamicCache does not have a legacy cache equivalent."
+        )
+
+    @classmethod
+    def from_legacy_cache(
+        cls, past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None
+    ) -> "DynamicCache":
+        raise NotImplementedError(
+            "HybridMambaAttentionDynamicCache does not have a legacy cache equivalent."
+        )
+
+    # Copied from modeling_mamba2.py
+    def update_conv_state(
+        self, layer_idx: int, new_conv_state: torch.Tensor, cache_init: bool = False
+    ) -> torch.Tensor:
+        if cache_init:
+            self.conv_states[layer_idx] = new_conv_state.to(self.conv_states.device)
+        else:
+            self.conv_states[layer_idx] = self.conv_states[layer_idx].roll(
+                shifts=-1, dims=-1
+            )
+            self.conv_states[layer_idx][:, :, -1] = new_conv_state[:, 0, :].to(
+                self.conv_states.device
+            )
+        return self.conv_states[layer_idx]
+
+    def update_ssm_state(self, layer_idx: int, new_ssm_state: torch.Tensor):
+        self.ssm_states[layer_idx] = new_ssm_state.to(self.ssm_states.device)
+        return self.ssm_states[layer_idx]
+
+    def reset(self):
+        self.conv_states.zero_()
+        self.ssm_states.zero_()
+
+
+class MambaRMSNormGated(torch.nn.Module):
+    def __init__(self, hidden_size, group_size, eps=1e-5):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+        self.group_size = group_size
+
+    # jan28b version
+    def forward(self, hidden_states, gate=None):
+        return rmsnorm_fn(
+            x=hidden_states,
+            weight=self.weight,
+            bias=None,  # No bias
+            z=gate,
+            eps=self.variance_epsilon,
+            group_size=self.group_size,
+            norm_before_gate=False,
+        )
+
+
+class NemotronHMamba2Mixer(nn.Module):
+    """
+    Compute ∆, A, B, C, and D the state space parameters and compute the `contextualized_states`.
+    A, D are input independent (see Mamba paper [1] Section 3.5.2 "Interpretation of A" for why A isn't selective)
+    ∆, B, C are input-dependent (this is a key difference between Mamba and the linear time invariant S4,
+    and is why Mamba is called **selective** state spaces)
+    """
+
+    def __init__(self, config: NemotronHConfig, layer_idx: int):
+        super().__init__()
+        self.num_heads = config.mamba_num_heads
+        self.hidden_size = config.hidden_size
+        self.ssm_state_size = config.ssm_state_size
+        self.conv_kernel_size = config.conv_kernel
+        self.intermediate_size = config.mamba_num_heads * config.mamba_head_dim
+        self.layer_idx = layer_idx
+        self.use_conv_bias = config.use_conv_bias
+        self.activation = config.mamba_hidden_act
+        self.act = ACT2FN[config.mamba_hidden_act]
+
+        self.layer_norm_epsilon = config.layer_norm_epsilon
+
+        self.n_groups = config.n_groups
+        self.head_dim = config.mamba_head_dim
+        self.chunk_size = config.chunk_size
+
+        self.time_step_limit = config.time_step_limit
+        self.time_step_min = config.time_step_min
+        self.time_step_max = config.time_step_max
+
+        self.conv_dim = self.intermediate_size + 2 * self.n_groups * self.ssm_state_size
+        self.conv1d = nn.Conv1d(
+            in_channels=self.conv_dim,
+            out_channels=self.conv_dim,
+            bias=config.use_conv_bias,
+            kernel_size=config.conv_kernel,
+            groups=self.conv_dim,
+            padding=config.conv_kernel - 1,
+        )
+
+        # projection of the input hidden states
+        projection_size = self.intermediate_size + self.conv_dim + self.num_heads
+        self.in_proj = nn.Linear(
+            self.hidden_size,
+            projection_size,
+            bias=config.use_bias,
+        )
+        # selective projection used to make dt, B and C input dependant
+
+        # time step projection (discretization)
+        # instantiate once and copy inv_dt in init_weights of PretrainedModel
+        self.dt_bias = nn.Parameter(torch.ones(self.num_heads))
+
+        # S4D real initialization. These are not discretized!
+        # The core is to load them, compute the discrete states, then write the updated state. Keeps the memory bounded
+        A = torch.arange(1, self.num_heads + 1)
+        self.A_log = nn.Parameter(torch.log(A))
+        self.A_log._no_weight_decay = True
+        self.norm = MambaRMSNormGated(
+            self.intermediate_size,
+            eps=self.layer_norm_epsilon,
+            group_size=self.intermediate_size // self.n_groups,
+        )
+        self.D = nn.Parameter(torch.ones(self.num_heads))
+        self.D._no_weight_decay = True
+
+        self.out_proj = nn.Linear(
+            self.intermediate_size, self.hidden_size, bias=config.use_bias
+        )
+        self.use_bias = config.use_bias
+
+        if not is_fast_path_available:
+            logger.warning_once(
+                "The fast path is not available because on of `(selective_state_update, causal_conv1d_fn, causal_conv1d_update)`"
+                " is None. Falling back to the naive implementation. To install follow https://github.com/state-spaces/mamba/#installation and"
+                " https://github.com/Dao-AILab/causal-conv1d"
+            )
+
+    def cuda_kernels_forward(
+        self,
+        hidden_states: torch.Tensor,
+        cache_params: Optional[HybridMambaAttentionDynamicCache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+    ):
+        # 1. Gated MLP's linear projection
+        hidden_states = apply_mask_to_padding_states(hidden_states, attention_mask)
+        projected_states = self.in_proj(hidden_states)
+
+        # Set up dimensions for reshapes later
+        batch_size, seq_len, _ = hidden_states.shape
+        groups_time_state_size = self.n_groups * self.ssm_state_size
+        d_mlp = (
+            projected_states.shape[-1]
+            - 2 * self.intermediate_size
+            - 2 * self.n_groups * self.ssm_state_size
+            - self.num_heads
+        ) // 2
+
+        # Single step calculations via cache
+        if (
+            cache_params is not None
+            and cache_position is not None
+            and cache_position[0] > 0
+        ):
+            _, _, gate, hidden_states_B_C, dt = projected_states.squeeze(1).split(
+                [d_mlp, d_mlp, self.intermediate_size, self.conv_dim, self.num_heads],
+                dim=-1,
+            )
+
+            # 2. Convolution sequence transformation
+            hidden_states_B_C = causal_conv1d_update(
+                hidden_states_B_C,
+                cache_params.conv_states[self.layer_idx],
+                self.conv1d.weight.squeeze(1),
+                self.conv1d.bias,
+                self.activation,
+            )
+
+            hidden_states, B, C = torch.split(
+                hidden_states_B_C,
+                [
+                    self.intermediate_size,
+                    groups_time_state_size,
+                    groups_time_state_size,
+                ],
+                dim=-1,
+            )
+
+            # 3. SSM transformation
+            A = -torch.exp(self.A_log.float())  # (nheads,)
+            A = (
+                A[:, None, ...][:, :, None]
+                .expand(-1, self.head_dim, self.ssm_state_size)
+                .to(dtype=torch.float32)
+            )
+            dt = dt[:, :, None].expand(-1, -1, self.head_dim)
+            dt_bias = self.dt_bias[:, None, ...].expand(-1, self.head_dim)
+            D = self.D[:, None, ...].expand(-1, self.head_dim)
+            B = B.view(batch_size, self.n_groups, B.shape[1] // self.n_groups)
+            C = C.view(batch_size, self.n_groups, C.shape[1] // self.n_groups)
+            hidden_states_reshaped = hidden_states.view(
+                batch_size, self.num_heads, self.head_dim
+            )
+            hidden_states = selective_state_update(
+                cache_params.ssm_states[self.layer_idx],
+                hidden_states_reshaped,
+                dt,
+                A,
+                B,
+                C,
+                D,
+                z=None,
+                dt_bias=dt_bias,
+                dt_softplus=True,
+            )
+            hidden_states = hidden_states.view(
+                batch_size, self.num_heads * self.head_dim
+            )
+            hidden_states = self.norm(hidden_states, gate)
+
+            # 4. Final linear projection
+            out = self.out_proj(hidden_states)[:, None, ...]
+
+        # Fused calculations or step by step if no initialized cache is found
+        else:
+            A = -torch.exp(
+                self.A_log.float()
+            )  # (num_heads) or (intermediate_size, state_size)
+            dt_limit_kwargs = (
+                {}
+                if self.time_step_limit == (0.0, float("inf"))
+                else {"dt_limit": self.time_step_limit}
+            )
+
+            # 2-4. Fused kernel for conv1d, SSM, and the final projection
+            if self.training and cache_params is None:
+                out = mamba_split_conv1d_scan_combined(
+                    projected_states,
+                    self.conv1d.weight.squeeze(1),
+                    self.conv1d.bias,
+                    self.dt_bias,
+                    A,
+                    D=self.D,
+                    chunk_size=self.chunk_size,
+                    seq_idx=None,  # was seq_idx
+                    activation=self.activation,
+                    rmsnorm_weight=self.norm.weight,
+                    rmsnorm_eps=self.norm.variance_epsilon,
+                    outproj_weight=self.out_proj.weight,
+                    outproj_bias=self.out_proj.bias,
+                    headdim=self.head_dim,
+                    ngroups=self.n_groups,
+                    norm_before_gate=False,
+                    return_final_states=False,
+                    **dt_limit_kwargs,
+                )
+
+            else:
+                _, _, gate, hidden_states_B_C, dt = projected_states.split(
+                    [
+                        d_mlp,
+                        d_mlp,
+                        self.intermediate_size,
+                        self.conv_dim,
+                        self.num_heads,
+                    ],
+                    dim=-1,
+                )
+
+                # 2. Convolution sequence transformation
+                # Init cache
+                if cache_params is not None:
+                    hidden_states_B_C_transposed = hidden_states_B_C.transpose(1, 2)
+                    conv_states = nn.functional.pad(
+                        hidden_states_B_C_transposed,
+                        (
+                            cache_params.conv_kernel_size
+                            - hidden_states_B_C_transposed.shape[-1],
+                            0,
+                        ),
+                    )
+                    cache_params.update_conv_state(
+                        layer_idx=self.layer_idx,
+                        new_conv_state=conv_states,
+                        cache_init=True,
+                    )
+
+                if self.activation not in ["silu", "swish"]:
+                    hidden_states_B_C = self.act(
+                        self.conv1d(hidden_states_B_C.transpose(1, 2))[
+                            ..., :seq_len
+                        ].transpose(1, 2)
+                    )
+                else:
+                    hidden_states_B_C = causal_conv1d_fn(
+                        x=hidden_states_B_C.transpose(1, 2),
+                        weight=self.conv1d.weight.squeeze(1),
+                        bias=self.conv1d.bias,
+                        activation=self.activation,
+                    ).transpose(1, 2)
+                hidden_states_B_C = apply_mask_to_padding_states(
+                    hidden_states_B_C, attention_mask
+                )
+                hidden_states, B, C = torch.split(
+                    hidden_states_B_C,
+                    [
+                        self.intermediate_size,
+                        groups_time_state_size,
+                        groups_time_state_size,
+                    ],
+                    dim=-1,
+                )
+
+                # 3. SSM transformation
+                scan_output, ssm_state = mamba_chunk_scan_combined(
+                    hidden_states.view(batch_size, seq_len, -1, self.head_dim),
+                    dt,
+                    A,
+                    B.view(batch_size, seq_len, self.n_groups, -1),
+                    C.view(batch_size, seq_len, self.n_groups, -1),
+                    chunk_size=self.chunk_size,
+                    D=self.D,
+                    z=None,
+                    seq_idx=None,
+                    return_final_states=True,
+                    dt_bias=self.dt_bias,
+                    dt_softplus=True,
+                    **dt_limit_kwargs,
+                )
+
+                # Init cache
+                if ssm_state is not None and cache_params is not None:
+                    cache_params.update_ssm_state(
+                        layer_idx=self.layer_idx, new_ssm_state=ssm_state
+                    )
+
+                scan_output = scan_output.view(batch_size, seq_len, -1)
+
+                # Multiply "gate" branch and apply extra normalization layer
+                scan_output = self.norm(scan_output, gate)
+
+                # 4. Final linear projection
+                out = self.out_proj(scan_output)
+        return out
+
+    # fmt: off
+    def torch_forward(self, input_states, cache_params: Optional[HybridMambaAttentionDynamicCache]=None, cache_position:Optional[torch.LongTensor]=None, attention_mask: Optional[torch.Tensor]=None):
+        batch_size, seq_len, _ = input_states.shape
+        dtype = input_states.dtype
+
+        # 1. Gated MLP's linear projection
+        input_states = apply_mask_to_padding_states(input_states, attention_mask)
+        projected_states = self.in_proj(input_states)
+        d_mlp = (projected_states.shape[-1] - 2 * self.intermediate_size - 2 * self.n_groups * self.ssm_state_size-self.num_heads) // 2
+        _, _, gate, hidden_states_B_C, dt = projected_states.split(
+                [d_mlp, d_mlp, self.intermediate_size,  self.conv_dim, self.num_heads], dim=-1
+        )
+
+        # 2. Convolution sequence transformation
+        if cache_params is not None and cache_position is not None and cache_position[0] > 0:
+            cache_params.update_conv_state(layer_idx=self.layer_idx, new_conv_state=hidden_states_B_C, cache_init=False)
+
+            # We need to guarantee that anything regarding the cache is on the same device
+            conv_states = cache_params.conv_states[self.layer_idx].to(device=self.conv1d.weight.device)
+
+            hidden_states_B_C = torch.sum(
+                conv_states * self.conv1d.weight.squeeze(1), dim=-1
+            )
+            if self.use_conv_bias:
+                hidden_states_B_C = hidden_states_B_C + self.conv1d.bias
+            hidden_states_B_C = self.act(hidden_states_B_C)
+        else:
+            # Init cache
+            if cache_params is not None:
+                hidden_states_B_C_transposed = hidden_states_B_C.transpose(1, 2)
+                conv_states = nn.functional.pad(
+                    hidden_states_B_C_transposed, (cache_params.conv_kernel_size - hidden_states_B_C_transposed.shape[-1], 0)
+                )
+                cache_params.update_conv_state(layer_idx=self.layer_idx, new_conv_state=conv_states, cache_init=True)
+
+            hidden_states_B_C = self.act(self.conv1d(hidden_states_B_C.transpose(1, 2))[..., :seq_len].transpose(1, 2))
+
+        hidden_states_B_C = apply_mask_to_padding_states(hidden_states_B_C, attention_mask)
+        hidden_states, B, C = torch.split(
+            hidden_states_B_C,
+            [self.intermediate_size, self.n_groups * self.ssm_state_size, self.n_groups * self.ssm_state_size],
+            dim=-1
+        )
+
+        # 3. SSM transformation
+        A = -torch.exp(self.A_log.float())                            # [num_heads]
+        if cache_params is not None and cache_position is not None and cache_position[0] > 0:
+            # We need to guarantee that anything regarding the cache is on the same device
+            cache_device = cache_params.ssm_states.device
+
+            # Note: there is no need to pad parameter matrices here, as there is just one new token
+            # for batched generation
+            dt = dt[:, 0, :][:, None, ...]
+            dt = dt.transpose(1, 2).expand(batch_size, dt.shape[-1], self.head_dim)
+            # [num_heads] -> [num_heads, head_dim]
+            dt_bias = self.dt_bias[..., None].expand(self.dt_bias.shape[0], self.head_dim)
+
+            dt = torch.nn.functional.softplus(dt + dt_bias.to(dt.dtype))
+            dt = torch.clamp(dt, self.time_step_limit[0], self.time_step_limit[1])
+            A = A[..., None, None].expand(self.num_heads, self.head_dim, self.ssm_state_size).to(dtype=torch.float32)
+            # [bsz, num_heads, head_dim, state_size]
+            dA = (torch.exp(dt[..., None] * A)).to(device=cache_device)
+
+            # Discretize B
+            # [bsz, n_groups * state_size] -> [bsz, n_groups, 1, state_size] ->
+            # -> [bsz, n_groups, group to head repetition factor, state_size] -> [bsz, num_heads, state_size]
+            B = B.reshape(batch_size, self.n_groups, -1)[..., None, :]
+            B = B.expand(batch_size, self.n_groups, self.num_heads // self.n_groups, B.shape[-1]).contiguous()
+            B = B.reshape(batch_size, -1, B.shape[-1])
+            # [bsz, num_heads, head_dim, state_size]
+            dB = dt[..., None] * B[..., None, :]
+
+            # Discretize x into dB
+            # [bsz, intermediate_size] -> [bsz, num_heads, head_dim]
+            hidden_states = hidden_states.reshape(batch_size, -1, self.head_dim)
+            dBx = (dB * hidden_states[..., None]).to(device=cache_device)
+
+            # State calculation
+            cache_params.update_ssm_state(
+                layer_idx=self.layer_idx,
+                new_ssm_state=cache_params.ssm_states[self.layer_idx] * dA + dBx
+            )
+
+            # Subsequent output
+            # [bsz, n_groups * state_size] -> [bsz, num_heads, state_size]
+            C = C.reshape(batch_size, self.n_groups, -1)[..., None, :]
+            C = C.expand(batch_size, self.n_groups, self.num_heads // self.n_groups, C.shape[-1]).contiguous()
+            C = C.reshape(batch_size, -1, C.shape[-1])
+            # [bsz, num_heads, head_dim]
+
+            ssm_states = cache_params.ssm_states[self.layer_idx].to(device=C.device, dtype=C.dtype)  # Shape: [b, h, d, n]
+            # Reshape ssm_states to merge the first two dimensions
+            ssm_states_reshaped = ssm_states.view(batch_size * self.num_heads, self.head_dim, self.ssm_state_size)  # Shape: [b*h, d, n]
+            C_reshaped = C.view(batch_size * self.num_heads, self.ssm_state_size, 1)  # Shape: [b*h, n, 1]
+            y = torch.bmm(ssm_states_reshaped, C_reshaped)
+            y = y.view(batch_size, self.num_heads, self.head_dim)
+
+            # D skip connection
+            # [num_heads] -> [num_heads, head_dim]
+            D = self.D[..., None].expand(self.D.shape[0], self.head_dim)
+            y = (y + hidden_states * D).to(y.dtype)
+
+            # [bsz, num_heads, head_dim] -> [bsz, 1, intermediate_size]
+            y = y.reshape(batch_size, -1)[:, None, ...]
+        else:
+            # begin ssd naive implementation without einsums
+            dt = nn.functional.softplus(dt + self.dt_bias)
+            dt = torch.clamp(dt, self.time_step_limit[0], self.time_step_limit[1])
+            hidden_states = hidden_states.reshape(batch_size, seq_len, -1, self.head_dim).float()
+            B = B.reshape(batch_size, seq_len, -1, self.ssm_state_size).float()
+            C = C.reshape(batch_size, seq_len, -1, self.ssm_state_size).float()
+            B = B.repeat(1, 1, self.num_heads // self.n_groups, 1)
+            C = C.repeat(1, 1, self.num_heads // self.n_groups, 1)
+            pad_size = (self.chunk_size - seq_len % self.chunk_size) % self.chunk_size
+
+            D_residual = self.D[..., None] * pad_tensor_by_size(hidden_states, pad_size)
+
+            # Discretize x and A
+            hidden_states = hidden_states * dt[..., None]
+            A = A.to(hidden_states.dtype) * dt
+
+            # Rearrange into blocks/chunks
+            hidden_states, A, B, C = [reshape_into_chunks(t, pad_size, self.chunk_size) for t in (hidden_states, A, B, C)]
+
+            # [bsz, -1, chunk_size, num_heads] -> [bsz, num_heads, -1, chunk_size]
+            A = A.permute(0, 3, 1, 2)
+            A_cumsum = torch.cumsum(A, dim=-1)
+
+            # 1. Compute the output for each intra-chunk (diagonal blocks)
+            # This is the analog of a causal mask
+            L = torch.exp(segment_sum(A))
+
+            # Contraction of C and B to get G (attention-weights like)
+            G_intermediate = C[:, :, :, None, :, :] * B[:, :, None, :, :, :]  # shape: (b, c, l, s, h, n)
+            G = G_intermediate.sum(dim=-1)  # shape: (b, c, l, s, h)
+
+            # Compute M, equivalent to applying attention mask to weights
+            M_intermediate = G[..., None] * L.permute(0, 2, 3, 4, 1)[..., None]
+            M = M_intermediate.sum(dim=-1)
+
+            # Compute Y_diag (apply to values)
+            Y_diag = (M[..., None] * hidden_states[:, :, None]).sum(dim=3)
+
+            # 2. Compute the state for each intra-chunk
+            # (right term of low-rank factorization of off-diagonal blocks; B terms)
+            decay_states = torch.exp((A_cumsum[:, :, :, -1:] - A_cumsum))
+            B_decay = B * decay_states.permute(0, -2, -1, 1)[..., None]
+            states = (B_decay[..., None, :] * hidden_states[..., None]).sum(dim=2)
+
+            # 3. Compute the inter-chunk SSM recurrence; produces correct SSM states at chunk boundaries
+            # (middle term of factorization of off-diag blocks; A terms)
+            if cache_params is not None and cache_position is not None and cache_position[0] > 0:
+                previous_states = cache_params.ssm_states[self.layer_idx][:, None, ...].to(device=states.device)
+            else:
+                previous_states = torch.zeros_like(states[:, :1])
+            states = torch.cat([previous_states, states], dim=1)
+            decay_chunk = torch.exp(segment_sum(nn.functional.pad(A_cumsum[:, :, :, -1], (1, 0))))
+            decay_chunk = decay_chunk.transpose(1, 3)
+            new_states = (decay_chunk[..., None, None] * states[:, :, None, ...]).sum(dim=1)
+            states, ssm_state = new_states[:, :-1], new_states[:, -1]
+
+            # 4. Compute state -> output conversion per chunk
+            # (left term of low-rank factorization of off-diagonal blocks; C terms)
+            state_decay_out = torch.exp(A_cumsum)
+            C_times_states = (C[..., None, :] * states[:, :, None, ...])
+            state_decay_out_permuted = state_decay_out.permute(0, 2, 3, 1)
+            Y_off = (C_times_states.sum(-1) * state_decay_out_permuted[..., None])
+
+            # Add output of intra-chunk and inter-chunk terms (diagonal and off-diagonal blocks)
+            y = Y_diag + Y_off
+            # [bsz, -1, self.chunk_size, num_heads, head_dim] -> [bsz, (padded) seq_len, num_heads, head_dim]
+            y = y.reshape(batch_size, -1, self.num_heads, self.head_dim)
+
+            y = y + D_residual
+            # Cutting off padded chunks
+            if pad_size > 0:
+                y = y[:, :seq_len, :, :]
+            y = y.reshape(batch_size, seq_len, -1)
+
+            # Init cache
+            if ssm_state is not None and cache_params is not None:
+                cache_params.update_ssm_state(layer_idx=self.layer_idx, new_ssm_state=ssm_state)
+
+        scan_output = self.norm(y, gate)
+
+        # end ssd naive
+
+        # 4. Final linear projection
+        contextualized_states = self.out_proj(scan_output.to(dtype))  # [batch, seq_len, hidden_size]
+        return contextualized_states
+    # fmt: on
+
+    def forward(
+        self,
+        hidden_states,
+        cache_params: Optional[HybridMambaAttentionDynamicCache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+    ):
+        if is_fast_path_available and "cuda" in self.in_proj.weight.device.type:
+            return self.cuda_kernels_forward(
+                hidden_states, cache_params, cache_position, attention_mask
+            )
+        dtype = hidden_states.dtype
+        if (
+            attention_mask is not None
+            and attention_mask.shape[1] > 1
+            and attention_mask.shape[0] > 1
+        ):
+            # tune out hidden states for pad tokens, see https://github.com/state-spaces/mamba/issues/66
+            hidden_states = (hidden_states * attention_mask[:, :, None]).to(dtype)
+
+        return self.torch_forward(
+            hidden_states, cache_params, cache_position, attention_mask
+        )
+
+
+class NemotronHRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        NemotronHRMSNorm is equivalent to T5LayerNorm and LlamaRMSNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        # Weights are in float32
+        return (self.weight.to(torch.float32) * hidden_states).to(input_dtype)
+
+
+class NemotronHBlock(nn.Module):
+    def __init__(self, config, layer_idx):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.residual_in_fp32 = config.residual_in_fp32
+        self.norm = NemotronHRMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
+
+        # M: Mamba2, *: Attention, -: MLP
+        self.block_type = config.layers_block_type[layer_idx]
+        if self.block_type == "mamba":
+            self.mixer = NemotronHMamba2Mixer(config, layer_idx=layer_idx)
+        elif self.block_type == "attention":
+            self.mixer = NEMOTRONH_ATTENTION_CLASSES[config._attn_implementation](
+                config, layer_idx=layer_idx
+            )
+        elif self.block_type == "mlp":
+            self.mixer = NemotronHMLP(config, layer_idx=layer_idx)
+        else:
+            raise ValueError(
+                f"Invalid layer pattern {config.hybrid_override_pattern[layer_idx]}"
+            )
+
+    def forward(
+        self,
+        hidden_states,
+        cache_params: Optional[HybridMambaAttentionDynamicCache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+    ):
+        with torch.cuda.stream(torch.cuda.default_stream(hidden_states.device)):
+            # * Use torch.cuda.stream() to avoid NaN issues when using multiple GPUs
+            residual = hidden_states
+            hidden_states = self.norm(hidden_states.to(dtype=self.norm.weight.dtype))
+            if self.residual_in_fp32:
+                residual = residual.to(torch.float32)
+
+            if self.block_type == "mamba":
+                hidden_states = self.mixer(
+                    hidden_states,
+                    cache_params=cache_params,
+                    cache_position=cache_position,
+                )
+            elif self.block_type == "attention":
+                hidden_states = self.mixer(hidden_states, cache_position=cache_position)
+                hidden_states = hidden_states[0]
+            elif self.block_type == "mlp":
+                hidden_states = self.mixer(hidden_states)
+            else:
+                raise ValueError(f"Invalid block_type: {self.block_type}")
+
+            hidden_states = residual + hidden_states
+            return hidden_states
+
+
+# Copied from transformers.models.nemotron.modeling_nemotron Nemotron->NemotronH
+class NemotronHMLP(nn.Module):
+    def __init__(self, config, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            logger.warning_once(
+                f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will "
+                "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.up_proj = nn.Linear(
+            self.hidden_size, self.intermediate_size, bias=config.mlp_bias
+        )
+        self.down_proj = nn.Linear(
+            self.intermediate_size, self.hidden_size, bias=config.mlp_bias
+        )
+        self.act_fn = ACT2FN[config.mlp_hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.up_proj(x)))
+
+
+# Copied from transformers.models.llama.modeling_llama.repeat_kv
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(
+        batch, num_key_value_heads, n_rep, slen, head_dim
+    )
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+class NemotronHAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: NemotronHConfig, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            logger.warning_once(
+                f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will "
+                "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+
+        self.attention_dropout = config.attention_dropout
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        if config.attention_head_dim is not None:
+            self.head_dim = config.attention_head_dim
+        else:
+            self.head_dim = config.hidden_size // config.num_attention_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.is_causal = True
+
+        self.q_proj = nn.Linear(
+            self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.k_proj = nn.Linear(
+            self.hidden_size,
+            self.num_key_value_heads * self.head_dim,
+            bias=config.attention_bias,
+        )
+        self.v_proj = nn.Linear(
+            self.hidden_size,
+            self.num_key_value_heads * self.head_dim,
+            bias=config.attention_bias,
+        )
+        self.o_proj = nn.Linear(
+            self.head_dim * self.num_heads, self.hidden_size, bias=config.attention_bias
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        # position_embeddings: Tuple[torch.Tensor, torch.Tensor], #TODO
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[HybridMambaAttentionDynamicCache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+        key_states = key_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+        value_states = value_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+
+        if past_key_value is not None:
+            key_states, value_states = past_key_value.update(
+                key_states, value_states, self.layer_idx
+            )
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        causal_mask = attention_mask
+        if attention_mask is not None:  # no matter the length, we just slice it
+            causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+
+        if query_states.device.type == "cuda" and attention_mask is not None:
+            query_states = query_states.contiguous()
+            key_states = key_states.contiguous()
+            value_states = value_states.contiguous()
+
+        is_causal = True if causal_mask is None and q_len > 1 else False
+
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=causal_mask,
+            dropout_p=self.attention_dropout if self.training else 0.0,
+            is_causal=is_causal,
+        )
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        # attn_output = attn_output.view(bsz, q_len, self.hidden_size)
+        attn_output = attn_output.view(bsz, q_len, self.num_heads * self.head_dim)
+
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+# Adapted from transformers.models.mistral.modeling_mistral.MistralFlashAttention2 with Mistral->Jamba
+# class JambaFlashAttention2(JambaAttention):
+class NemotronHFlashAttention2(NemotronHAttention):
+    """
+    Jamba flash attention module. This module inherits from `JambaAttention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[HybridMambaAttentionDynamicCache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ):
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        # Flash attention requires the input to have the shape
+        # batch_size x seq_length x head_dim x hidden_dim
+        # therefore we just need to keep the original shape
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim)
+        key_states = key_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+        value_states = value_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+
+        if past_key_value is not None:
+            key_states, value_states = past_key_value.update(
+                key_states, value_states, self.layer_idx
+            )
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+        dropout_rate = 0.0 if not self.training else self.attention_dropout
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in float16 just to be sure everything works as expected.
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        # Reashape to the expected shape for Flash Attention
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        attn_output = _flash_attention_forward(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            q_len,
+            dropout=dropout_rate,
+            sliding_window=getattr(self.config, "sliding_window", None),
+            is_causal=self.is_causal,
+            use_top_left_mask=self._flash_attn_uses_top_left_mask,
+        )
+
+        # attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = attn_output.reshape(
+            bsz, q_len, self.num_heads * self.head_dim
+        ).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+# Adapted from transformers.models.mistral.modeling_mistral.MistralSdpaAttention with Mistral->Jamba
+# class JambaSdpaAttention(JambaAttention):
+class NemotronHSdpaAttention(NemotronHAttention):
+    """
+    Jamba attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from
+    `JambaAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to
+    SDPA API.
+    """
+
+    # Adapted from NemotronHAttention.forward
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[HybridMambaAttentionDynamicCache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if output_attentions:
+            # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
+            logger.warning_once(
+                "NemotronHModel is using NemotronHSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, "
+                'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+            )
+            return super().forward(
+                hidden_states=hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+            )
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+        key_states = key_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+        value_states = value_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+
+        if past_key_value is not None:
+            key_states, value_states = past_key_value.update(
+                key_states, value_states, self.layer_idx
+            )
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        causal_mask = attention_mask
+        if attention_mask is not None:
+            causal_mask = causal_mask[:, :, :, : key_states.shape[-2]]
+
+        # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
+        # Reference: https://github.com/pytorch/pytorch/issues/112577.
+        if query_states.device.type == "cuda" and attention_mask is not None:
+            query_states = query_states.contiguous()
+            key_states = key_states.contiguous()
+            value_states = value_states.contiguous()
+
+        # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment
+        # in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling.
+        # The q_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create a causal mask in case q_len == 1.
+        is_causal = (
+            True if self.is_causal and causal_mask is None and q_len > 1 else False
+        )
+
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=causal_mask,
+            dropout_p=self.attention_dropout if self.training else 0.0,
+            is_causal=is_causal,
+        )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.view(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+NEMOTRONH_ATTENTION_CLASSES = {
+    "eager": NemotronHAttention,
+    "flash_attention_2": NemotronHFlashAttention2,
+    "sdpa": NemotronHSdpaAttention,
+}
+
+
+# Copied from transformers.models.mamba.modeling_mamba2.Mamba2PreTrainedModel
+class NemotronHPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = NemotronHConfig
+    base_model_prefix = "backbone"
+    _no_split_modules = ["NemotronHBlock"]
+    supports_gradient_checkpointing = True
+    _is_stateful = True
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, NemotronHMamba2Mixer):
+            module.A_log._no_weight_decay = True
+            module.D._no_weight_decay = True
+
+            dt = torch.exp(
+                torch.rand(self.config.mamba_num_heads)
+                * (
+                    math.log(self.config.time_step_max)
+                    - math.log(self.config.time_step_min)
+                )
+                + math.log(self.config.time_step_min)
+            ).clamp(min=self.config.time_step_floor)
+
+            # # Inverse of softplus: https://github.com/pytorch/pytorch/issues/72759
+            inv_dt = dt + torch.log(-torch.expm1(-dt))
+            with torch.no_grad():
+                module.dt_bias.copy_(inv_dt)
+            module.dt_bias._no_reinit = True
+
+        if isinstance(module, nn.Linear):
+            if module.bias is not None:
+                if not getattr(module.bias, "_no_reinit", False):
+                    nn.init.zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            nn.init.normal_(module.weight, std=self.config.initializer_range)
+
+        # TODO: Check
+        if self.config.rescale_prenorm_residual:
+            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+            #
+            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+            for name, p in module.named_parameters():
+                if name in ["out_proj.weight"]:
+                    # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
+                    # Following Pytorch init, except scale by 1/sqrt(2 * n_layer)
+                    # We need to reinit p since this code could be called multiple times
+                    # Having just p *= scale would repeatedly scale it down
+                    nn.init.kaiming_uniform_(p, a=math.sqrt(5))
+                    with torch.no_grad():
+                        p /= math.sqrt(self.config.num_hidden_layers)
+
+
+@dataclass
+# Copied from transformers.models.mamba.modeling_mamba2.Mamba2Output with MAMBA2->NemotronH,Mamba2->NemotronH
+class NemotronHOutput(ModelOutput):
+    """
+    Class for the NemotronH model outputs.
+
+    Args:
+        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        cache_params (`HybridMambaAttentionDynamicCache`):
+            The state of the model at the last time step. Can be used in a forward method with the next `input_ids` to
+            avoid providing the old `input_ids`.
+
+            Includes both the State space model state matrices after the selective scan, and the Convolutional states
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+    """
+
+    last_hidden_state: Optional[torch.FloatTensor] = None
+    cache_params: Optional[HybridMambaAttentionDynamicCache] = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+
+
+@dataclass
+# Copied from transformers.models.mamba2.modeling_mamba2.MambaCausalLMOutput with Mamba2->NemotronH
+class NemotronHCausalLMOutput(ModelOutput):
+    """
+    Base class for causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        cache_params (`HybridMambaAttentionDynamicCache`):
+            The state of the model at the last time step. Can be used in a forward method with the next `input_ids` to
+            avoid providing the old `input_ids`.
+
+            Includes both the State space model state matrices after the selective scan, and the Convolutional states
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    logits: Optional[torch.FloatTensor] = None
+    cache_params: Optional[HybridMambaAttentionDynamicCache] = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+
+
+NEMOTRONH_START_DOCSTRING = r"""
+
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`NemotronHConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+NEMOTRONH_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`, *optional*):
+            Indices of input sequence tokens in the vocabulary.
+
+            If `cache_params.seqlen_offset>0`, only `input_ids` that do not have their past calculated should be passed as
+            `input_ids`.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        position_ids (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings.
+        cache_params (`HybridMambaAttentionDynamicCache`, *optional*):
+            If passed along, the model uses the previous state in all the blocks (which will give the output for the
+            `input_ids` provided as if the model add `state_input_ids + input_ids` as context).
+        use_cache (`bool`, *optional*):
+            If set to `True`, the `cache_params` is returned and can be used to quickly generate the next logits.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        cache_position (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            The position of the current input in the cache. This is used to ensure that the cache is correctly updated.
+            If `cache_params` is passed, `cache_position` should also be passed.
+        attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+"""
+
+
+@add_start_docstrings(
+    "The bare NemotronH Model transformer outputting raw hidden-states without any specific head on top.",
+    NEMOTRONH_START_DOCSTRING,
+)
+class NemotronHModel(NemotronHPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size)
+        self.layers = nn.ModuleList(
+            [
+                NemotronHBlock(config, layer_idx=idx)
+                for idx in range(config.num_hidden_layers)
+            ]
+        )
+
+        self.gradient_checkpointing = False
+        self.norm_f = NemotronHRMSNorm(
+            config.hidden_size, eps=config.layer_norm_epsilon
+        )
+        # Initialize weights and apply final processing
+        self._register_load_state_dict_pre_hook(self.load_hook)
+        self.post_init()
+
+    def load_hook(self, state_dict, prefix, *args):
+        for k in state_dict:
+            if "embedding." in k:
+                state_dict[k.replace("embedding.", "embeddings.")] = state_dict.pop(k)
+                break
+
+    def get_input_embeddings(self):
+        return self.embeddings
+
+    def set_input_embeddings(self, new_embeddings):
+        self.embeddings = new_embeddings
+
+    @add_start_docstrings_to_model_forward(NEMOTRONH_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=NemotronHOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        inputs_embeds: Optional[torch.LongTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        cache_params: Optional[HybridMambaAttentionDynamicCache] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        **kwargs,
+    ) -> Union[Tuple, NemotronHOutput]:
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        # use_cache = use_cache if use_cache is not None else self.config.use_cache
+        use_cache = (
+            use_cache
+            if use_cache is not None
+            else (self.config.use_cache if not self.training else False)
+        )
+
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        if (input_ids is None) ^ (inputs_embeds is not None):  # ^ is python for xor
+            raise ValueError(
+                "You must specify exactly one of input_ids or inputs_embeds"
+            )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embeddings(input_ids)
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
+            )
+            use_cache = False
+
+        # From zamba_modeling.py
+        if use_cache and cache_params is None:
+            logger.warning_once(
+                "NemotronH requires an initialized `NemotronHHybridDynamicCache` to return a cache. None was "
+                "provided, so no cache will be returned."
+            )
+
+        hidden_states = inputs_embeds
+
+        if cache_position is None:
+            cache_position = torch.arange(
+                hidden_states.shape[1], device=hidden_states.device
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = self._update_causal_mask(
+            attention_mask, inputs_embeds, cache_position
+        )
+        mamba_mask = self._update_mamba_mask(attention_mask, cache_position)
+
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        # Until HERE
+
+        for layer_idx, mixer_block in enumerate(self.layers):
+            # Depending on the layer type we opt for 2D base attention mask (Mamba) or 4D causal mask (Attention)
+            if mixer_block.block_type == "mamba":
+                layer_mask = mamba_mask
+            elif mixer_block.block_type == "attention":
+                layer_mask = causal_mask
+            elif mixer_block.block_type == "mlp":
+                layer_mask = None
+            else:
+                raise ValueError(f"Invalid block_type: {self.block_type}")
+
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                hidden_states = self._gradient_checkpointing_func(
+                    mixer_block.__call__,
+                    hidden_states,
+                    cache_params,
+                    cache_position,
+                    layer_mask,
+                )
+            else:
+                hidden_states = mixer_block(
+                    hidden_states,
+                    cache_params=cache_params,
+                    cache_position=cache_position,
+                    attention_mask=layer_mask,
+                )
+
+            # TODO: Store attentions
+            # if output_attentions:
+            #     if layer_outputs[1] is not None:
+            #         # append attentions only of attention layers. Mamba layers return `None` as the attention weights
+            #         all_self_attns += (layer_outputs[1],)
+
+            # TODO (Check): should it happen before the forward pass?
+            # if output_hidden_states:
+            #     all_hidden_states = all_hidden_states + (hidden_states,)
+
+        hidden_states = self.norm_f(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, cache_params, all_hidden_states]
+                if v is not None
+            )
+
+        return NemotronHOutput(
+            last_hidden_state=hidden_states,
+            cache_params=cache_params if use_cache else None,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+    # Copied from transformers.models.jamba.modeling_jamba.JambaModel._update_causal_mask
+    def _update_causal_mask(self, attention_mask, input_tensor, cache_position):
+        if self.config._attn_implementation == "flash_attention_2":
+            if attention_mask is not None and 0.0 in attention_mask:
+                return attention_mask
+            return None
+
+        dtype, device = input_tensor.dtype, input_tensor.device
+        min_dtype = torch.finfo(dtype).min
+        sequence_length = input_tensor.shape[1]
+        target_length = cache_position[-1] + 1
+
+        causal_mask = torch.full(
+            (sequence_length, target_length),
+            fill_value=min_dtype,
+            dtype=dtype,
+            device=device,
+        )
+        if sequence_length != 1:
+            causal_mask = torch.triu(causal_mask, diagonal=1)
+        causal_mask *= torch.arange(
+            target_length, device=device
+        ) > cache_position.reshape(-1, 1)
+        causal_mask = causal_mask[None, None, :, :].expand(
+            input_tensor.shape[0], 1, -1, -1
+        )
+        if attention_mask is not None:
+            causal_mask = (
+                causal_mask.clone()
+            )  # copy to contiguous memory for in-place edit
+            if attention_mask.dim() == 2:
+                mask_length = attention_mask.shape[-1]
+                padding_mask = causal_mask[..., :mask_length].eq(0.0) * attention_mask[
+                    :, None, None, :
+                ].eq(0.0)
+                causal_mask[..., :mask_length] = causal_mask[
+                    ..., :mask_length
+                ].masked_fill(padding_mask, min_dtype)
+
+        if (
+            self.config._attn_implementation == "sdpa"
+            and attention_mask is not None
+            and attention_mask.device.type == "cuda"
+        ):
+            # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when
+            # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path.
+            # Details: https://github.com/pytorch/pytorch/issues/110213
+            causal_mask = AttentionMaskConverter._unmask_unattended(
+                causal_mask, min_dtype
+            )
+
+        return causal_mask
+
+    def _update_mamba_mask(self, attention_mask, cache_position):
+        """
+        No need for zeroing states when
+            1. Cached forward
+            2. Attending to all inputs
+        """
+        mamba_mask = attention_mask
+        if cache_position[0] > 0 or (
+            attention_mask is not None and torch.all(attention_mask == 1)
+        ):
+            mamba_mask = None
+        return mamba_mask
+
+
+@add_start_docstrings(
+    """
+    The NEMOTRONH Model transformer with a language modeling head on top (linear layer with weights not tied to the input
+    embeddings).
+    """,
+    NEMOTRONH_START_DOCSTRING,
+)
+class NemotronHForCausalLM(NemotronHPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.backbone = NemotronHModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.backbone.get_input_embeddings()
+
+    def set_input_embeddings(self, new_embeddings):
+        return self.backbone.set_input_embeddings(new_embeddings)
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def get_decoder(self):
+        return self.model
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        cache_position=None,
+        position_ids=None,
+        use_cache=True,
+        **kwargs,
+    ):
+        # Copy from https://github.com/huggingface/transformers/blob/main/src/transformers/models/jamba/modeling_jamba.py
+        # Overwitten -- uses `cache_params` as opposed to `past_key_values`
+        empty_past_kv = past_key_values is None
+
+        # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens
+        # Exception 1: when passing input_embeds, input_ids may be missing entries
+        # Exception 2: some generation methods do special slicing of input_ids, so we don't need to do it here
+        # Exception 3: with synced GPUs cache_position may go out of bounds, but we only want dummy token in that case.
+        #              (we can't check exception 3 while compiling)
+        if not empty_past_kv:
+            if (
+                inputs_embeds is not None  # Exception 1
+                or cache_position[-1] >= input_ids.shape[1]  # Exception 3
+            ):
+                input_ids = input_ids[:, -cache_position.shape[0] :]
+            elif (
+                input_ids.shape[1] != cache_position.shape[0]
+            ):  # Default case (the "else", a no op, is Exception 2)
+                input_ids = input_ids[:, cache_position]
+        else:
+            past_key_values = HybridMambaAttentionDynamicCache(
+                self.config, input_ids.shape[0], self.dtype, device=self.device
+            )
+
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if not empty_past_kv:
+                position_ids = position_ids[:, -input_ids.shape[1] :]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and empty_past_kv:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {
+                "input_ids": input_ids.contiguous()
+            }  # `contiguous()` needed for compilation use cases
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": use_cache,
+                "attention_mask": attention_mask,
+                "logits_to_keep": self.config.num_logits_to_keep,
+                "cache_position": cache_position,
+            }
+        )
+        return model_inputs
+
+    @add_start_docstrings_to_model_forward(NEMOTRONH_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=NemotronHCausalLMOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        cache_params: Optional[HybridMambaAttentionDynamicCache] = None,
+        labels: Optional[torch.LongTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        **kwargs,  # for now we need this for generation
+    ) -> Union[Tuple, NemotronHCausalLMOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        nemotron_h_outputs = self.backbone(
+            input_ids,
+            cache_params=cache_params,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            attention_mask=attention_mask,
+        )
+        hidden_states = nemotron_h_outputs[0]
+
+        # TODO: Check zamba_modeling.py: https://github.com/huggingface/transformers/blob/d7188ba600e36d3fd191b12e19f1b3bb81a8404f/src/transformers/models/zamba/modeling_zamba.py#L1284C1-L1286C2
+        # logits = self.lm_head(hidden_states.to(self.lm_head.weight.dtype)).float()
+        logits = self.lm_head(hidden_states.to(self.lm_head.weight.dtype)).float()
+
+        loss = None
+        if labels is not None:
+            # move labels to correct device to enable model parallelism
+            labels = labels.to(logits.device)
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(
+                shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1)
+            )
+
+        if not return_dict:
+            output = (logits,) + nemotron_h_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return NemotronHCausalLMOutput(
+            loss=loss,
+            logits=logits,
+            cache_params=nemotron_h_outputs.cache_params,
+            hidden_states=nemotron_h_outputs.hidden_states,
+            attentions=nemotron_h_outputs.attentions,
+        )

special_tokens_map.json

@@ -0,0 +1,44 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,51 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}