Release WeMM-Embedding-2B-Quantized: Native Sub-1.3GB SafeTensors with <1.0% degradation

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,113 @@
+{
+  "architectures": [
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_token_id": 248056,
+  "model_type": "qwen3_5",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_output_gate": true,
+    "bos_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": 248044,
+    "full_attention_interval": 4,
+    "head_dim": 256,
+    "hidden_act": "silu",
+    "hidden_size": 2048,
+    "initializer_range": 0.02,
+    "intermediate_size": 6144,
+    "layer_types": [
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention"
+    ],
+    "linear_conv_kernel_dim": 4,
+    "linear_key_head_dim": 128,
+    "linear_num_key_heads": 16,
+    "linear_num_value_heads": 16,
+    "linear_value_head_dim": 128,
+    "mamba_ssm_dtype": "float32",
+    "max_position_embeddings": 262144,
+    "mlp_only_layers": [],
+    "model_type": "qwen3_5_text",
+    "mtp_num_hidden_layers": 1,
+    "mtp_use_dedicated_embeddings": false,
+    "num_attention_heads": 8,
+    "num_hidden_layers": 24,
+    "num_key_value_heads": 2,
+    "pad_token_id": null,
+    "partial_rotary_factor": 0.25,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        11,
+        11,
+        10
+      ],
+      "partial_rotary_factor": 0.25,
+      "rope_theta": 10000000,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": false,
+    "use_cache": true,
+    "vocab_size": 248078
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.2.0",
+  "video_token_id": 248057,
+  "vision_config": {
+    "deepstack_visual_indexes": [],
+    "depth": 24,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1024,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "model_type": "qwen3_5",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 2048,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 248054,
+  "vision_start_token_id": 248053,
+  "auto_map": {
+    "AutoModel": "modeling_wemm_embedding.WeMMEmbedding",
+    "AutoModelForCausalLM": "modeling_wemm_embedding.WeMMEmbedding"
+  },
+  "matryoshka_dimensions": [
+    64,
+    128,
+    256,
+    512,
+    1024,
+    2048
+  ]
+}

config_sentence_transformers.json

@@ -0,0 +1,11 @@
+{
+  "__version__": {
+    "pytorch": "2.11.0+cu128",
+    "sentence_transformers": "5.7.0",
+    "transformers": "5.2.0"
+  },
+  "default_prompt_name": null,
+  "model_type": "SentenceTransformer",
+  "prompts": {},
+  "similarity_fn_name": "cosine"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b92c565dd4530b3cecc25eb1147b6e0a6592cd3ac5b734b01f5aa64fc4ccfef
+size 1543511130

modeling_st_wemm.py

@@ -0,0 +1,70 @@
+"""Sentence Transformers module for WeMM-Embedding.
+
+Reproduces the `transformers` usage from the model card inside a Sentence Transformers
+pipeline: vision inputs are prepared with `qwen_vl_utils.process_vision_info` and the
+embedding is read from `WeMMEmbedding.embedding`, which pools the `<embedding>` position and
+L2-normalizes. Everything else (batching, prompts, truncation, `encode_query` /
+`encode_document`, similarity) comes from the stock `Transformer` module.
+"""
+
+from __future__ import annotations
+
+import inspect
+from typing import Any
+
+from sentence_transformers.models import Transformer
+
+
+class WeMMTransformer(Transformer):
+    """`Transformer` that prepares images and videos the way the model card's snippet does."""
+
+    def __init__(self, model_name_or_path: str, **kwargs: Any) -> None:
+        super().__init__(model_name_or_path, **kwargs)
+        vision_config = getattr(self.config, "vision_config", None)
+        self.image_patch_size = int(getattr(vision_config, "patch_size", 16))
+
+        # `embedding` hands its **kwargs to the inner model, so filtering on its own signature
+        # would drop `pixel_values`. Filter on the inner model's parameters instead, plus the
+        # processor's input names for anything the model only accepts as **kwargs.
+        inner_model = getattr(self.model, "model", self.model)
+        signature = set(inspect.signature(inner_model.forward).parameters)
+        signature |= set(getattr(self.processor, "model_input_names", ()))
+        for modality_params in self.modality_config.values():
+            method_name = modality_params["method"]
+            if method_name != "forward":
+                self._method_signature_cache.setdefault(method_name, signature)
+
+    def _apply_chat_template(
+        self,
+        messages: list[list[dict[str, Any]]],
+        modality_kwargs: dict[str, dict[str, Any]],
+        common_kwargs: dict[str, Any],
+        chat_template_kwargs: dict[str, Any],
+    ) -> dict[str, Any]:
+        """Render the chat template and prepare images / videos exactly as the model card does."""
+        from qwen_vl_utils import process_vision_info
+
+        chat_template_kwargs = {"add_generation_prompt": False, **chat_template_kwargs}
+        texts = [
+            self.processor.apply_chat_template(conversation, tokenize=False, **chat_template_kwargs)
+            for conversation in messages
+        ]
+        images, videos, video_kwargs = process_vision_info(
+            [list(conversation) for conversation in messages],
+            image_patch_size=self.image_patch_size,
+            return_video_kwargs=True,
+            return_video_metadata=True,
+        )
+        if videos is not None:
+            videos, video_metadata = (list(part) for part in zip(*videos))
+            video_kwargs = {**video_kwargs, "video_metadata": video_metadata}
+
+        return self.processor(
+            text=texts,
+            images=images,
+            videos=videos,
+            text_kwargs=modality_kwargs["text"],
+            images_kwargs=modality_kwargs["image"],
+            videos_kwargs={**modality_kwargs["video"], **video_kwargs},
+            common_kwargs=common_kwargs,
+        )

modeling_wemm_embedding.py

@@ -0,0 +1,98 @@
+
+import torch.nn as nn
+
+class FlatQuantW4A8Linear(nn.Module):
+    def __init__(self, in_features, out_features, bias=False, group_size=64):
+        super().__init__()
+        self.in_features = in_features
+        self.out_features = out_features
+        self.group_size = group_size
+        self.register_buffer("weight", torch.empty((out_features, in_features // 2), dtype=torch.uint8))
+        self.register_buffer("weight_scale", torch.empty((out_features, in_features // group_size), dtype=torch.bfloat16))
+        if bias:
+            self.bias = nn.Parameter(torch.empty(out_features, dtype=torch.bfloat16))
+        else:
+            self.register_parameter("bias", None)
+
+    def forward(self, x):
+        low = (self.weight & 0x0F).to(torch.int8)
+        high = (self.weight >> 4).to(torch.int8)
+        low = torch.where(low >= 8, low - 16, low)
+        high = torch.where(high >= 8, high - 16, high)
+        unpacked = torch.empty(self.out_features, self.in_features, device=x.device, dtype=x.dtype)
+        unpacked[:, 0::2] = low.to(x.dtype)
+        unpacked[:, 1::2] = high.to(x.dtype)
+        scales = self.weight_scale.to(x.dtype).repeat_interleave(self.group_size, dim=1)
+        return F.linear(x, unpacked * scales, self.bias.to(x.dtype) if self.bias is not None else None)
+
+class FlatQuantFP8Linear(nn.Module):
+    def __init__(self, in_features, out_features, bias=False):
+        super().__init__()
+        self.in_features = in_features
+        self.out_features = out_features
+        self.register_buffer("weight", torch.empty((out_features, in_features), dtype=torch.float8_e4m3fn))
+        self.register_buffer("weight_scale", torch.empty((), dtype=torch.bfloat16))
+        if bias:
+            self.bias = nn.Parameter(torch.empty(out_features, dtype=torch.bfloat16))
+        else:
+            self.register_parameter("bias", None)
+
+    def forward(self, x):
+        w_deq = (self.weight.float() * self.weight_scale.float()).to(x.dtype)
+        return F.linear(x, w_deq, self.bias.to(x.dtype) if self.bias is not None else None)
+
+class FlatQuantW4A8Embedding(nn.Module):
+    def __init__(self, num_embeddings, embedding_dim, group_size=64):
+        super().__init__()
+        self.num_embeddings = num_embeddings
+        self.embedding_dim = embedding_dim
+        self.group_size = group_size
+        self.register_buffer("weight", torch.empty((num_embeddings, embedding_dim // 2), dtype=torch.uint8))
+        self.register_buffer("weight_scale", torch.empty((num_embeddings, embedding_dim // group_size), dtype=torch.bfloat16))
+
+    def forward(self, input_ids):
+        w_packed = self.weight[input_ids]
+        scales = self.weight_scale[input_ids]
+        low = (w_packed & 0x0F).to(torch.int8)
+        high = (w_packed >> 4).to(torch.int8)
+        low = torch.where(low >= 8, low - 16, low)
+        high = torch.where(high >= 8, high - 16, high)
+        *bdims, packed_dim = w_packed.shape
+        unpacked = torch.empty(*bdims, packed_dim * 2, device=input_ids.device, dtype=torch.bfloat16)
+        unpacked[..., 0::2] = low.to(torch.bfloat16)
+        unpacked[..., 1::2] = high.to(torch.bfloat16)
+        return unpacked * scales.to(torch.bfloat16).repeat_interleave(self.group_size, dim=-1)
+
+import torch
+import torch.nn.functional as F
+from transformers import Qwen3_5ForConditionalGeneration
+
+
+class WeMMEmbedding(Qwen3_5ForConditionalGeneration):
+    def embedding(self, input_ids=None, attention_mask=None, **kwargs):
+        # transformers < 5.15 reuses the rope_deltas cached by the previous multimodal
+        # forward for a text-only one, which shifts its position ids.
+        self.model.rope_deltas = None
+
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            **kwargs
+        )
+        
+        last_hidden_state = outputs.last_hidden_state
+
+        if attention_mask is not None:
+            eos_positions = attention_mask.sum(dim=1) - 1
+        else:
+            eos_positions = torch.full((last_hidden_state.shape[0],), last_hidden_state.shape[1] - 1, device=last_hidden_state.device)
+
+        eos_positions = eos_positions.clamp(min=0)
+        
+        batch_indices = torch.arange(last_hidden_state.size(0), device=last_hidden_state.device)
+        
+        embeddings = last_hidden_state[batch_indices, eos_positions]
+        
+        embeddings = F.normalize(embeddings, dim=-1)
+        
+        return embeddings

modules.json

@@ -0,0 +1,8 @@
+[
+  {
+    "idx": 0,
+    "name": "0",
+    "path": "",
+    "type": "modeling_st_wemm.WeMMTransformer"
+  }
+]

processor_config.json

@@ -0,0 +1,60 @@
+{
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Qwen2VLImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "merge_size": 2,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 16777216,
+      "shortest_edge": 65536
+    },
+    "temporal_patch_size": 2
+  },
+  "processor_class": "Qwen3VLProcessor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "fps": 2,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_frames": 768,
+    "merge_size": 2,
+    "min_frames": 4,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "longest_edge": 234881024,
+      "shortest_edge": 4096
+    },
+    "temporal_patch_size": 2,
+    "video_processor_type": "Qwen3VLVideoProcessor"
+  }
+}

sentence_bert_config.json

@@ -0,0 +1,37 @@
+{
+    "transformer_task": "feature-extraction",
+    "modality_config": {
+        "text": {
+            "method": "embedding",
+            "method_output_name": null
+        },
+        "image": {
+            "method": "embedding",
+            "method_output_name": null
+        },
+        "video": {
+            "method": "embedding",
+            "method_output_name": null
+        },
+        "image+text": {
+            "method": "embedding",
+            "method_output_name": null
+        },
+        "text+video": {
+            "method": "embedding",
+            "method_output_name": null
+        },
+        "message": {
+            "method": "embedding",
+            "method_output_name": null,
+            "format": "structured"
+        }
+    },
+    "module_output_name": "sentence_embedding",
+    "processing_kwargs": {
+        "chat_template": {
+            "chat_template": "sentence_transformers"
+        }
+    },
+    "unpad_inputs": false
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40e444c744512f423da4c8443c47c21e22ff76056ba4e9796a81c04c13a9daf0
+size 19990378

tokenizer_config.json

@@ -0,0 +1,32 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": true,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}