Instructions to use lomahony/pythia-2.8b-helpful-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use lomahony/pythia-2.8b-helpful-sft with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="lomahony/pythia-2.8b-helpful-sft")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("lomahony/pythia-2.8b-helpful-sft")
model = AutoModelForCausalLM.from_pretrained("lomahony/pythia-2.8b-helpful-sft")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use lomahony/pythia-2.8b-helpful-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "lomahony/pythia-2.8b-helpful-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lomahony/pythia-2.8b-helpful-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/lomahony/pythia-2.8b-helpful-sft

SGLang

How to use lomahony/pythia-2.8b-helpful-sft with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "lomahony/pythia-2.8b-helpful-sft" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lomahony/pythia-2.8b-helpful-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "lomahony/pythia-2.8b-helpful-sft" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lomahony/pythia-2.8b-helpful-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use lomahony/pythia-2.8b-helpful-sft with Docker Model Runner:
```
docker model run hf.co/lomahony/pythia-2.8b-helpful-sft
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Pythia-2.8b supervised finetuned using TRLx library with the helpful subset of Anthropic-hh-rlhf dataset for 1 epoch.

Checkpoints are also uploaded.

Fully reproducible finetuning code is available on GitHub

wandb log

See Pythia-2.8b for model details (paper).

See further details of these models in the paper Attributing Mode Collapse in the Fine-Tuning of Large Language Models.

You can cite these models if they are helpful as follows:

@inproceedings{o2024attributing,
  title={Attributing Mode Collapse in the Fine-Tuning of Large Language Models},
  author={O’Mahony, Laura and Grinsztajn, Leo and Schoelkopf, Hailey and Biderman, Stella},
  booktitle={ICLR 2024, Mathematical and Empirical Understanding of Foundation Models (ME-FoMo) workshop},
  year={2024}
}

hf (pretrained=lomahony/pythia-2.8b-helpful-sft), gen_kwargs: (None), limit: None, num_fewshot: 0, batch_size: 16

Tasks	Version	Filter	Metric	Value		Stderr
arc_challenge	1	none	acc	0.2901	±	0.0133
		none	acc_norm	0.3404	±	0.0138
arc_easy	1	none	acc	0.6469	±	0.0098
		none	acc_norm	0.5766	±	0.0101
boolq	2	none	acc	0.6361	±	0.0084
hellaswag	1	none	acc	0.4557	±	0.0050
		none	acc_norm	0.5984	±	0.0049
lambada_openai	1	none	perplexity	5.2226	±	0.1377
		none	acc	0.6210	±	0.0068
openbookqa	1	none	acc	0.2640	±	0.0197
		none	acc_norm	0.3760	±	0.0217
piqa	1	none	acc	0.7481	±	0.0101
		none	acc_norm	0.7481	±	0.0101
sciq	1	none	acc	0.8800	±	0.0103
		none	acc_norm	0.8180	±	0.0122
wikitext	2	none	word_perplexity	13.4928	±	N/A
		none	byte_perplexity	1.6268	±	N/A
		none	bits_per_byte	0.7020	±	N/A
winogrande	1	none	acc	0.6125	±	0.0137

hf (pretrained=lomahony/pythia-2.8b-helpful-sft), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: 16

Tasks	Version	Filter	n-shot	Metric	Value		Stderr
arc_challenge	1	none	5	acc	0.3285	±	0.0137
		none	5	acc_norm	0.3677	±	0.0141
arc_easy	1	none	5	acc	0.6873	±	0.0095
		none	5	acc_norm	0.6835	±	0.0095
boolq	2	none	5	acc	0.6670	±	0.0082
hellaswag	1	none	5	acc	0.4542	±	0.0050
		none	5	acc_norm	0.5963	±	0.0049
lambada_openai	1	none	5	perplexity	7.4076	±	0.2095
		none	5	acc	0.5486	±	0.0069
openbookqa	1	none	5	acc	0.2680	±	0.0198
		none	5	acc_norm	0.3620	±	0.0215
piqa	1	none	5	acc	0.7568	±	0.0100
		none	5	acc_norm	0.7486	±	0.0101
sciq	1	none	5	acc	0.9380	±	0.0076
		none	5	acc_norm	0.9330	±	0.0079
wikitext	2	none	5	word_perplexity	13.4928	±	N/A
		none	5	byte_perplexity	1.6268	±	N/A
		none	5	bits_per_byte	0.7020	±	N/A
winogrande	1	none	5	acc	0.5935	±	0.0138