Modelpublic

kaist-ai/mistral-orpo-capybara-7k

sourceHugging Facemitupdated 3y agoView on Hugging Face

26likes8.6kdownloads

Model Card

Mistral-ORPO-Capybara-7k (7B)

Mistral-ORPO is a fine-tuned version of mistralai/Mistral-7B-v0.1 using the [odds ratio preference optimization (ORPO)](https://arxiv.org/abs/2403.07691). With ORPO, the model directly learns the preference without the supervised fine-tuning warmup phase.

Mistral-ORPO-ORPO-Capybara-7k is fine-tuned for 2.5 hours on four A100s exclusively on the 7k instances of the distilled Capybara paired multi-turn conversation dataset, argilla/distilabel-capybara-dpo-7k-binarized, by Argilla.

—Github Repository: https://github.com/xfactlab/orpo

👍 Model Performance

1) AlpacaEval & MT-Bench

Model Name	Size	Align	MT-Bench	AlpacaEval 2.0 (LC)
Mistral-<tt>ORPO</tt>-Capybara-7k	7B	<tt>ORPO</tt>	7.44	15.9
Mistral-<tt>ORPO</tt>-β	7B	<tt>ORPO</tt>	7.32	14.7
Zephyr β	7B	DPO	7.34	13.2
TULU-2-DPO	13B	DPO	7.00	11.6
Llama-2-Chat	7B	RLHF	6.27	5.4
Llama-2-Chat	13B	RLHF	6.65	8.4

2) IFEval

Model Type	Prompt-Strict	Prompt-Loose	Inst-Strict	Inst-Loose
Mistral-ORPO-Capybara-7k	0.5083	0.5083	0.5827	0.6127
Mistral-ORPO-⍺	0.5009	0.5083	0.5995	0.6163
Mistral-ORPO-β	0.5287	0.5564	0.6355	0.6619

🗺️ MT-Bench by Category

image/png

🖥️ Inference

python

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("kaist-ai/mistral-orpo-capybara-7k")
tokenizer = AutoTokenizer.from_pretrained("kaist-ai/mistral-orpo-capybara-7k")
# Apply chat template
query = [{'role': 'user', 'content': 'Hi! How are you doing?'}]
prompt = tokenizer.apply_chat_template(query, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors='pt')
# Generation with specific configurations
output = model.generate(
  **inputs,
  max_new_tokens=128,
  do_sample=True,
  temperature=0.7
)
response = tokenizer.batch_decode(output)
#<|user|>
#Hi! How are you doing?</s>
#<|assistant|>
#I'm doing well, thank you! How are you?</s>

📎 Citation

@misc{hong2024orpo,
      title={ORPO: Monolithic Preference Optimization without Reference Model}, 
      author={Jiwoo Hong and Noah Lee and James Thorne},
      year={2024},
      eprint={2403.07691},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}