Modelpublic

anirudhmamgain/latest_phi2_grpo_combined_rewards_strict

sourceHugging Faceapache-2.0updated 11mo agoView on Hugging Face

0likes7downloads

Model Card

Phi-2 GRPO Combined Rewards (Strict)

This is a fine-tuned version of Phi-2 using GRPO with combined reward signals (factual accuracy + format adherence + chain-of-thought strictness).

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("anirudhmamgain/latest_phi2_grpo_combined_rewards_strict", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("anirudhmamgain/latest_phi2_grpo_combined_rewards_strict", trust_remote_code=True)