olmo3-7b-grpo-purerl-creativity-step5

Olmo3-7B trained with GRPO (Pure RL, no SFT) on Creativity dataset. Checkpoint: step 5 (val ACC=0.022)

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Echoandland/olmo3-7b-grpo-purerl-creativity-step5")
tokenizer = AutoTokenizer.from_pretrained("Echoandland/olmo3-7b-grpo-purerl-creativity-step5")