sagnikM/qwen_qwen_step100_posterior

HiLL Qwen2.5-7B posterior — step 100

The privileged hint generator converted from the step-100 FSDP checkpoint of the HiLL Qwen2.5-7B/OpenThoughts run.

from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "sagnikM/qwen_qwen_step100_posterior"
model = AutoModelForCausalLM.from_pretrained(repo_id, torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(repo_id)

Only model, configuration, and tokenizer artifacts are uploaded; optimizer and trainer state remain in the original VERL checkpoint.