finegrained-grpo-groundplus-0825

Finegrained GRPO GroundPlus 0825

This repository contains the stronger-grounding model used to generate test_data_GROUNDPLUS_VLLM_0825_answers_evaluated_openai_gpt_5_4.

Repository Layout

• merged/: merged full model. Use this for direct vLLM serving.
• adapter/: LoRA adapter checkpoint.

Recommended Environment

pip install "vllm==0.8.1" "transformers==4.50.0"

Load With vLLM Serve

python - <<'PY'
from huggingface_hub import snapshot_download
snapshot_download(
    repo_id="chaitalibh/finegrained-grpo-groundplus-0825",
    local_dir="./finegrained-grpo-groundplus-0825",
    allow_patterns=["merged/*", "README.md"],
)
PY

vllm serve ./finegrained-grpo-groundplus-0825/merged \
  --served-model-name GROUNDPLUS_VLLM_0825 \
  --dtype bfloat16 \
  --max-model-len 8192 \
  --trust-remote-code

Or with Git LFS:

git lfs install
git clone https://huggingface.co/chaitalibh/finegrained-grpo-groundplus-0825
vllm serve ./finegrained-grpo-groundplus-0825/merged \
  --served-model-name GROUNDPLUS_VLLM_0825 \
  --dtype bfloat16 \
  --max-model-len 8192 \
  --trust-remote-code

The adapter/ folder is kept for LoRA workflows. The merged/ folder is the intended path for normal vLLM deployment.