OpenEnv-HR-Agent (LoRA adapter)

PEFT LoRA adapter for SentinelHire / OpenEnv HR hiring agent demos.
Base checkpoint: unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit (see adapter_config.json).

Use in the app

Set:

SENTINEL_ADAPTER_REPO=Vijay-1807/OpenEnv-HR-Agent

or download this repo snapshot and point SENTINEL_ADAPTER_PATH at the folder containing adapter_model.safetensors.

Training

Trained with GRPO / Unsloth in this project (train_qwen_grpo.py).
Code: github.com/Vijay-1807/OpenEnv-HR-Agent

Reward curve during RL training: see reward_curve.png in this repo.

Limitations

Inference requires a GPU with enough VRAM for the 4-bit base plus adapter, or a configured CPU/offload path. See the GitHub README for Streamlit setup.