arun-misra/ai-soar-training

AI-SOAR GRPO Training Space

Trains Qwen2.5-7B-Instruct with LoRA using GRPO reinforcement learning on the Network Defense environment.

Setup

1. 1.Set Secrets in Space settings:
2. 2.HF_TOKEN — HuggingFace token with write access
3. 3.WANDB_API_KEY — your wandb API key

1. 1.Hardware: A10G small (24 GB VRAM) minimum

Files needed in this Space repo

app.py
requirements.txt
models.py               ← copy from vir_env/
server/                 ← copy from vir_env/server/
  __init__.py
  vir_env_environment.py
  app.py