Commit History

  1. Upload folder using huggingface_hub

    Verified signature
    4cd2128
  2. Fix: decode with skip_special_tokens=False to preserve <think> tags for stripping

    Verified signature
    231d76f
  3. Fix Qwen3 thinking mode + increase max_new_tokens: training/rollout.py

    Verified signature
    f788adf
  4. Sync InvoiceGuard code for GRPO training job

    Verified signature
    d1e9e92
  5. Sync InvoiceGuard code for GRPO training job

    Verified signature
    9a88af0