| base_model: Qwen/Qwen3.5-9B | |
| library_name: peft | |
| tags: | |
| - tau2-bench | |
| - qwen3.5 | |
| - lora | |
| # Qwen3.5-9B τ²-bench SFT LoRA | |
| PEFT adapter only — base weights stay on `Qwen/Qwen3.5-9B`. | |
| | Field | Value | | |
| | --- | --- | | |
| | TRAIN_MODE | `exp` | | |
| | LoRA rank | 32 | | |
| | W&B | [https://wandb.ai/alchemxz/decagon-posttraining-sft/runs/bsmjocsv](https://wandb.ai/alchemxz/decagon-posttraining-sft/runs/bsmjocsv) | | |
| ## vLLM (no merge upload) | |
| ```bash | |
| vllm serve Qwen/Qwen3.5-9B --enable-lora --lora-modules sft=lilyzhng/qwen3.5-9b-tau2-retail-sft-lora \ | |
| --max-lora-rank 32 --dtype bfloat16 | |
| ``` | |