README.md

---
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
```