---
base_model: openai/gpt-oss-20b
library_name: peft
tags:
- terminal-bench
- gpt-oss
- lora
- terminus-2
---

# gpt-oss-20b TB2 SFT LoRA (PEFT)

LoRA adapter from Tinker SFT on Nemotron 500 terminus-2 trajectories.

| Field | Value |
| --- | --- |
| Base model | `openai/gpt-oss-20b` |
| LoRA rank | 32 |
| Checkpoint | step **250** |
| Tinker | `tinker://bf2715a6-440c-54c8-876c-a83603f51849:train:0/sampler_weights/000250` |
| W&B | [https://wandb.ai/alchemxz/afterquery-tb2-sft/runs/a4rbl6xl](https://wandb.ai/alchemxz/afterquery-tb2-sft/runs/a4rbl6xl) |

## Serve (vLLM)

```bash
vllm serve openai/gpt-oss-20b --enable-lora --enable-mixed-moe-lora-format \
  --lora-modules '{"name":"tb2-sft","path":"lilyzhng/gpt-oss-20b-tb2-sft-lora","is_3d_lora_weight":false}' \
  --trust-remote-code
```

Harbor eval model: `hosted_vllm/tb2-sft` with `api_base=<vllm_url>/v1`.