---
license: apache-2.0
base_model:
- openai/gpt-oss-120b
---

This is the BF16 version and cannot be hosted with vLLM. TensorRT-LLM is supported but not tested.
For the MXFP4 version that is vLLM compatible, check out [gpt-oss-120b-uncensored-mxfp4](https://huggingface.co/huizimao/gpt-oss-120b-uncensored-mxfp4/)

GGUF is available at [bartowski/huizimao_gpt-oss-120b-uncensored-bf16-GGUF](https://huggingface.co/bartowski/huizimao_gpt-oss-120b-uncensored-bf16-GGUF) (Shout out to Bartowski!)

Finetuning is done by LoRA on [Amazon FalseReject](https://huggingface.co/datasets/AmazonScience/FalseReject) train set with 800 samples.

PTQ is done with [NVIDIA ModelOpt](https://github.com/NVIDIA/TensorRT-Model-Optimizer)

Evaluation results obtained on [Amazon FalseReject](https://huggingface.co/datasets/AmazonScience/FalseReject) test set with 300 samples.

| Model Variants | False refusal rate |
|----------|-------------------|
| gpt-oss-120b original (MXFP4) | 70% |
| LoRA (BF16) - this model | 6% |
| LoRA + PTQ (MXFP4) | 24% |

Code example, documentation, and further QAT checkpoints will be released soon.