Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model:
|
| 4 |
+
- openai/gpt-oss-120b
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
This is the BF16 version and cannot be hosted with vLLM. TensorRT-LLM is supported but not tested.
|
| 8 |
+
For the MXFP4 version that is vLLM compatible, check out [gpt-oss-120b-uncensored-mxfp4](https://huggingface.co/huizimao/gpt-oss-120b-uncensored-mxfp4/)
|
| 9 |
+
|
| 10 |
+
Finetuning is done by LoRA on [Amazon FalseReject](https://huggingface.co/datasets/AmazonScience/FalseReject) train set with 800 samples.
|
| 11 |
+
|
| 12 |
+
PTQ is done with [NVIDIA ModelOpt](https://github.com/NVIDIA/TensorRT-Model-Optimizer)
|
| 13 |
+
|
| 14 |
+
Evaluation results obtained on [Amazon FalseReject](https://huggingface.co/datasets/AmazonScience/FalseReject) test set with 300 samples.
|
| 15 |
+
|
| 16 |
+
| Model Variants | False refusal rate |
|
| 17 |
+
|----------|-------------------|
|
| 18 |
+
| gpt-oss-20b original (MXFP4) | 70% |
|
| 19 |
+
| LoRA (BF16) - this model | 6% |
|
| 20 |
+
| LoRA + PTQ (MXFP4) | 24% |
|
| 21 |
+
|
| 22 |
+
Code example, documentation, and further QAT checkpoints will be released soon.
|