Consider using noneUsername/saiga_nemo_12b-W8A8-Dynamic-Per-Token-better. I tweaked the quantization parameters to get better results.
vllm (pretrained=/root/autodl-tmp/saiga_nemo_12b,add_bos_token=true,tensor_parallel_size=2,max_model_len=2048), gen_kwargs: (None), limit: 250.0, num_fewshot: 5, batch_size: auto
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|---|
| gsm8k | 3 | flexible-extract | 5 | exact_match | ↑ | 0.808 | ± | 0.0250 |
| | strict-match | 5 | exact_match | ↑ | 0.760 | ± | 0.0271 |
vllm (pretrained=/root/autodl-tmp/output,add_bos_token=true,tensor_parallel_size=2,max_model_len=2048,dtype=bfloat16), gen_kwargs: (None), limit: 250.0, num_fewshot: 5, batch_size: auto
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|---|
| gsm8k | 3 | flexible-extract | 5 | exact_match | ↑ | 0.792 | ± | 0.0257 |
| | strict-match | 5 | exact_match | ↑ | 0.768 | ± | 0.0268 |
