vllm (pretrained=/root/autodl-tmp/output,add_bos_token=true,tensor_parallel_size=2,max_model_len=2048,gpu_memory_utilization=0.97,enforce_eager=true), 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 | ↑ | 0.872 | ± | 0.0212 |
| strict-match | 5 | exact_match | ↑ | 0.860 | ± | 0.0220 |
Since the transformers library lacks solar-related definitions, I edited the py files in the transformers library to complete the quantization. The changed files and the versions of the transformers library are detailed in the zip archive.