noneUsername/Mistral-Nemo-Instruct-2407-W8A8-Dynamic-Per-Token-better

Note: This model is no longer the optimal W8A8 quantization, please consider using a better quantization model I made later: noneUsername/Mistral-Nemo-Instruct-2407-abliterated-W8A8-Dynamic-Per-Token

vllm (pretrained=/root/autodl-tmp/Mistral-Nemo-Instruct-2407,add_bos_token=true,tensor_parallel_size=2,max_model_len=4096,gpu_memory_utilization=0.85,swap_space=0), 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	↑	0.800	±	0.0253
		strict-match	5	exact_match	↑	0.784	±	0.0261

vllm (pretrained=/root/autodl-tmp/output,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	↑	0.792	±	0.0257
		strict-match	5	exact_match	↑	0.776	±	0.0264

I found some rules about quantization parameters and achieved better results.