List your company

INT2 Quantization on LLaMA 7B

8.76Memory Cost (GB)OPTQ
8.139212.329616.5220.7104Feb 14, 2024

Evaluation results

MethodLinks
OPTQTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.028.76
Z-FOLDTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.028.76
OPTQTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.028.76
Z-FOLDTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.028.76
OmniQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0212.61
OmniQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0212.61
aespaTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0221.69
aespaTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0221.69
AffineQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0224.28
AffineQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0224.28