List your company

INT2 Quantization on OPT-6.7B

11.68Memory Footprint (GB)OmniQuant
11.34413.61215.8818.148Feb 14, 2024

Evaluation results

MethodLinks
OmniQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0211.68
OmniQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0211.68
aespaTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0212.26
aespaTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0212.26
OPTQTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0213.07
Z-FOLDTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0213.07
OPTQTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0213.07
Z-FOLDTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0213.07
AffineQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.0220.08
AffineQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.0220.08