INT2 Quantization on OPT-6.7B
11.68Memory Footprint (GB)OmniQuant
Evaluation Results
| Method | Links | |
|---|---|---|
| OmniQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 11.68 | |
| OmniQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 11.68 | |
| aespaTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 12.26 | |
| aespaTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 12.26 | |
| OPTQTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 13.07 | |
| Z-FOLDTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 13.07 | |
| OPTQTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 13.07 | |
| Z-FOLDTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 13.07 | |
| AffineQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 20.08 | |
| AffineQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 20.08 |