INT2 Quantization on OPT-2.7B
6.43Memory Footprint (GB)OPTQ
Evaluation Results
| Method | Links | |
|---|---|---|
| OPTQTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 6.43 | |
| Z-FOLDTarget=layer-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 6.43 | |
| OPTQTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 6.43 | |
| Z-FOLDTarget=layer-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 6.43 | |
| aespaTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 6.84 | |
| aespaTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 6.84 | |
| OmniQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 7.09 | |
| OmniQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 7.09 | |
| AffineQuantTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 12.25 | |
| AffineQuantTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 12.25 | |
| BRECQTarget=attention-wise reconstruction, Hardware=NVIDIA A100 GPU2024.02 | 27.79 | |
| BRECQTarget=attention-wise reconstruction, Hardware=single NVIDIA A100 GPU2024.02 | 27.79 |