Language Modeling on C4 Llama-3.1-8B-Instruct (val)
9.09PerplexityBF16 Baseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BF16 BaselineAvg. Bitwidth=16.0002025.07 | 9.09 | |
| HPTQAvg. Bitwidth=4.1252025.07 | 9.99 | |
| SSQR-5%Avg. Bitwidth=5.7252025.07 | 10.03 | |
| SSQR-4%Avg. Bitwidth=5.4052025.07 | 10.05 | |
| SSQR-3%Avg. Bitwidth=5.0852025.07 | 10.11 | |
| SSQR-2%Avg. Bitwidth=4.7652025.07 | 10.2 | |
| SSQR-1%Avg. Bitwidth=4.4452025.07 | 10.3 | |
| GPTQAvg. Bitwidth=4.1252025.07 | 10.46 | |
| HPTQAvg. Bitwidth=3.1252025.07 | 11.04 | |
| SSQR-5%Avg. Bitwidth=4.7252025.07 | 11.13 | |
| SSQR-4%Avg. Bitwidth=4.4052025.07 | 11.31 | |
| SSQR-3%Avg. Bitwidth=4.0852025.07 | 11.54 | |
| SSQR-2%Avg. Bitwidth=3.7652025.07 | 11.75 | |
| SSQR-1%Avg. Bitwidth=3.4452025.07 | 12.35 | |
| GPTQAvg. Bitwidth=3.1252025.07 | 13.16 | |
| SSQR-5%Avg. Bitwidth=3.7252025.07 | 16.13 | |
| HPTQAvg. Bitwidth=2.1252025.07 | 16.37 | |
| SSQR-4%Avg. Bitwidth=3.4052025.07 | 17.77 | |
| SSQR-3%Avg. Bitwidth=3.0852025.07 | 20.03 | |
| SSQR-2%Avg. Bitwidth=2.7652025.07 | 25.21 | |
| SSQR-1%Avg. Bitwidth=2.4452025.07 | 39.96 | |
| GPTQAvg. Bitwidth=2.1252025.07 | 71.33 |