Language Modeling on WikiText-103 (Perplexity and Efficiency)
13.92PerplexityMCWC (K = 4)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MCWC (K = 4)Model=Pythia-2.8B, bits per model parameter=4.22026.05 | 13.92 | 1,480 | 4.6 | |
| AWQ W4Model=Pythia-2.8B, bits per model parameter=4.22026.05 | 14.28 | 1,475 | 4.4 |