Language Modeling on Pythia-160M
14.47Delta Perplexity (ΔPPL)Worst
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Worstk=12026.04 | 14.47 | — | |
| Randomk=12026.04 | 14.5 | — | |
| Zone-Aware Spectral Layer Pruningk=12026.04 | 35.21 | 1.2 | |
| Last-Nk=12026.04 | 43.14 | — | |
| Zone-Aware Spectral Layer Pruningk=22026.04 | 101.03 | 1.1 | |
| Last-Nk=22026.04 | 111.94 | — | |
| Randomk=22026.04 | 123.59 | — | |
| Zone-Aware Spectral Layer Pruningk=32026.04 | 195.23 | 2 | |
| Last-Nk=32026.04 | 389.77 | — | |
| Zone-Aware Spectral Layer Pruningk=42026.04 | 501.59 | 1.3 | |
| Randomk=32026.04 | 591.06 | — | |
| Last-Nk=42026.04 | 633.19 | — | |
| Randomk=42026.04 | 1,224 | — | |
| Worstk=22026.04 | 34,875 | — | |
| Worstk=32026.04 | 156,415 | — | |
| Worstk=42026.04 | 5,959,385 | — |