Language Modeling on 12 diverse evaluation prompts (held-out)
21.45Held-out PerplexityBLOOM-1b7 (Stock)
Evaluation results
| Method | Links | |
|---|---|---|
| BLOOM-1b7 (Stock)Surgical Stage=Stock, Params trained=—2026.03 | 21.45 | |
| Surgical Reinitialization (Pass 1)Surgical Stage=Pass 1, Epochs=E3, Params trained=302M (17.5%)2026.03 | 27.52 | |
| Surgical Reinitialization (Pass 2)Surgical Stage=Pass 2, Epochs=E1, Params trained=40M (2.3%)2026.03 | 28.76 |