Model Compression Analysis on Llama-3.1-8B, Qwen3-14B, Phi-4 Checkpoints
2,974Model Size (MB)BitDelta (scalar)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BitDelta (scalar)Model=Llama-3.1-8B-Instruct, Base Model=Llama-3.1-8B2025.12 | 2,974 | — | |
| Vector (row/col)Model=Llama-3.1-8B-Instruct, Base Model=Llama-3.1-8B2025.12 | 2,980 | — | |
| BitDelta (scalar)Model=Phi-4-reasoning, Base Model=Phi-42025.12 | 3,760 | — | |
| Vector (row/col)Model=Phi-4-reasoning, Base Model=Phi-42025.12 | 3,768 | — | |
| Vector (row/col)Model=Qwen3-14B, Base Model=Qwen3-14B-Base2025.12 | 4,774 | — | |
| BitDelta (scalar)Model=Qwen3-14B, Base Model=Qwen3-14B-Base2025.12 | 4,775 | — |