Huang2020/Qwen3-4B-Domino-b16

Full fine-tune (fewer layers) vs Qwen/Qwen3-4B. Huang2020/Qwen3-4B-Domino-b16 matches part of Qwen/Qwen3-4B (56 of its 398 tensors; the rest of Qwen/Qwen3-4B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 5 of 5 layers.

Tensors changed: Norms 21/21, MLP 15/15, Attention 20/20

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse