nebius/MTP-DeepSeek-V3-0324

Full fine-tune (fewer layers) vs deepseek-ai/DeepSeek-V3-0324. nebius/MTP-DeepSeek-V3-0324 matches part of deepseek-ai/DeepSeek-V3-0324 (1563 of its 91991 tensors; the rest of deepseek-ai/DeepSeek-V3-0324 has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 1 of 1 layers.

Tensors changed: Attention 10/10, Norms 7/7, MLP 28/30, other 2/2

On the Hub · Downloads on Model Pulse