ai-sage/GigaChat3.5-432B-A28B-Reasoning-bf16

Full fine-tune vs ai-sage/GigaChat3.5-432B-A28B-base. Fine-tune: 100% of the weights changed, across 41 of 41 layers.

Tensors changed: lm_head 1/1, Embeddings 2/2, Norms 600/600, MLP 2071/2071, Attention 246/246, other 2/2

2 derivatives checked