miromind-ai/MiroThinker-v1.5-235B

Full fine-tune vs Qwen/Qwen3-235B-A22B-Thinking-2507. Fine-tune: 100% of the weights changed, across 94 of 94 layers.

Tensors changed: Embeddings 1/1, MLP 2245/2245, Norms 276/377, Attention 376/376, lm_head 1/1

3 derivatives checked