i-Coder/iCoder-27B

Full fine-tune vs Qwen/Qwen3.6-27B. Fine-tune: 94% of the weights changed, across 64 of 64 layers.

Tensors changed: Embeddings 0/4, MLP 192/304, Attention 397/556, lm_head 1/1, Norms 160/319

2 derivatives checked