EVA-UNIT-01/EVA-Qwen3-Next-v0.0

Partial fine-tune vs Qwen/Qwen3-Coder-Next-Base. Partial change: 57% of the weights changed (attention, mlp, norms), in 48 of 48 layers.

Tensors changed: Embeddings 0/1, Norms 36/157, Attention 48/264, MLP 2326/2577, lm_head 0/1

2 derivatives checked