RUC-AIBOX/AEWM
Full fine-tune (fewer layers) vs Qwen/Qwen3.5-35B-A3B. RUC-AIBOX/AEWM matches part of Qwen/Qwen3.5-35B-A3B (1026 of its 1811 tensors; the rest of Qwen/Qwen3.5-35B-A3B has no counterpart). On the shared part: Fine-tune: 99% of the weights changed, across 40 of 40 layers.
Tensors changed: MLP 280/392, lm_head 1/1, Embeddings 1/4, Attention 251/388, Norms 126/241