Logics-MLLM/Logics-PPT-Qwen3.6-35B-A3B-SFT
Partial fine-tune (fewer layers) vs Qwen/Qwen3.6-35B-A3B. Qwen/Qwen3.6-35B-A3B matches part of Logics-MLLM/Logics-PPT-Qwen3.6-35B-A3B-SFT (946 of its 31666 tensors; the rest of Logics-MLLM/Logics-PPT-Qwen3.6-35B-A3B-SFT has no counterpart). On the shared part: Partial change: 82% of the weights changed (attention, embeddings, lm_head, mlp, norms), in 40 of 40 layers.
Tensors changed: Embeddings 2/4, Attention 364/388, MLP 297/312, Norms 172/241, lm_head 1/1
1 derivatives checked
- Logics-MLLM/Logics-PPT-Qwen3.6-35B-A3B-RL: Partial fine-tune