logic65/Whittle-Reflex-25B-A3B
Partial fine-tune vs logic65/Whittle-Qwen-3.8-35B-A3B. Partial change: 5% of the weights changed (attention, embeddings, mlp), in 40 of 40 layers.
Tensors changed: lm_head 0/1, Embeddings 2/4, Attention 130/410, MLP 120/400, Norms 0/154, other 0/5
1 derivatives checked
- logic65/Whittle-Reflex-25B-A3B-GGUF: Not comparable yet