Mapika/decider-2b-vision
Partial fine-tune vs Qwen/Qwen3.5-2B-Base. Partial change: 77% of the weights changed (attention, embeddings, mlp, norms), in 24 of 24 layers.
Tensors changed: Attention 227/264, Norms 88/177, Embeddings 3/4, MLP 155/172
2 derivatives checked
- mradermacher/decider-2b-vision-GGUF: Same weights, converted
- mindchain/decider-2b-vision-GGUF: Quantization
1 other repos hold byte-for-byte identical weight files.