MeerDevelopment/Qevi-2B
Partial fine-tune vs Qwen/Qwen3-VL-2B-Instruct. Partial change: 85% of the weights changed (attention, embeddings, mlp, norms), in 28 of 28 layers.
Tensors changed: Embeddings 3/4, Norms 156/217, MLP 186/196, Attention 208/208
1 derivatives checked
- mradermacher/Qevi-2B-GGUF: Same weights, converted (fewer layers)