OpenDataArena/MMFineReason-2B

Full fine-tune vs Qwen/Qwen3-VL-2B-Instruct. Fine-tune: 100% of the weights changed, across 28 of 28 layers.

Tensors changed: Embeddings 4/4, Norms 166/217, MLP 182/196, Attention 208/208

2 derivatives checked

• mradermacher/MMFineReason-2B-i1-GGUF: Quantization (fewer layers)
• mradermacher/MMFineReason-2B-GGUF: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse