allenai/SAGE-MM-Qwen3-VL-4B-SFT
Full fine-tune vs Qwen/Qwen3-VL-4B-Instruct. Fine-tune: 91% of the weights changed, across 36 of 36 layers.
Tensors changed: Embeddings 1/4, Norms 128/249, MLP 108/220, Attention 144/240
4 derivatives checked
- prithivMLmods/SAGE-MM-Qwen3-VL-4B-SFT-GGUF: Same weights, converted (fewer layers)
- mradermacher/SAGE-MM-Qwen3-VL-4B-SFT-i1-GGUF: Quantization (fewer layers)
- mradermacher/SAGE-MM-Qwen3-VL-4B-SFT-GGUF: Same weights, converted (fewer layers)
- allenai/SAGE-MM-Qwen3-VL-4B-SFT_RL: Partial fine-tune