mlfoundations/Gelato-30B-A3B
Full fine-tune vs Qwen/Qwen3-VL-30B-A3B-Instruct. Fine-tune: 96% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 3/4, Norms 160/309, MLP 246/268, Attention 300/300, lm_head 0/1
3 derivatives checked
- prithivMLmods/Gelato-30B-A3B-f32-AIO-GGUF: Same weights, converted
- mradermacher/Gelato-30B-A3B-i1-GGUF: Quantization
- mradermacher/Gelato-30B-A3B-GGUF: Quantization