OX-PIXL/SpatialThinker-3B
Full fine-tune vs Qwen/Qwen2.5-VL-3B-Instruct. Fine-tune: 100% of the weights changed, across 36 of 36 layers.
Tensors changed: Embeddings 2/2, Norms 73/138, MLP 276/304, Attention 379/380
2 derivatives checked
- mradermacher/SpatialThinker-3B-i1-GGUF: Quantization (fewer layers)
- mradermacher/SpatialThinker-3B-GGUF: Same weights, converted (fewer layers)
1 other repos hold byte-for-byte identical weight files.