numind/NuExtract3
Full fine-tune vs Qwen/Qwen3.5-4B. Fine-tune: 97% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 4/5, MLP 189/199, Attention 279/324, Norms 130/209, other 0/1
8 derivatives checked
- numind/NuExtract3-W8A8: Quantization
- numind/NuExtract3-GGUF: Quantization
- numind/NuExtract3-W4A16: Quantization
- prithivMLmods/NuExtract3-GGUF: Same weights, converted
- numind/NuExtract3-mlx-4bits: Quantization
- numind/NuExtract3-mlx-8bits: Quantization
- Vishva007/NuExtract3-W4A16-AutoRound-GPTQ: Quantization
- numind/NuExtract3-mlx-mxfp8: Quantization