docling-project/SmolDocling-256M-preview
Full fine-tune vs HuggingFaceTB/SmolVLM-256M-Instruct. Fine-tune: 100% of the weights changed, across 30 of 30 layers.
Tensors changed: lm_head 1/1, other 1/1, Embeddings 4/4, Norms 88/111, MLP 138/138, Attention 216/216
5 derivatives checked
- docling-project/SmolDocling-256M-preview-mlx-bf16: Same weights, repacked
- mradermacher/SmolDocling-256M-preview-i1-GGUF: Quantization (fewer layers)
- mradermacher/SmolDocling-256M-preview-GGUF: Same weights, converted (fewer layers)
- docling-project/ChemicalOCR: Partial fine-tune
- cstr/smoldocling-GGUF: Not comparable yet
3 other repos hold byte-for-byte identical weight files.