HuggingFaceTB/SmolVLM-256M-Instruct
Full fine-tune (fewer layers) vs HuggingFaceTB/SmolLM2-135M-Instruct. HuggingFaceTB/SmolLM2-135M-Instruct matches part of HuggingFaceTB/SmolVLM-256M-Instruct (271 of its 471 tensors; the rest of HuggingFaceTB/SmolVLM-256M-Instruct has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 30 of 30 layers.
Tensors changed: Norms 61/61, MLP 90/90, Attention 120/120
11 derivatives checked
- HuggingFaceTB/SmolVLM2-256M-Video-Instruct: Full fine-tune
- docling-project/SmolDocling-256M-preview: Full fine-tune
- ggml-org/SmolVLM-256M-Instruct-GGUF: Same weights, converted (fewer layers)
- mradermacher/SmolVLM-256M-Instruct-GGUF: Same weights, converted (fewer layers)
- Mungert/SmolDocling-256M-preview-GGUF: Full fine-tune (fewer layers)
- mradermacher/SmolVLM-256M-Instruct-i1-GGUF: Quantization (fewer layers)
- BIOMEDICA/BMC-smolvlm1-256M: Full fine-tune
- pierretokns/SmolVLM-256M-Instruct-GGUF: Quantization (fewer layers)
- moot20/SmolVLM-256M-Instruct-MLX: Same weights, converted
- shreydan/SmolVLM-256M-Detection: Partial fine-tune
- vidore/ColSmolVLM-Instruct-256M-base: Same weights, repacked
7 other repos hold byte-for-byte identical weight files.