HuggingFaceTB/SmolVLM-500M-Instruct
Full fine-tune (fewer layers) vs HuggingFaceTB/SmolLM2-360M-Instruct. HuggingFaceTB/SmolLM2-360M-Instruct matches part of HuggingFaceTB/SmolVLM-500M-Instruct (289 of its 489 tensors; the rest of HuggingFaceTB/SmolVLM-500M-Instruct has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Norms 65/65, MLP 96/96, Attention 128/128
9 derivatives checked
- HuggingFaceTB/SmolVLM2-500M-Video-Instruct: Full fine-tune
- ggml-org/SmolVLM-500M-Instruct-GGUF: Same weights, converted (fewer layers)
- YuankaiLuo/SimVLA-LIBERO: Partial fine-tune (fewer layers)
- mradermacher/SmolVLM-500M-Instruct-GGUF: Same weights, converted (fewer layers)
- mradermacher/SmolVLM-500M-Instruct-i1-GGUF: Quantization (fewer layers)
- ShayonSarker/SmolVLM-500M-Instruct-Q4_K_M-GGUF: Quantization (fewer layers)
- pierretokns/SmolVLM-500M-Instruct-GGUF: Quantization (fewer layers)
- sugiv/cardvaultplus-500m: LoRA-style
- vidore/ColSmolVLM-Instruct-500M-base: Same weights, repacked
2 other repos hold byte-for-byte identical weight files.