SmolVLM 256M Instruct: memory fit on every Jetson module.
Method v1.0 · dataset 2026-09-07 · verified 2026-09-07 · last updated September 2026
0.26 B parameters (text 0.17 B + vision tower) · 30 layers · 8,192-token max context. Hugging Face: HuggingFaceTB/SmolVLM-256M-Instruct.
Model facts
| Fact | Value | Note |
|---|---|---|
| Parameters | 0.257 B | text 0.172 B + vision 0.085 B (estimated) |
| Layers | 30 | |
| Hidden size | 576 | |
| Attention heads | 9 | |
| KV heads | 3 | |
| Head dimension | 64 | |
| Max context | 8,192 tokens | |
| dtype | bfloat16 | |
| Vocabulary | 49,280 |
HuggingFaceTB · hf:HuggingFaceTB/SmolVLM-256M-Instruct · config.json · verified 2026-09-07
“"num_attention_heads": 9, "num_key_value_heads": 3, "head_dim": 64”
Parameters source
- Hugging Face · hf:HuggingFaceTB/SmolVLM-256M-Instruct · verified 2026-09-07 · model.safetensors (blob metadata via HF API) · class A
“model.safetensors size 513028808 bytes”
Vision tower
| Fact | Value |
|---|---|
| Layers | 12 |
| Hidden size | 768 |
| Patch size | 16 |
| Image size | 512 |
| Model type | idefics3 (SigLIP-based, 93M-param encoder per model card) |
Vision config source
- HuggingFaceTB · hf:HuggingFaceTB/SmolVLM-256M-Instruct · vision_config · verified 2026-09-07 · config.json vision_config · class A
“"hidden_size": 768, "num_hidden_layers": 12, "image_size": 512”
mmproj (vision weights): GGUF mmproj Q8_0 (vision), 0.104 GB.
mmproj source
- ggml-org (llama.cpp maintainers) · hf:ggml-org/SmolVLM-256M-Instruct-GGUF · verified 2026-09-07 · SmolVLM-256M-Instruct-GGUF repo file listing · class A
“mmproj-SmolVLM-256M-Instruct-Q8_0.gguf size 103769856 bytes”
Image tokens: 64 tokens per image (512² patch / 16 = 32² tokens, pixel-shuffle ×4 → 64 tokens per 512×512 tile; one tile assumed; class D).
Published quantised artefacts
A file the publisher or a community mirror actually ships, used as the class-A weight figure when the requested quant matches. Any quant without a row here falls back to parameters × bytes-per-parameter (class D for block quants).
| Quant | Format | File | Size | Source |
|---|---|---|---|---|
| INT8 / GGUF Q8_0 | GGUF Q8_0 (text) | SmolVLM-256M-Instruct-Q8_0.gguf | 0.175 GB | ggml-org (llama.cpp maintainers) · hf:ggml-org/SmolVLM-256M-Instruct-GGUF · verified 2026-09-07 |
Jetson tokens/s measurements
Published or archived throughput numbers, not modelled. Class C (external measured benchmark).
No Jetson-measured tokens/s reported for this model yet.
Every module × every quantisation
Verdict and total memory at 4096-token context, 1 sequence, llama.cpp, headless. Q4 cells link to the static breakdown page; every other cell links to the live engine at that quantisation.
Notes
No Q4_K_M published for this size; Q8_0 is the smallest quant in the ggml-org repo. Model card states 'run inference on one image with under 1GB of GPU RAM' (generic GPU, not Jetson-specific).
Method and data: Model Memory Fit methodology. Full registry: model-memory.json.
Change the context, concurrency or runtime.
The live engine covers any context length, sequence count, KV precision and runtime, with a permanent link.