The story so far
Weights, every dated article, and repo commits on one rail.
21 repo commits since the weights landed
Teal is the weights (2025-02-08), grey ticks are repo commits.
SmolVLM2-2.2B-Instruct
HuggingFaceTB Apache 2.0 out 20 months ago
2.25B parameters, images and text in, text out, 8K tokens of context.
Verified 2026-10-25 Rank #492 in LLMs · score 76.8 method v1 Sources 3 Downloads 30d 75.1k · #347 of 619 tracked
Do people use it?75.1kdownloads, 30 days#347 of 619 we track · below the median+0 likes, downloads −41% this week336 likesHugging Face2026-10-24
Where it's used
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 20 quantized builds and 100 Spaces.
- As a service0no API provider listed
- On own hardware20quantized builds
- Built on7746 fine-tunes · 31 adapters · 0 merges
- In apps100+Hugging Face Spaces
Top Spaces running it: HuggingFaceTB/SmolVLM2 ♥ 83 · HuggingFaceTB/SmolVLM2-HighlightGenerator ♥ 59 · jimenezcarrero/cookAIware ♥ 37 · HuggingFaceTB/SmolVLM2-XSPFGenerator ♥ 32
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 1.3 GB at INT4, 2.7 GB at FP8, 5.4 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
What people built on it
97 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
97 derivatives on the Hub, led by 46 fine-tunes
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
Benchmarks
18 rows
Benchmarks 18 rows
measured third-party leaderboard rowauthor-reported number from the model card or eval files (18)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported AI2D · Understand images and video | 70 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported ChartQA · Understand images and video | 68.84 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported DocVQA · Extract text from documents | 79.98 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported MLVU · Understand images and video | 55.2 | 2.2B | model author | 2026-10-07 |
| author-reported MLVU · Understand images and video | 47.3 | 500M | model author | 2026-10-07 |
| author-reported MLVU · Understand images and video | 40.6 | 256M | model author | 2026-10-07 |
| author-reported MMMU · Understand images and video | 42 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported MMStar · Understand images and video | 46 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported MVBench | 46.27 | 2.2B | model author | 2026-10-07 |
| author-reported MVBench | 39.73 | 500M | model author | 2026-10-07 |
| author-reported MVBench | 32.7 | 256M | model author | 2026-10-07 |
| author-reported MathVista · Understand images and video | 51.5 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported OCRBench · Extract text from documents | 72.9 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported Science_QA | 90 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported TextVQA · Understand images and video | 73.21 | SmolVLM2 2.2B | model author | 2026-10-07 |
| author-reported Video-MME · Understand images and video | 52.1 | 2.2B | model author | 2026-10-07 |
| author-reported Video-MME · Understand images and video | 42.2 | 500M | model author | 2026-10-07 |
| author-reported Video-MME · Understand images and video | 33.7 | 256M | model author | 2026-10-07 |
What the card says
1 cited sentence
Use-case evidence
4 use cases
What it is good at 4 use cases
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
- Understand images and videoauthor-reportedAuthor reports Mathvista 51.5Author reports MMMU 42Author reports MMStar 46+9 more
- Extract text from documentsauthor-reportedAuthor reports OCRBench 72.9Author reports DocVQA Val 79.98decision-model estimate p=0.78
- Chat and follow instructionsstructuraldecision-model estimate p=0.73name: Instructtag: conversational+1 more
- Run on a laptop or phonestructuraldecision-model estimate p=1.00fits: single-gpu
Facts
22 fields
Listed in LLMs
Facts
- SmolVLM2-2.2B-Instruct has 2.25B parameters. (safetensors metadata, observed 2026-10-24)
- Context length is 8,192 tokens (8K). (config.json max_position_embeddings, observed 2026-10-24)
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
- Created 2025-02-08, last modified 2025-04-08. (HF API, observed 2026-10-24)
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | SmolVLMForConditionalGeneration |
| Model type | smolvlm |
| Dtype | float32 |
| Params by dtype | F32: 2,246,784,880 |
| Size band | M |
| MoE | no |
| Context length | 8,192 (config.json max_position_embeddings) |
| Languages | en |
| Papers | arXiv:2504.05299 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · SmolVLMForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Hardware fit
3 estimates
Hardware fit ESTIMATE
Weights-only memory for SmolVLM2-2.2B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 5.4 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 2.7 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 1.3 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
40 variants
Lineage
Base model: HuggingFaceTB/SmolVLM-Instruct
quantized: 20finetune: 46adapter: 31
Top variants (40 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| ggml-org/SmolVLM2-2.2B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | ggml-org | 42 | 236k |
| second-state/SmolVLM2-2.2B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | second-state | 3 | 16.1k |
| mradermacher/SmolVLM2-2.2B-Instruct-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 2 | 10.6k |
| mradermacher/SmolVLM2-2.2B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 4 | 7.2k |
| mlx-community/SmolVLM2-2.2B-Instruct-mlx | finetune | mlx | 2.25B | M | apache-2.0 | mlx-community | 10 | 5.6k |
| gaianet/SmolVLM2-2.2B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | gaianet | 0 | 2.2k |
| richardyoung/SmolVLM2-2.2B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | richardyoung | 0 | 1.2k |
| litert-community/SmolVLM2-2.2B | quantized | — | — | — | apache-2.0 | litert-community | 1 | 957 |
| pierretokns/SmolVLM2-2.2B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | pierretokns | 0 | 960 |
| lewiswatson/Frame2KG-SmolVLM-2.2b-JSON | quantized | GGUF | — | — | — | lewiswatson | 0 | 917 |
| echarlaix/SmolVLM2-2.2B-Instruct-openvino | finetune | — | — | — | apache-2.0 | echarlaix | 0 | 557 |
| merve/SmolVLM2-2.2B-DocVQA | finetune | — | 2.25B | M | — | merve | 3 | 323 |
| Luigi/edge-fall-vlm-2.2b | quantized | GGUF | 2.25B | M | apache-2.0 | Luigi | 0 | 369 |
| huggingface-mic/smolvlm-mic | finetune | — | 6.58M | S | — | huggingface-mic | 0 | 341 |
| random654367gmailcom/SmolVLM2-2.2B-Instruct-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | random654367gmailcom | 0 | 331 |
| kelnei/slugvision-2.2b | quantized | GGUF | 2.25B | M | gemma | kelnei | 0 | 309 |
| davanstrien/SmolVLM2-2.2B-Instruct | finetune | — | 2.25B | M | — | davanstrien | 0 | 281 |
| anuragpradhan/SmolVLM2-2.2B-Instruct-bnb-4bit | quantized | -bnb | 422M | S | apache-2.0 | anuragpradhan | 0 | 277 |
| Chopinab/SmolVLM2-2.2B-Instruct-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | Chopinab | 0 | 209 |
| NANI-Nithin/SmolVLM-Hallucination-Defense-Merged | finetune | — | 2.25B | M | apache-2.0 | NANI-Nithin | 2 | 107 |
| introvoyz041/SmolVLM2-2.2B-Instruct-Q4_0-GGUF | quantized | Q4_0 | — | — | apache-2.0 | introvoyz041 | 0 | 199 |
| Chopinab/SmolVLM2-2.2B-Instruct-IQ3_M-GGUF | quantized | GGUF | — | — | apache-2.0 | Chopinab | 0 | 195 |
| cedpsam/SmolVLM2-2.2B-Instruct-ao-autoquant | quantized | — | — | — | apache-2.0 | cedpsam | 0 | 174 |
| Chopinab/SmolVLM2-2.2B-Instruct-Q5_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | Chopinab | 0 | 157 |
| Chopinab/SmolVLM2-2.2B-Instruct-Q8_0-GGUF | quantized | Q8_0 | — | — | apache-2.0 | Chopinab | 0 | 147 |
| darjja/smolvlm2-food-finetuned | finetune | — | 2.25B | M | apache-2.0 | darjja | 0 | 118 |
| NANI-Nithin/SmolVLM-Hallucination-Defense | adapter | — | — | — | apache-2.0 | NANI-Nithin | 1 | 62 |
| lewiswatson/Frame2KG-SmolVLM-2.2b-CT | quantized | GGUF | — | — | — | lewiswatson | 0 | 99 |
| NLP-Orange-Problem/AttentionSeekers | adapter | — | — | — | apache-2.0 | NLP-Orange-Problem | 1 | 48 |
| EZCon/SmolVLM2-2.2B-Instruct-mlx | finetune | mlx | 2.25B | M | apache-2.0 | EZCon | 0 | 74 |
| robot-learning-group47/eval3_lora_fixed_Chengming | adapter | — | — | — | — | robot-learning-group47 | 0 | 73 |
| MRIII0917/SmolVLM2-2.2B-Instruct-video-feedback | finetune | — | 2.25B | M | apache-2.0 | MRIII0917 | 0 | 64 |
| robot-learning-group47/eval3_phase1_celeb | adapter | — | — | — | — | robot-learning-group47 | 0 | 63 |
| robot-learning-group47/eval3_yann_left_repair_adapter | adapter | — | 2.25B | M | — | robot-learning-group47 | 0 | 61 |
| Firemedic15/smolvlm2-2.2b-instruct-sft-jobs | finetune | — | 2.25B | M | — | Firemedic15 | 0 | 60 |
| JackArr/okvqa_smolvlm_k_6m_mixedbread_run3 | adapter | — | — | — | — | JackArr | 0 | 59 |
| b3slim/SmolVLM2-2.2B-Instruct-video-feedback | adapter | — | 2.25B | M | apache-2.0 | b3slim | 0 | 54 |
| JackArr/okvqa_smolvlm_k_6m_mixedbread_run2 | adapter | — | — | — | — | JackArr | 0 | 51 |
| lukesutor/SmolVLM2-2.2B-ActivityTracking | finetune | — | 2.25B | M | — | lukesutor | 0 | 46 |
| robot-learning-group47/eval3_phase2_TOY | adapter | — | — | — | — | robot-learning-group47 | 0 | 45 |
Repository changelog
21 commits
Repository changelog 21 commits
| Date | Commit | SHA |
|---|---|---|
| 2025-04-08 | Update README.md | 482adb5 |
| 2025-03-06 | Update README.md | 274fe49 |
| 2025-03-03 | Update README.md | 42a4cc9 |
| 2025-03-03 | Update README.md | 83e8cba |
| 2025-03-03 | Update README.md | f83c1bb |
| 2025-02-27 | Test video tag (#13) | e00155d |
| 2025-02-27 | update tag (#11) | a9e1eea |
| 2025-02-26 | Code snippets (#10) | 8af4ecd |
| 2025-02-23 | Fix code typo for Simple and Video Inference examples (#9) | e92909e |
| 2025-02-21 | Fix dtype processing in README example (#5) | 41348c8 |
| 2025-02-21 | SmolVLM2-500M-Video -> SmolVLM2-2.2B (#3) | 54cb403 |
| 2025-02-20 | Update README.md | 06ac4f8 |
| 2025-02-18 | Fix chat template (#1) | 5719134 |
| 2025-02-17 | Update README.md | 520a778 |
| 2025-02-17 | Update README.md | 920ac89 |
| 2025-02-13 | Upload processor | 8d39814 |
| 2025-02-12 | Upload 3 files | 70402a4 |
| 2025-02-10 | Upload SmolVLMForConditionalGeneration | 67b63ec |
| 2025-02-08 | Upload processor | 8e96b17 |
| 2025-02-08 | Upload SmolVLMForConditionalGeneration | f33043d |
| 2025-02-08 | initial commit | 9fbd096 |
Papers
1 paper
Papers 1
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Similar models
3 models
Similar models
- Qwen3.5-2B · 2.27B same size band (M), same task (image-text-to-text), more providers, permissive license
- Qwen3-VL-2B-Instruct · 2.13B same size band (M), same task (image-text-to-text), more providers, permissive license
- Qwen2-VL-2B-Instruct · 2.21B same size band (M), same task (image-text-to-text), permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:40 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:11 UTC · Last verified 2026-10-25
Follow SmolVLM2-2.2B-Instruct
Watch this model
- /feeds/models/HuggingFaceTB--SmolVLM2-2.2B-Instruct.xml(RSS 2.0) · every change to SmolVLM2-2.2B-Instruct
- /feeds/families/HuggingFaceTB--SmolVLM2.xml(RSS 2.0) · every change across the SmolVLM2 family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.