- Status
- Verified
- Base model
- …
- Lineage
- …
- Unchanged from base
- …
- Trending
- #12
- Downloads, 30 days
- 45.3k
- Weights
- 4.1 GB
- Revision
SmolVLM2-256M-Video is a lightweight multimodal model designed to analyze video content. The model processes videos, images, and text inputs to generate text outputs - whether answering questions about media files, comparing visual content, or transcribing text from images.
At a glance
- Task
- Vision language
- Input
- image, text
- Output
- text
- Parameters
- 256M
- Architecture
- Smolvlm
- Context
- 8K tokens
- Precision
- FP32
- Format
- Safetensors
- Library
- transformers
- License
- Apache 2.0Commercial use
- Languages
- en
- Base model
- Quantized from HuggingFaceTB/SmolVLM-256M-Instruct
- Released
- Feb 2025
- Updated
- Nov 2026
- Likes
- 178
- Downloads, all time
- 2,277,815
Architecture
- Layers
- 30
- Hidden size
- 576
- Attention
- 9 heads, grouped query, 3 KV heads
- Vocabulary
- 49,280
- Positions
- 8,192
- RoPE theta
- 100,000
- Tied embeddings
- No
- Vision encoder
- smolvlm_vision
Family
Models built on SmolVLM2-256M-Video-Instruct.
Run it
Loads with Transformers AutoModelForMultimodalLM and AutoProcessor, pinned to the indexed revision.
from transformers import AutoModelForMultimodalLM, AutoProcessor
model = AutoModelForMultimodalLM.from_pretrained("HuggingFaceTB/SmolVLM2-256M-Video-Instruct", revision="9cda5c165b28c68a52e9bb542d2b89f0f6963709")
processor = AutoProcessor.from_pretrained("HuggingFaceTB/SmolVLM2-256M-Video-Instruct", revision="9cda5c165b28c68a52e9bb542d2b89f0f6963709")Through the hub: the same tools, each file from a source that is up (Hugging Face, ModelScope, IPFS), at this revision. The last line checks every file against the hash the tree names.
export HF_ENDPOINT=https://gethologram.ai
hf download HuggingFaceTB/SmolVLM2-256M-Video-Instruct --quiet && hf cache verify HuggingFaceTB/SmolVLM2-256M-Video-InstructPapers
- SmolVLM: Redefining small and efficient multimodal modelsAndrés Marafioti et al., 2025
Datasets
Spaces
Used in 24 Spaces.
Read the full model card

SmolVLM2-256M-Video
SmolVLM2-256M-Video is a lightweight multimodal model designed to analyze video content. The model processes videos, images, and text inputs to generate text outputs - whether answering questions about media files, comparing visual content, or transcribing text from images. Despite its compact size, requiring only 1.38GB of GPU RAM for video inference. This efficiency makes it particularly well-suited for on-device applications that require specific domain fine-tuning and computational resources may be limited.
Model Summary
- Developed by: Hugging Face 🤗
- Model type: Multi-modal model (image/multi-image/video/text)
- Language(s) (NLP): English
- License: Apache 2.0
- Architecture: Based on Idefics3 (see technical summary)
Resources
- Demo: Video Highlight Generator
- Blog: Blog post
Uses
SmolVLM2 can be used for inference on multimodal (video / image / text) tasks where the input consists of text queries along with video or one or more images. Text and media files can be interleaved arbitrarily, enabling tasks like captioning, visual question answering, and storytelling based on visual content. The model does not support image or video generation.
To fine-tune SmolVLM2 on a specific task, you can follow the fine-tuning tutorial.
Evaluation
We evaluated the performance of the SmolVLM2 family on the following scientific benchmarks:
| Size | Video-MME | MLVU | MVBench |
|---|---|---|---|
| 2.2B | 52.1 | 55.2 | 46.27 |
| 500M | 42.2 | 47.3 | 39.73 |
| 256M | 33.7 | 40.6 | 32.7 |
How to get started
You can use transformers to load, infer and fine-tune SmolVLM. Make sure you have num2words, flash-attn and latest transformers installed. You can load the model as follows.
from transformers import AutoProcessor, AutoModelForImageTextToText
import torch
model_path = "HuggingFaceTB/SmolVLM2-256M-Video-Instruct"
processor = AutoProcessor.from_pretrained(model_path)
model = AutoModelForImageTextToText.from_pretrained(
model_path,
torch_dtype=torch.bfloat16,
_attn_implementation="flash_attention_2"
).to("cuda")
Simple Inference
You preprocess your inputs directly using chat templates and directly passing them
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
{"type": "text", "text": "Can you describe this image?"},
]
},
]
inputs = processor.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
).to(model.device, dtype=torch.bfloat16)
generated_ids = model.generate(**inputs, do_sample=False, max_new_tokens=64)
generated_texts = processor.batch_decode(
generated_ids,
skip_special_tokens=True,
)
print(generated_texts[0])
Video Inference
To use SmolVLM2 for video inference, make sure you have decord installed.
messages = [
{
"role": "user",
"content": [
{"type": "video", "path": "path_to_video.mp4"},
{"type": "text", "text": "Describe this video in detail"}
]
},
]
inputs = processor.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
).to(model.device, dtype=torch.bfloat16)
generated_ids = model.generate(**inputs, do_sample=False, max_new_tokens=64)
generated_texts = processor.batch_decode(
generated_ids,
skip_special_tokens=True,
)
print(generated_texts[0])
Multi-image Interleaved Inference
You can interleave multiple media with text using chat templates.
import torch
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "What is the similarity between these two images?"},
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg"},
]
},
]
inputs = processor.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
).to(model.device, dtype=torch.bfloat16)
generated_ids = model.generate(**inputs, do_sample=False, max_new_tokens=64)
generated_texts = processor.batch_decode(
generated_ids,
skip_special_tokens=True,
)
print(generated_texts[0])
Model optimizations
Misuse and Out-of-scope Use
SmolVLM is not intended for high-stakes scenarios or critical decision-making processes that affect an individual's well-being or livelihood. The model may produce content that appears factual but may not be accurate. Misuse includes, but is not limited to:
- Prohibited Uses:
- Evaluating or scoring individuals (e.g., in employment, education, credit)
- Critical automated decision-making
- Generating unreliable factual content
- Malicious Activities:
- Spam generation
- Disinformation campaigns
- Harassment or abuse
- Unauthorized surveillance
License
SmolVLM2 is built upon SigLIP as image encoder and SmolLM2 for text decoder part.
We release the SmolVLM2 checkpoints under the Apache 2.0 license.
Citation information
You can cite us in the following way:
@article{marafioti2025smolvlm,
title={SmolVLM: Redefining small and efficient multimodal models},
author={Andrés Marafioti and Orr Zohar and Miquel Farré and Merve Noyan and Elie Bakouch and Pedro Cuenca and Cyril Zakka and Loubna Ben Allal and Anton Lozhkov and Nouamane Tazi and Vaibhav Srivastav and Joshua Lochner and Hugo Larcher and Mathieu Morlon and Lewis Tunstall and Leandro von Werra and Thomas Wolf},
journal={arXiv preprint arXiv:2504.05299},
year={2025}
}
Training Data
SmolVLM2 used 3.3M samples for training originally from ten different datasets: LlaVa Onevision, M4-Instruct, Mammoth, LlaVa Video 178K, FineVideo, VideoStar, VRipt, Vista-400K, MovieChat and ShareGPT4Video. In the following plots we give a general overview of the samples across modalities and the source of those samples.
Data Split per modality
| Data Type | Percentage |
|---|---|
| Image | 34.4% |
| Text | 20.2% |
| Video | 33.0% |
| Multi-image | 12.3% |
Granular dataset slices per modality
Text Datasets
| Dataset | Percentage |
|---|---|
| llava-onevision/magpie_pro_ft3_80b_mt | 6.8% |
| llava-onevision/magpie_pro_ft3_80b_tt | 6.8% |
| llava-onevision/magpie_pro_qwen2_72b_tt | 5.8% |
| llava-onevision/mathqa | 0.9% |
Multi-image Datasets
| Dataset | Percentage |
|---|---|
| m4-instruct-data/m4_instruct_multiimage | 10.4% |
| mammoth/multiimage-cap6 | 1.9% |
Image Datasets
| Dataset | Percentage |
|---|---|
| llava-onevision/other | 17.4% |
| llava-onevision/vision_flan | 3.9% |
| llava-onevision/mavis_math_metagen | 2.6% |
| llava-onevision/mavis_math_rule_geo | 2.5% |
| llava-onevision/sharegpt4o | 1.7% |
| llava-onevision/sharegpt4v_coco | 1.5% |
| llava-onevision/image_textualization | 1.3% |
| llava-onevision/sharegpt4v_llava | 0.9% |
| llava-onevision/mapqa | 0.9% |
| llava-onevision/qa | 0.8% |
| llava-onevision/textocr | 0.8% |
Video Datasets
| Dataset | Percentage |
|---|---|
| llava-video-178k/1-2m | 7.3% |
| llava-video-178k/2-3m | 7.0% |
| other-video/combined | 5.7% |
| llava-video-178k/hound | 4.4% |
| llava-video-178k/0-30s | 2.4% |
| video-star/starb | 2.2% |
| vista-400k/combined | 2.2% |
| vript/long | 1.0% |
| ShareGPT4Video/all | 0.8% |
Derived on Nov 18, 2026 from Hugging Face at revision 9cda5c16, README.md , config.json .
38 files, 4.1 GB. Every download is checked against its address.