Hugging FaceIdefics3ForConditionalGeneration

SmolVLM 256M Instruct — Hardware Requirements & GPU Compatibility

Vision

SmolVLM-256M Instruct is Hugging Face's 256-million-parameter vision-language model, built on a compact SmolLM2-135M language backbone paired with a small 93-million-parameter SigLIP vision encoder. Hugging Face calls it the smallest publicly available multimodal model, designed for image captioning, visual question answering, and basic text transcription from images rather than open-ended chat. Its size makes it practical to run even on CPUs or entry-level GPUs, and on-device deployment is a stated use case. Context length is limited to 8,192 tokens. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in January 2025. Compared to the larger 2.2B SmolVLM2 sibling, it uses a smaller vision encoder and more aggressive image-token compression to further cut memory and latency, at some cost to accuracy.

256M params · dense8K contextApache 2.0Jan 2025135.3K downloads425 likes25.2K quant downloads

Based on SmolLM2 135M Instruct

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Runs great on a 24 GB GPU

Runs from 8 GB (Q4_K_M · 8K)Best on 24 GB (Q4_K_M · ~1310 tok/s · 8K)Macs from 8 GB (Q4_K_M · ~143 tok/s)

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run SmolVLM 256M Instruct

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Ways to runOther ways

each opens step-by-step

Run in cloud

Your own private copy · Q4_K_M · 24 GB GPU

The easiest way to try this model
Running in a few minutes, easy setup
Chat right here or connect any app

$5+ bonus for new users

Run SmolVLM 256M Instruct

0.6 GB VRAM at 8K context · fits the RTX 4090 (23.4 GB free)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/mradermacher/SmolVLM-256M-Instruct-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=8192 ollama serve.
$/set parameter num_ctx 8192
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/mradermacher/SmolVLM-256M-Instruct-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/mradermacher/SmolVLM-256M-Instruct-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 8192. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf mradermacher/SmolVLM-256M-Instruct-GGUF:Q4_K_M -c 8192 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/mradermacher/SmolVLM-256M-Instruct-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf mradermacher/SmolVLM-256M-Instruct-GGUF:Q4_K_M -c 8192 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve HuggingFaceTB/SmolVLM-256M-Instruct --max-model-len 8192
Runs the original weights, not the Q4_K_M GGUF; needs ~1 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model moot20/SmolVLM-256M-Instruct-MLX
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model moot20/SmolVLM-256M-Instruct-MLX

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$mradermacher/SmolVLM-256M-Instruct-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
RTX 3080 Ti (12 GB)CheapestCheapest11 GB free	~1186 tok/s	$0.029	Deploy on Vast.aifrom $0.12/hr (opens in new tab)	12 GB · 11 GB free · ~1186 tok/s · $0.029 / 1M tokens
RTX 3060 12GB11 GB free	~468 tok/s	$0.036	Deploy on Vast.aifrom $0.061/hr (opens in new tab)	11 GB free · ~468 tok/s · $0.036 / 1M tokens
RTX 4060 (8 GB)7 GB free	~354 tok/s	$0.067	Deploy on Vast.aifrom $0.086/hr (opens in new tab)	8 GB · 7 GB free · ~354 tok/s · $0.067 / 1M tokens
RTX A4000 (16 GB)15 GB free	~582 tok/s	$0.042	Deploy on Vast.aifrom $0.088/hr (opens in new tab)	16 GB · 15 GB free · ~582 tok/s · $0.042 / 1M tokens
RTX 4070 SUPER (12 GB)11 GB free	~655 tok/s	$0.044	Deploy on Vast.aifrom $0.10/hr (opens in new tab)	12 GB · 11 GB free · ~655 tok/s · $0.044 / 1M tokens
RTX PRO 4500 (32 GB)31 GB free	~1165 tok/s	$0.081	Deploy on RunPodfrom $0.34/hr (opens in new tab)	32 GB · 31 GB free · ~1165 tok/s · $0.081 / 1M tokens
RTX 4090 (24 GB)23 GB free	~1310 tok/s	$0.072	Deploy on RunPodfrom $0.34/hr (opens in new tab)	24 GB · 23 GB free · ~1310 tok/s · $0.072 / 1M tokens
A40 (48 GB)47 GB free	~905 tok/s	$0.11	Deploy on RunPodfrom $0.35/hr (opens in new tab)	48 GB · 47 GB free · ~905 tok/s · $0.11 / 1M tokens
L4 (24 GB)23 GB free	~390 tok/s	$0.31	Deploy on RunPodfrom $0.44/hr (opens in new tab)	24 GB · 23 GB free · ~390 tok/s · $0.31 / 1M tokens

Compare GPUs →

Your own private copy of SmolVLM 256M Instruct · Q4_K_M · OpenAI-compatible API

Open the llama.cpp listing on RunPod
Open on RunPodServerless (opens in new tab)
New to RunPod? Sign up and add credit first.
2
Fill in the form
Press Deploy and fill in:
Model
mradermacher/SmolVLM-256M-Instruct-GGUF
Quantization
Q4_K_M
Context Size
16384
After pasting the model press Esc — don’t pick a suggestion. Press Next.
Create the endpoint
Keep the default 24 GB and 48 GB, then Create Endpoint. Right away: Manage → Edit endpoint → Max workers 1 → Save (changing settings later restarts the worker).
The first start downloads the model, usually 5–15 min. After that, replies after a pause come in seconds.
Create an API key
RunPod → Settings → API Keys → Create. Pick “Restricted” with access to this endpoint only.

Use it

Copy the endpoint URL from its page on RunPod (any of its links works) and paste it with the key. Both stay in this browser.

Endpoint URL

API key

Chats and keys stay only in this browser.

https://api.runpod.ai/v2/<ENDPOINT_ID>/openai/v1

Put your endpoint ID in place of <ENDPOINT_ID>. The API key is the one from step 2.

from openai import OpenAI
client = OpenAI(base_url="https://api.runpod.ai/v2/<ENDPOINT_ID>/openai/v1", api_key="rpa_...")
r = client.chat.completions.create(
    model="any",
    messages=[{"role": "user", "content": "Hello!"}])
print(r.choices[0].message.content)

Works with any OpenAI client.

https://api.runpod.ai/v2/<ENDPOINT_ID>/openai/v1

Put your endpoint ID in place of <ENDPOINT_ID>. The API key is the one from step 2.

Free on Windows, macOS, Linux, iOS, Android and the web.

Open the web app ↗ (opens in a new tab)Download ↗ (opens in a new tab)

1. Settings → Model Provider → Add → Add Custom Provider
2. API Mode OpenAI API Compatible; paste the API key and the base URL as API Host
3. Model → Fetch, pick the model
4. New chat → pick the provider → start chatting

Free and open source for Windows, macOS and Linux.

Download Jan ↗ (opens in a new tab)

1.Settings → next to Model Providers press +
2.API format OpenAI-compatible; Base URL is the base URL above (with /v1); API key is your key; Create
3.If no model appears, click + in Models and type any
4.New chat → pick the model → start chatting

How Much VRAM Does SmolVLM 256M Instruct Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
0.4 GB	—	—	0.1 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Fits: 8K context · ~1560 tok/s
0.4 GB	—	—	0.1 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Fits: 8K context · ~1560 tok/s
0.4 GB	—	—	0.1 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Fits: 8K context · ~1524 tok/s
0.4 GB	—	—	0.1 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Fits: 8K context · ~1524 tok/s
0.5 GB	—	—	0.1 GB	2-bit · small K-quant — 2-bit small K-quant	Fits: 8K context · ~1456 tok/s
0.5 GB	—	—	0.1 GB	3-bit · i-quant — Importance-weighted 3-bit	Fits: 8K context · ~1456 tok/s
0.5 GB	—	—	0.1 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Fits: 8K context · ~1456 tok/s
0.5 GB	—	—	0.1 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Fits: 8K context · ~1424 tok/s
0.5 GB	—	—	0.1 GB	3-bit small — 3-bit small quantization	Fits: 8K context · ~1424 tok/s
0.5 GB	—	—	0.1 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Fits: 8K context · ~1424 tok/s
0.5 GB	—	—	0.1 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 8K context · ~1424 tok/s
0.5 GB	—	—	0.1 GB	3-bit medium — 3-bit medium quantization	Fits: 8K context · ~1394 tok/s
0.5 GB	—	—	0.1 GB	4-bit · legacy — 4-bit legacy quantization	Fits: 8K context · ~1365 tok/s
0.5 GB	—	—	0.1 GB	3-bit large — 3-bit large quantization	Fits: 8K context · ~1365 tok/s
0.5 GB	—	—	0.1 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Fits: 8K context · ~1337 tok/s
0.5 GB	—	—	0.1 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Fits: 8K context · ~1337 tok/s
0.5 GB	—	—	0.1 GB	4-bit small — 4-bit small quantization	Fits: 8K context · ~1337 tok/s
0.5 GB	—	—	0.1 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Fits: 8K context · ~1337 tok/s
0.5 GB	—	—	0.1 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 8K context · ~1310 tok/s
0.5 GB	—	—	0.1 GB	5-bit small — 5-bit small quantization	Fits: 8K context · ~1260 tok/s
0.5 GB	—	—	0.1 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 8K context · ~1236 tok/s
0.6 GB	—	—	0.2 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 8K context · ~1170 tok/s
0.6 GB	—	—	0.2 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 8K context · ~1092 tok/s
0.9 GB	—	—	~0.5 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Fits: 8K context · ~762 tok/s

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

0.6GBat 8K context

Weights: 0.2 GB
KV cache: 0.2 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 23.4 GB free · ~1310 tok/s

≥50 tok/s needs ≥38 GB/s · RTX 5090 ~2330, RTX 3090 Ti ~1310

Which Hardware Can Run SmolVLM 256M Instruct?

Q4_K_M · 0.5 GB

SmolVLM 256M Instruct (Q4_K_M) requires 0.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ GB is recommended. Using the full 8K context window can add up to 0.1 GB, bringing total usage to 0.6 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run SmolVLM 256M Instruct, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Q4_K_M · 0.6 GB · ~354 tok/s · 8K0.6~3548K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 0.6 GB · ~468 tok/s · 8K0.6~4688K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 0.6 GB · ~582 tok/s · 8K0.6~5828K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 0.6 GB · ~1310 tok/s · 8K0.6~13108K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 0.6 GB · ~168 tok/s · 8K0.6~1688KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 0.6 GB · ~764 tok/s · 8K0.6~7648KMacs →

Alternatives to SmolVLM 256M Instruct

Similar models

3tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
Gemma 3 270M ITGoogle	—	1.2 GB	~3640	same size
Functiongemma 270M ITGoogle	—	1.6 GB	~1111	same size
Tiny MixtralTitanML	—	0.5 GB	~2329	same size

Fine-tunes & variants(1)community builds of SmolVLM 256M Instruct

SmolVLM2 256M Video Instruct

Specifications

Publisher: Hugging Face
Parameters: 256M
Architecture: Idefics3ForConditionalGeneration
Context Length: 8,192 tokens
Vocabulary Size: 49,280
Release Date: 2025-01-17
License: Apache 2.0

Resources

Hugging Face model card (opens in a new tab)

Frequently Asked Questions

How much VRAM does SmolVLM 256M Instruct need?

SmolVLM 256M Instruct requires 0.5 GB of VRAM at Q4_K_M, or 0.9 GB at BF16.

VRAM = Weights + KV Cache + Overhead

Weights = 256M × 4.8 bits ÷ 8 = 0.2 GB

KV Cache + Overhead ≈ 0.3 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 0.4 GB (at full 8K context)

VRAM usage by quantization

Q4_K_M

0.5 GB

Q4_K_M + full context

0.6 GB

Learn more about VRAM estimation →

What's the best quantization for SmolVLM 256M Instruct?

For SmolVLM 256M Instruct, Q4_K_M (0.5 GB) offers the best balance of quality and VRAM usage. Q5_K_S (0.5 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 0.4 GB.

VRAM requirement by quantization

IQ2_XXS

0.4 GB~53%

IQ3_XS

0.5 GB~73%

Q4_0

0.5 GB~85%

IQ4_NL

0.5 GB~88%

Q4_K_M ★

0.5 GB~89%

BF16

0.9 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run SmolVLM 256M Instruct on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run SmolVLM 256M Instruct. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.5 GB of usable unified memory (RAM minus macOS overhead).

Can I run SmolVLM 256M Instruct locally?

Yes — SmolVLM 256M Instruct can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is SmolVLM 256M Instruct?

At Q4_K_M, SmolVLM 256M Instruct can reach ~9600 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~1310 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 0.5 × 0.65 = ~10400 tok/s

Estimated speed at Q4_K_M (0.5 GB)

NVIDIA B200

~10400 tok/s

NVIDIA GeForce RTX 4090

~1310 tok/s

NVIDIA B300

~10400 tok/s

AMD Instinct MI350X

~9600 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of SmolVLM 256M Instruct?

At Q4_K_M, the download is about 0.15 GB. The full-precision BF16 version is 0.51 GB. The smallest option (IQ2_XXS) is 0.07 GB.

Which GPUs can run SmolVLM 256M Instruct?

52 consumer GPUs can run SmolVLM 256M Instruct at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.

Which devices can run SmolVLM 256M Instruct?

163 devices with unified memory can run SmolVLM 256M Instruct at Q4_K_M (0.5 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.