SmolVLM2 2.2B Instruct — Hardware Requirements & GPU Compatibility
Vision
SmolVLM2-2.2B Instruct is Hugging Face's 2.2-billion-parameter vision-language model, the largest member of the SmolVLM2 family, built to process interleaved text, images, and video within a single conversation. Beyond captioning and visual question answering, it is trained on additional video-instruction data for tasks like summarizing clips or answering questions about video content. At this size it still runs on a single consumer GPU, useful for on-device, resource-constrained deployment. The model supports a context window of 8,192 tokens. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in February 2025 alongside smaller 500M and 256M SmolVLM2 siblings. Unlike those variants, which favor raw efficiency, the 2.2B model is tuned to be the most capable of the three on both image and video benchmarks.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 8K)Best on 24 GB (Q4_K_M · ~443 tok/s · 8K)Macs from 8 GB (Q4_K_M · ~48 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run SmolVLM2 2.2B 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.
Recommended
ollama run hf.co/ggml-org/SmolVLM2-2.2B-Instruct-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 21.7 GB free · 8K context
Download: SmolVLM2-2.2B-Instruct-Q4_K_M.gguf1.11 GBggml-org
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (9 GB headroom) · Q4_K_M
- Generation speed
- ~401 tok/s
- generation speed
- Cost per 1M output tokens
- $0.085
- per 1M output tokens
How Much VRAM Does SmolVLM2 2.2B Instruct Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 1.1 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 8K context · ~624 tok/s | |
| 1.1 GB | — | 3-bit small — 3-bit small quantization | Fits: 8K context · ~607 tok/s | |
| 1.2 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 8K context · ~546 tok/s | |
| 1.2 GB | — | 4-bit · legacy — 4-bit legacy quantization | Fits: 8K context · ~528 tok/s | |
| 1.5 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 8K context · ~443 tok/s | |
| 1.8 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 8K context · ~372 tok/s | |
| 2.0 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 8K context · ~321 tok/s | |
| 2.5 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 8K context · ~265 tok/s |
Selected Q4_K_M
4-bit medium quantization2.3GBat 8K context
- Weights
- 0.9 GB
- KV cache
- 1.0 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 21.8 GB free · ~443 tok/s
≥50 tok/s needs ≥114 GB/s · RTX 5090 ~787, RTX 3090 Ti ~443
Which Hardware Can Run SmolVLM2 2.2B Instruct?
Q4_K_M · 1.5 GBSmolVLM2 2.2B Instruct (Q4_K_M) requires 1.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 2+ GB is recommended. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run SmolVLM2 2.2B Instruct, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 5.8 GB free · up to 8K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~334Simulate output on RTX 3080RTX 2080 Ti~271Simulate output on RTX 2080 TiRTX 3070 Ti~267Simulate output on RTX 3070 TiM3*8 GB~48Simulate output on iMac 24" M3 (8 GB)Orin Nano~45Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~32Simulate output on iPhone 1720 GB
fits · 17.8 GB free · up to 8K ctx
1 GPU
RX 7900 XT~324Simulate output on RX 7900 XT32–40 GB
fits · 29.8 GB free · up to 8K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~787Simulate output on RTX 5090M5 Max36 GB~218Simulate output on Mac Studio M5 Max (36 GB)M532 GB~73Simulate output on MacBook Air 13" M5 (32 GB)Orin~90Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~59Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| LFM3 Omni 2BLiquid AI | — | 2.4 GB | ~385 | same size |
| Qwen3.5 2BAlibaba | — | 1.9 GB | ~370 | same size |
| Qwen3 VL 2B InstructAlibaba | — | 3.5 GB | ~362 | same size |
| Qwen2 VL 2B InstructAlibaba | — | 2.1 GB | ~390 | same size |
| Kimi K3 DSparkRadixArk | — | 2.2 GB | ~381 | same size |
| EXAONE 3.5 2.4B InstructLGAI-EXAONE | — | 3.4 GB | ~412 | same size |
Specifications
- Publisher
- Hugging Face
- Parameters
- 2.2B
- Architecture
- SmolVLMForConditionalGeneration
- Context Length
- 8,192 tokens
- Vocabulary Size
- 49,280
- Release Date
- 2025-02-08
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does SmolVLM2 2.2B Instruct need?
SmolVLM2 2.2B Instruct requires 1.5 GB of VRAM at Q4_K_M, or 4.9 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 2.2B × 4.8 bits ÷ 8 = 1.3 GB
KV Cache + Overhead ≈ 0.2 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.
VRAM usage by quantization
What's the best quantization for SmolVLM2 2.2B Instruct?
For SmolVLM2 2.2B Instruct, Q4_K_M (1.5 GB) offers the best balance of quality and VRAM usage. Q5_0 (1.5 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 0.7 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run SmolVLM2 2.2B Instruct on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run SmolVLM2 2.2B Instruct. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 1.5 GB of usable unified memory (RAM minus macOS overhead).
Can I run SmolVLM2 2.2B Instruct locally?
Yes — SmolVLM2 2.2B Instruct can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is SmolVLM2 2.2B Instruct?
At Q4_K_M, SmolVLM2 2.2B Instruct can reach ~3243 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~443 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 ÷ 1.48 × 0.65 = ~3514 tok/s
Estimated speed at Q4_K_M (1.5 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of SmolVLM2 2.2B Instruct?
At Q4_K_M, the download is about 1.35 GB. The full-precision BF16 version is 4.49 GB. The smallest option (IQ2_XXS) is 0.62 GB.
Which GPUs can run SmolVLM2 2.2B Instruct?
52 consumer GPUs can run SmolVLM2 2.2B 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 SmolVLM2 2.2B Instruct?
163 devices with unified memory can run SmolVLM2 2.2B Instruct at Q4_K_M (1.5 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.