SmolLM3 3B needs about 3.0 GB of memory at Q4_K_M (a 1.9 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 1.9 GB (Q4_K_M) to 6.2 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.4 GB.
SmolLM3 3B – a 3.1B parameter open model with long context and hybrid reasoning
SmolLM3 3B is a 3.1 billion parameter decoder‑only transformer. It supports up to 128 k tokens with YARN extrapolation and is trained on 11.2 T tokens. The model is fully open under Apache‑2.0 and includes multilingual support for six languages.
Good at
- Hybrid reasoning with extended thinking
- Long‑context generation up to 128k tokens
- Multilingual support for six languages
Limits
- No vision or audio capabilities
- Requires 8 k context for memory estimates
- License is Apache‑2.0 – no commercial restrictions
Best for
- Chat and instruction following
- Long‑form content creation
- Multilingual text generation
Who it’s for. Developers and researchers needing a small, open language model with strong reasoning and long‑context capabilities. Suitable for local deployment with llama.cpp or other inference engines.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 1.9 GB | 3.0 GB | Fits | Get file → | |
| Q8_0 | 3.3 GB | 4.4 GB | Fits | Get file → | |
| F16 | 6.2 GB | 7.3 GB | Tight | Get file → |
For 8 GB VRAM, use Q4_K_M (1.92 GB file, ~3 GB VRAM at 8k context). For 12 GB VRAM, Q8_0 (3.28 GB file, ~4.4 GB VRAM). If higher precision is needed and memory allows, F16 (6.16 GB file, ~7.3 GB VRAM). Estimates are approximate.
What GPU do you need for SmolLM3 3B?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is Q4_K_M (1.9 GB file, about 3.0 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (3.3 GB file, about 4.4 GB in use).
- 8 GB VRAM: the best quant that fits is F16 (6.2 GB file, about 7.3 GB in use).
Which SmolLM3 3B quant should you choose?
Q4_K_M is the usual starting point for SmolLM3 3B: it balances file size and quality well. Here it is a 1.9 GB file.
The Q8_0 file is 3.3 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about SmolLM3 3B
How much VRAM does SmolLM3 3B need?
About 3.0 GB at Q4_K_M with an 8K context window. That is the 1.9 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run SmolLM3 3B on an 8 GB GPU?
Yes, with F16: the 6.2 GB file needs about 7.3 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of SmolLM3 3B GGUF?
It depends on the quant: Q4_K_M is 1.9 GB, Q8_0 is 3.3 GB, F16 is 6.2 GB.
What context length does SmolLM3 3B support?
The GGUF header lists a trained context length of 65,536 tokens (64K). The KV cache grows with the context you use: about 0.6 GB at 8K and 2.4 GB at 32K.
Where can I download SmolLM3 3B GGUF?
The files are on Hugging Face in the ggml-org/SmolLM3-3B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does SmolLM3 3B use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run SmolLM3 3B on 8 GB VRAM?
Yes, the Q4_K_M quantized file is 1.92 GB and requires about 3 GB VRAM at 8k context.
Does SmolLM3 support vision or audio?
No, the model only handles text input and output.
What context length does the model support?
It is trained on 64k tokens and can generate up to 128k tokens using YARN extrapolation.