starcoder2-3b needs about 2.6 GB of memory at Q4_K_M (a 1.8 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 12 quants, from 1.1 GB (Q2_K) to 3.2 GB (Q8_0).
StarCoder2‑3B: 3B‑parameter code generation model
StarCoder2‑3B is a 3.03B‑parameter transformer decoder trained on 17 programming languages. It uses grouped‑query attention and a 16,384‑token context window with a 4,096‑token sliding window. The model is not an instruction model and is best used for code completion and generation.
Good at
- Code completion
- Fill‑in‑the‑middle generation
- Supports 17 programming languages
Limits
- Not an instruction model
- Requires context for good results
- No vision support
Best for
- Developers needing code snippets
- Fine‑tuning on specific languages
- Automated code generation
Who it’s for. Developers and researchers who need a lightweight code generation model. It is suitable for environments with limited GPU memory and for projects that require a 3B‑parameter model with a large context window.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 1.1 GB | 1.9 GB | Fits | Get file → | |
| Q3_K_S | 1.3 GB | 2.1 GB | Fits | Get file → | |
| Q3_K_M | 1.5 GB | 2.3 GB | Fits | Get file → | |
| Q3_K_L | 1.7 GB | 2.4 GB | Fits | Get file → | |
| Q4_0 | 1.7 GB | 2.5 GB | Fits | Get file → | |
| Q4_K_S | 1.7 GB | 2.5 GB | Fits | Get file → | |
| Q4_K_M | 1.8 GB | 2.6 GB | Fits | Get file → | |
| Q5_0 | 2.1 GB | 2.8 GB | Fits | Get file → | |
| Q5_K_S | 2.1 GB | 2.8 GB | Fits | Get file → | |
| Q5_K_M | 2.2 GB | 2.9 GB | Fits | Get file → | |
| Q6_K | 2.5 GB | 3.2 GB | Fits | Get file → | |
| Q8_0 | 3.2 GB | 4.0 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K or Q4_K_S (≈2.5 GB memory at 8 k context). For 12 GB VRAM, Q5_K_M or Q5_0 (≈2.9 GB). For 16 GB VRAM, Q6_K (≈3.2 GB) or Q8_0 (≈4 GB). Estimates are approximate.
What GPU do you need for starcoder2-3b?
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 Q8_0 (3.2 GB file, about 4.0 GB in use).
Which starcoder2-3b quant should you choose?
Q4_K_M is the usual starting point for starcoder2-3b: it balances file size and quality well. Here it is a 1.8 GB file.
The Q8_0 file is 3.2 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.
If memory is tight, the smallest file is Q2_K at 1.1 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about starcoder2-3b
How much VRAM does starcoder2-3b need?
About 2.6 GB at Q4_K_M with an 8K context window. That is the 1.8 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 starcoder2-3b on an 8 GB GPU?
Yes, with Q8_0: the 3.2 GB file needs about 4.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of starcoder2-3b GGUF?
It depends on the quant: Q2_K is 1.1 GB, Q3_K_S is 1.3 GB, Q3_K_M is 1.5 GB, Q3_K_L is 1.7 GB, Q4_0 is 1.7 GB, Q4_K_S is 1.7 GB and 6 more.
What context length does starcoder2-3b support?
The GGUF header lists a trained context length of 16,384 tokens (16K). The KV cache grows with the context you use: about 0.3 GB at 8K and 0.5 GB at 16K.
Where can I download starcoder2-3b GGUF?
The files are on Hugging Face in the second-state/StarCoder2-3B-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 starcoder2-3b use?
Hugging Face lists the license as bigcode-openrail-m. 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 StarCoder2‑3B on 8 GB VRAM?
Yes, using Q4_K or Q4_K_S quantizations, which use about 2.5 GB of VRAM at 8 k context.
Does StarCoder2‑3B support natural language instructions?
No, it is not an instruction model; commands like "Write a function that computes the square root" do not work well.
What license applies to StarCoder2‑3B?
The model is released under the bigcode‑openrail‑m license.