ST

Starcoder

StarCoderGGUF by mradermacherUpdated 2024-09-08
Download on Hugging Face
Parameters15.8B
Context8K
Layers40
Architecturestarcoder
Smallest quant3.7 GB
Quick answer

Starcoder needs about 10.8 GB of memory at Q4_K_M (a 10.1 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 24 quants, from 3.7 GB (IQ1_S) to 13.1 GB (Q6_K).

15.8B Starcoder model optimized for code generation

Starcoder is a 15.8B parameter model from the StarCoder family, trained on 8192 context tokens. It is available in multiple GGUF quantizations, each with different file sizes and memory footprints. The model is licensed under bigcode-openrail-m.

Good at

  • Code generation
  • Programming assistance
  • Language‑specific code snippets

Limits

  • No vision support
  • Large context requires significant VRAM
  • License restricts commercial use

Best for

  • Writing boilerplate code
  • Debugging assistance
  • Learning new APIs

Who it’s for. Developers and researchers needing a code‑centric LLM. The model is suitable for local deployment via llama.cpp with a suitable quantization.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S3.7 GB 4.3 GB
Fits Get file →
IQ1_M4.0 GB 4.6 GB
Fits Get file →
IQ2_XXS4.5 GB 5.2 GB
Fits Get file →
IQ2_XS4.9 GB 5.6 GB
Fits Get file →
IQ2_S5.3 GB 5.9 GB
Fits Get file →
IQ2_M5.7 GB 6.3 GB
Fits Get file →
Q2_K6.3 GB 7.0 GB
Fits Get file →
IQ3_XXS6.4 GB 7.1 GB
Fits Get file →
IQ3_XS6.9 GB 7.6 GB
Tight Get file →
IQ3_S7.1 GB 7.8 GB
Tight Get file →
Q3_K_S7.1 GB 7.8 GB
Tight Get file →
IQ3_M7.6 GB 8.3 GB
Too big Get file →
Q3_K_M8.4 GB 9.0 GB
Too big Get file →
IQ4_XS8.7 GB 9.3 GB
Too big Get file →
Q4_09.2 GB 9.8 GB
Too big Get file →
Q4_09.2 GB 9.8 GB
Too big Get file →
Q4_09.2 GB 9.8 GB
Too big Get file →
Q4_09.2 GB 9.9 GB
Too big Get file →
Q4_K_S9.3 GB 9.9 GB
Too big Get file →
Q3_K_L9.3 GB 9.9 GB
Too big Get file →
Q4_K_M10.1 GB 10.8 GB
Too big Get file →
Q5_K_S11.1 GB 11.8 GB
Too big Get file →
Q5_K_M11.7 GB 12.4 GB
Too big Get file →
Q6_K13.1 GB 13.8 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (4.3 GB memory) or IQ1_M (4.6 GB). If 12 GB is available, IQ2_S (5.9 GB) or Q3_K_M (9 GB) can be considered. Larger models like Q5_K_M need 12.4 GB memory and 11.7 GB file size.

What GPU do you need for Starcoder?

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: no quant fits at an 8K context.
  • 6 GB VRAM: the best quant that fits is IQ2_S (5.3 GB file, about 5.9 GB in use).
  • 8 GB VRAM: the best quant that fits is IQ3_S (7.1 GB file, about 7.8 GB in use).
  • 12 GB VRAM: the best quant that fits is Q5_K_S (11.1 GB file, about 11.8 GB in use).
  • 16 GB VRAM: the best quant that fits is Q6_K (13.1 GB file, about 13.8 GB in use).

Which Starcoder quant should you choose?

Q4_K_M is the usual starting point for Starcoder: it balances file size and quality well. Here it is a 10.1 GB file.

If memory is tight, the smallest file is IQ1_S at 3.7 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

Architecturestarcoder
Parameters15.8B
Layers40
Trained context8,192 tokens
KV cache (fp16)20 KB per token

Source

Quantized bymradermacher
FamilyStarCoder
Downloads (HF)3,555
Last updated2024-09-08

Frequently asked questions about Starcoder

How much VRAM does Starcoder need?

About 10.8 GB at Q4_K_M with an 8K context window. That is the 10.1 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 Starcoder on an 8 GB GPU?

Yes, with IQ3_S: the 7.1 GB file needs about 7.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Starcoder GGUF?

It depends on the quant: IQ1_S is 3.7 GB, IQ1_M is 4.0 GB, IQ2_XXS is 4.5 GB, IQ2_XS is 4.9 GB, IQ2_S is 5.3 GB, IQ2_M is 5.7 GB and 18 more.

What context length does Starcoder support?

The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.2 GB at 8K.

Where can I download Starcoder GGUF?

The files are on Hugging Face in the mradermacher/starcoder-i1-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 Starcoder 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 Starcoder on 8 GB VRAM?

Yes, with IQ1_S or IQ1_M quantizations, which use about 4.3–4.6 GB memory at 8 k context.

Does Starcoder support image input?

No, the model has no vision capabilities.

Is Starcoder free to use?

It is licensed under bigcode-openrail-m, which allows non‑commercial use but restricts commercial deployment.

Other GGUFs of starcoder

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More StarCoder models

All StarCoder →

Other medium models

All medium models →