DE

deepseek-ai_deepseek-coder-33b-instruct

DeepSeekGGUF by TheBlokeUpdated 2023-11-05
Download on Hugging Face
Parameters33.3B
Context16K
Layers62
Architecturellama
Smallest quant14.0 GB
Quick answer

deepseek-ai_deepseek-coder-33b-instruct needs about 22.5 GB of memory at Q4_K_M (a 19.9 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 12 quants, from 14.0 GB (Q2_K) to 35.4 GB (Q8_0).

DeepSeek Coder 33B Instruct – large‑scale code model in GGUF format

DeepSeek Coder 33B Instruct is a 33‑billion‑parameter model fine‑tuned on 2 B instruction tokens. It supports a 16 K context window and is optimized for code generation and completion. The model is available in multiple GGUF quantisations for efficient local inference.

Good at

  • Project‑level code completion
  • Multi‑language code generation
  • Fill‑in‑the‑blank code tasks

Limits

  • Requires large VRAM for full precision
  • No vision or multimodal support
  • License is other – check model license

Best for

  • Developers needing large‑scale code assistance
  • Code review automation
  • Complex code generation tasks

Who it’s for. Engineers and researchers who need a powerful code model for local inference. The model is suitable for users with sufficient GPU memory and who can manage the quantised GGUF files.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K14.0 GB 16.6 GB
Too big Get file →
Q3_K_S14.4 GB 17.0 GB
Too big Get file →
Q3_K_M16.1 GB 18.7 GB
Too big Get file →
Q3_K_L17.6 GB 20.1 GB
Too big Get file →
Q4_018.8 GB 21.4 GB
Too big Get file →
Q4_K_S18.9 GB 21.5 GB
Too big Get file →
Q4_K_M19.9 GB 22.5 GB
Too big Get file →
Q5_023.0 GB 25.5 GB
Too big Get file →
Q5_K_S23.0 GB 25.5 GB
Too big Get file →
Q5_K_M23.5 GB 26.1 GB
Too big Get file →
Q6_K27.4 GB 29.9 GB
Too big Get file →
Q8_035.4 GB 38.0 GB
Too big Get file →

For 8 GB VRAM, use Q4_K_M (22.5 GB memory at 8 K context) or Q4_K_S (21.5 GB). For 12 GB VRAM, Q4_0 (21.4 GB) or Q4_K_M (22.5 GB) are suitable. Higher memory budgets allow Q5_K_M or Q6_K for better quality.

What GPU do you need for deepseek-ai_deepseek-coder-33b-instruct?

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: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: the best quant that fits is Q4_K_M (19.9 GB file, about 22.5 GB in use).
  • 32 GB VRAM: the best quant that fits is Q6_K (27.4 GB file, about 29.9 GB in use).
  • 48 GB VRAM: the best quant that fits is Q8_0 (35.4 GB file, about 38.0 GB in use).

Which deepseek-ai_deepseek-coder-33b-instruct quant should you choose?

Q4_K_M is the usual starting point for deepseek-ai_deepseek-coder-33b-instruct: it balances file size and quality well. Here it is a 19.9 GB file.

The Q8_0 file is 35.4 GB, about 1.8 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 14.0 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

Architecturellama
Parameters33.3B
Layers62
Trained context16,384 tokens
KV cache (fp16)248 KB per token

Source

Quantized byTheBloke
FamilyDeepSeek
Licenseother
Downloads (HF)16,307
Last updated2023-11-05

Frequently asked questions about deepseek-ai_deepseek-coder-33b-instruct

How much VRAM does deepseek-ai_deepseek-coder-33b-instruct need?

About 22.5 GB at Q4_K_M with an 8K context window. That is the 19.9 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 deepseek-ai_deepseek-coder-33b-instruct on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 16.6 GB, so you would need a larger GPU or CPU offloading.

What is the file size of deepseek-ai_deepseek-coder-33b-instruct GGUF?

It depends on the quant: Q2_K is 14.0 GB, Q3_K_S is 14.4 GB, Q3_K_M is 16.1 GB, Q3_K_L is 17.6 GB, Q4_0 is 18.8 GB, Q4_K_S is 18.9 GB and 6 more.

What context length does deepseek-ai_deepseek-coder-33b-instruct support?

The GGUF header lists a trained context length of 16,384 tokens (16K). The KV cache grows with the context you use: about 2.1 GB at 8K and 4.2 GB at 16K.

Where can I download deepseek-ai_deepseek-coder-33b-instruct GGUF?

The files are on Hugging Face in the TheBloke/deepseek-coder-33B-instruct-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 deepseek-ai_deepseek-coder-33b-instruct use?

Hugging Face lists the license as other. 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 DeepSeek Coder 33B Instruct on 8 GB VRAM?

Use a Q4_K_M or Q4_K_S quantised GGUF; memory usage is around 21–22 GB at 8 K context, so 8 GB VRAM is insufficient.

Does the model support Chinese code?

Yes, the training data includes 13 % Chinese, so it can generate code in Chinese.

Is the model free for commercial use?

The model license allows commercial use, but check the Model License for details.

Other GGUFs of deepseek-coder-33b-instruct

Similar models

Browse all →

More from TheBloke

All by TheBloke →

More DeepSeek models

All DeepSeek →

Other medium models

All medium models →