DE

DeepSeek-Coder-V2-Instruct

DeepSeekGGUF by bullerwinsUpdated 2024-06-19
Download on Hugging Face
Parameters236B
Context160K
Layers60
Architecturedeepseek2
Smallest quant133.9 GB
Quick answer

DeepSeek-Coder-V2-Instruct needs about 208.0 GB of memory at Q5_K_M (a 167.2 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 5 quants, from 133.9 GB (Q4_K_S) to 250.6 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 161.1 GB.

DeepSeek-Coder-V2-Instruct, a 236B Mixture-of-Experts code model

DeepSeek-Coder-V2-Instruct is a 236B parameter model built on the deepseek2 architecture. It is a Mixture-of-Experts model trained on 163840 context tokens and focuses on coding tasks. The model is available in multiple GGUF quantizations.

Good at

  • Large context code generation
  • Multi‑language coding support
  • High‑quality code reasoning

Limits

  • Requires disabling Flash Attention in llama.cpp
  • Large VRAM needed for higher quantizations
  • No vision capabilities

Best for

  • Developers needing code completion
  • Code review assistants
  • Educational coding tools

Who it’s for. The model is suitable for developers, educators, and researchers who need a powerful open‑source code language model. It is ideal for those who can manage the memory requirements of a 236B Mixture‑of‑Experts architecture.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q4_K_S133.9 GB (7 parts) 174.6 GB
Too big Get file →
Q5_K_S162.3 GB (8 parts) 203.1 GB
Too big Get file →
Q5_K_M167.2 GB (8 parts) 208.0 GB
Too big Get file →
Q6_K193.5 GB (10 parts) 234.3 GB
Too big Get file →
Q8_0250.6 GB (12 parts) 291.4 GB
Too big Get file →

For 8 GB VRAM, use Q4_K_S (file size 133.88 GB, estimated memory 174.6 GB at 8 K context). For 12 GB VRAM, Q5_K_S (162.31 GB, 203.1 GB). For 16 GB VRAM, Q5_K_M (167.22 GB, 208 GB). Higher quants require more VRAM.

What GPU do you need for DeepSeek-Coder-V2-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: no quant fits at an 8K context.
  • 32 GB VRAM: no quant fits at an 8K context.
  • 48 GB VRAM: no quant fits at an 8K context.

Which DeepSeek-Coder-V2-Instruct quant should you choose?

Q5_K_M (167.2 GB) is the middle option in this repository. Closer to the original model than 4-bit, for a moderate increase in size. A good pick when you have memory to spare.

If memory is tight, the smallest file is Q4_K_S at 133.9 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturedeepseek2
Parameters236B (approximate, mixture of experts)
Layers60
Trained context163,840 tokens
KV cache (fp16)4800 KB per token

Source

Quantized bybullerwins
FamilyDeepSeek
Licenseother
Downloads (HF)899
Last updated2024-06-19

Frequently asked questions about DeepSeek-Coder-V2-Instruct

How much VRAM does DeepSeek-Coder-V2-Instruct need?

About 208.0 GB at Q5_K_M with an 8K context window. That is the 167.2 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-Coder-V2-Instruct on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 174.6 GB, so you would need a larger GPU or CPU offloading.

What is the file size of DeepSeek-Coder-V2-Instruct GGUF?

It depends on the quant: Q4_K_S is 133.9 GB, Q5_K_S is 162.3 GB, Q5_K_M is 167.2 GB, Q6_K is 193.5 GB, Q8_0 is 250.6 GB.

What context length does DeepSeek-Coder-V2-Instruct support?

The GGUF header lists a trained context length of 163,840 tokens (160K). The KV cache grows with the context you use: about 40.3 GB at 8K and 161.1 GB at 32K.

Where can I download DeepSeek-Coder-V2-Instruct GGUF?

The files are on Hugging Face in the bullerwins/DeepSeek-Coder-V2-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-Coder-V2-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-V2-Instruct on 8 GB VRAM?

Only the Q4_K_S quantization is feasible with 8 GB VRAM, but it still requires approximately 174.6 GB of memory at 8 K context, so it is not practical without additional optimizations.

Does this model support vision tasks?

No, the model is purely a code language model and has no vision capabilities.

Do I need to disable Flash Attention to use this model?

Yes, the GGUF quantized version may require disabling Flash Attention in llama.cpp for proper operation.

Similar models

Browse all →

More from bullerwins

All by bullerwins →

More DeepSeek models

All DeepSeek →

Other large models

All large models →