MI

Mixtral_7Bx5_MoE_30B

MistralGGUF by LoneStrikerUpdated 2024-02-13
Download on Hugging Face
Parameters29.8B
Context32K
Layers32
Architecturellama
Smallest quant12.9 GB
Quick answer

Mixtral_7Bx5_MoE_30B needs about 19.6 GB of memory at Q4_K_M (a 18.0 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 9 quants, from 12.9 GB (Q3_K_S) to 31.7 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

Mixtral_7Bx5_MoE_30B is a 29.8B-parameter Mistral model in GGUF format, quantized by LoneStriker. Its trained context length is 32K tokens. The repository has 9 quant files, from 12.9 GB (Q3_K_S) to 31.7 GB (Q8_0).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q3_K_S12.9 GB 14.4 GB
Too big Get file →
Q3_K_M14.3 GB 15.8 GB
Too big Get file →
Q3_K_L15.4 GB 17.0 GB
Too big Get file →
Q4_K_S16.9 GB 18.5 GB
Too big Get file →
Q4_K_M18.0 GB 19.6 GB
Too big Get file →
Q5_K_S20.5 GB 22.1 GB
Too big Get file →
Q5_K_M21.1 GB 22.7 GB
Too big Get file →
Q6_K24.4 GB 26.0 GB
Too big Get file →
Q8_031.7 GB 33.2 GB
Too big Get file →

What GPU do you need for Mixtral_7Bx5_MoE_30B?

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: the best quant that fits is Q3_K_M (14.3 GB file, about 15.8 GB in use).
  • 24 GB VRAM: the best quant that fits is Q5_K_M (21.1 GB file, about 22.7 GB in use).
  • 32 GB VRAM: the best quant that fits is Q6_K (24.4 GB file, about 26.0 GB in use).
  • 48 GB VRAM: the best quant that fits is Q8_0 (31.7 GB file, about 33.2 GB in use).

Which Mixtral_7Bx5_MoE_30B quant should you choose?

Q4_K_M is the usual starting point for Mixtral_7Bx5_MoE_30B: it balances file size and quality well. Here it is a 18.0 GB file.

The Q8_0 file is 31.7 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 Q3_K_S at 12.9 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters29.8B (approximate, mixture of experts)
Layers32
Trained context32,768 tokens
KV cache (fp16)128 KB per token

Source

Quantized byLoneStriker
FamilyMistral
Downloads (HF)64
Last updated2024-02-13

Frequently asked questions about Mixtral_7Bx5_MoE_30B

How much VRAM does Mixtral_7Bx5_MoE_30B need?

About 19.6 GB at Q4_K_M with an 8K context window. That is the 18.0 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 Mixtral_7Bx5_MoE_30B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 14.4 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Mixtral_7Bx5_MoE_30B GGUF?

It depends on the quant: Q3_K_S is 12.9 GB, Q3_K_M is 14.3 GB, Q3_K_L is 15.4 GB, Q4_K_S is 16.9 GB, Q4_K_M is 18.0 GB, Q5_K_S is 20.5 GB and 3 more.

What context length does Mixtral_7Bx5_MoE_30B support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.

Where can I download Mixtral_7Bx5_MoE_30B GGUF?

The files are on Hugging Face in the LoneStriker/Mixtral_7Bx5_MoE_30B-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 Mixtral_7Bx5_MoE_30B use?

Hugging Face lists the license as cc-by-nc-4.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Similar models

Browse all →

More from LoneStriker

All by LoneStriker →

More Mistral models

All Mistral →

Other medium models

All medium models →