MI

cloudyu Mixtral 13Bx2 MOE 22B

MistralGGUF by RichardErkhovUpdated 2024-08-12
Download on Hugging Face
Parameters21.5B
Context4K
Layers40
Architecturellama
Smallest quant7.9 GB
Quick answer

cloudyu Mixtral 13Bx2 MOE 22B needs about 20.2 GB of memory at Q4_K_M (a 13.0 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 17 quants, from 7.9 GB (Q2_K) to 14.8 GB (Q5_K_S).

cloudyu Mixtral 13Bx2 MOE 22B is a 21.5B-parameter Mistral model in GGUF format, quantized by RichardErkhov. Its trained context length is 4K tokens. The repository has 17 quant files, from 7.9 GB (Q2_K) to 14.8 GB (Q5_K_S).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K7.9 GB 11.8 GB
Too big Get file →
IQ3_XS8.8 GB 12.7 GB
Too big Get file →
IQ3_S9.3 GB 13.2 GB
Too big Get file →
Q3_K_S9.3 GB 13.2 GB
Too big Get file →
IQ3_M9.7 GB 13.5 GB
Too big Get file →
Q3_K10.4 GB 14.2 GB
Too big Get file →
Q3_K_M10.4 GB 14.2 GB
Too big Get file →
Q3_K_L11.3 GB 15.2 GB
Too big Get file →
IQ4_XS11.6 GB 15.4 GB
Too big Get file →
Q4_012.1 GB 16.0 GB
Too big Get file →
IQ4_NL12.2 GB 16.1 GB
Too big Get file →
Q4_K_S12.2 GB 16.1 GB
Too big Get file →
Q4_K13.0 GB 16.9 GB
Too big Get file →
Q4_K_M13.0 GB 16.9 GB
Too big Get file →
Q4_113.5 GB 17.3 GB
Too big Get file →
Q5_014.8 GB 18.7 GB
Too big Get file →
Q5_K_S14.8 GB 18.7 GB
Too big Get file →

What GPU do you need for cloudyu Mixtral 13Bx2 MOE 22B?

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 IQ3_XS (8.8 GB file, about 16.0 GB in use).
  • 24 GB VRAM: the best quant that fits is Q5_0 (14.8 GB file, about 22.0 GB in use).

Which cloudyu Mixtral 13Bx2 MOE 22B quant should you choose?

Q4_K_M is the usual starting point for cloudyu Mixtral 13Bx2 MOE 22B: it balances file size and quality well. Here it is a 13.0 GB file.

If memory is tight, the smallest file is Q2_K at 7.9 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
Parameters21.5B (approximate, mixture of experts)
Layers40
Trained context4,096 tokens
KV cache (fp16)800 KB per token

Source

Quantized byRichardErkhov
FamilyMistral
Downloads (HF)99
Last updated2024-08-12

Frequently asked questions about cloudyu Mixtral 13Bx2 MOE 22B

How much VRAM does cloudyu Mixtral 13Bx2 MOE 22B need?

About 20.2 GB at Q4_K_M with an 8K context window. That is the 13.0 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 cloudyu Mixtral 13Bx2 MOE 22B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 15.1 GB, so you would need a larger GPU or CPU offloading.

What is the file size of cloudyu Mixtral 13Bx2 MOE 22B GGUF?

It depends on the quant: Q2_K is 7.9 GB, IQ3_XS is 8.8 GB, IQ3_S is 9.3 GB, Q3_K_S is 9.3 GB, IQ3_M is 9.7 GB, Q3_K is 10.4 GB and 11 more.

What context length does cloudyu Mixtral 13Bx2 MOE 22B support?

The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 6.7 GB at 8K and 3.4 GB at 4K.

Where can I download cloudyu Mixtral 13Bx2 MOE 22B GGUF?

The files are on Hugging Face in the RichardErkhov/cloudyu_-_Mixtral_13Bx2_MOE_22B-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 cloudyu Mixtral 13Bx2 MOE 22B 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 RichardErkhov

All by RichardErkhov →

More Mistral models

All Mistral →

Other medium models

All medium models →