MI

Karasu-Mixtral-8x22B-v0.1

MistralGGUF by mradermacherUpdated 2024-08-02
Download on Hugging Face
Parameters141B
Context64K
Layers56
Architecturellama
Smallest quant29.6 GB
Quick answer

Karasu-Mixtral-8x22B-v0.1 needs about 40.3 GB of memory at IQ2_XXS (a 37.9 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 6 quants, from 29.6 GB (IQ1_S) to 46.7 GB (IQ2_M). Longer context needs more: at 32K tokens the KV cache alone adds about 7.5 GB.

Karasu-Mixtral-8x22B-v0.1 – a 141B parameter multilingual chat model

Karasu-Mixtral-8x22B-v0.1 is a finetuned version of Mixtral-8x22B-v0.1. It was trained on a multilingual chat dataset and shows high accuracy and fast inference. The model supports 65536 token context and is released under apache-2.0.

Good at

  • Multilingual chat
  • High factual recall
  • Logical reasoning

Limits

  • Large VRAM requirement
  • No vision support
  • No explicit chat training on base model

Best for

  • Chatbots requiring large context
  • Research on multilingual dialogue
  • Fact‑checking applications

Who it’s for. Developers and researchers who need a large‑parameter LLM for conversational tasks and can provide sufficient GPU memory. The model is suitable for use with vLLM or llama.cpp.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S29.6 GB 32.0 GB
Too big Get file →
IQ1_M32.7 GB 35.1 GB
Too big Get file →
IQ2_XXS37.9 GB 40.3 GB
Too big Get file →
IQ2_XS42.0 GB 44.4 GB
Too big Get file →
IQ2_S42.6 GB 45.0 GB
Too big Get file →
IQ2_M46.7 GB 49.1 GB
Too big Get file →

For 32GB VRAM, IQ1_S (29.65GB) is the smallest quant. IQ1_M (32.74GB) fits 35.1GB memory at 8k context. For 40GB VRAM, IQ2_XXS (37.89GB) or IQ2_XS (42GB) are options. Memory estimates are approximate.

What GPU do you need for Karasu-Mixtral-8x22B-v0.1?

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: the best quant that fits is IQ1_S (29.6 GB file, about 32.0 GB in use).
  • 48 GB VRAM: the best quant that fits is IQ2_S (42.6 GB file, about 45.0 GB in use).

Which Karasu-Mixtral-8x22B-v0.1 quant should you choose?

IQ2_XXS (37.9 GB) is the middle option in this repository. 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.

If memory is tight, the smallest file is IQ1_S at 29.6 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
Parameters141B (approximate, mixture of experts)
Layers56
Trained context65,536 tokens
KV cache (fp16)224 KB per token

Source

Quantized bymradermacher
FamilyMistral
Licenseapache-2.0
Downloads (HF)691
Last updated2024-08-02

Frequently asked questions about Karasu-Mixtral-8x22B-v0.1

How much VRAM does Karasu-Mixtral-8x22B-v0.1 need?

About 40.3 GB at IQ2_XXS with an 8K context window. That is the 37.9 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 Karasu-Mixtral-8x22B-v0.1 on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 32.0 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Karasu-Mixtral-8x22B-v0.1 GGUF?

It depends on the quant: IQ1_S is 29.6 GB, IQ1_M is 32.7 GB, IQ2_XXS is 37.9 GB, IQ2_XS is 42.0 GB, IQ2_S is 42.6 GB, IQ2_M is 46.7 GB.

What context length does Karasu-Mixtral-8x22B-v0.1 support?

The GGUF header lists a trained context length of 65,536 tokens (64K). The KV cache grows with the context you use: about 1.9 GB at 8K and 7.5 GB at 32K.

Where can I download Karasu-Mixtral-8x22B-v0.1 GGUF?

The files are on Hugging Face in the mradermacher/Karasu-Mixtral-8x22B-v0.1-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 Karasu-Mixtral-8x22B-v0.1 use?

Hugging Face lists the license as apache-2.0. 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 Karasu-Mixtral-8x22B-v0.1 on 8GB VRAM?

No, the smallest quant requires 29.65GB of VRAM at 8k context.

What license does the model use?

The model is released under apache-2.0.

Does the model support vision tasks?

No, the model does not support vision.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Mistral models

All Mistral →

Other large models

All large models →