What are the hardware requirements for Karasu Mixtral 8x22B V0.1 AWQ?

Karasu Mixtral 8x22B V0.1 AWQ requires approximately 73.7 GB of VRAM and supports a context window of 64K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Karasu Mixtral 8x22B V0.1 AWQ and how large is it?

Karasu Mixtral 8x22B V0.1 AWQ is developed by lightblue, a model with 140.6b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Karasu Mixtral 8x22B V0.1 AWQ?

Karasu Mixtral 8x22B V0.1 AWQ is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Karasu Mixtral 8x22B V0.1 AWQ by lightblue — VRAM 73.7GB, 64K context

Name: Karasu Mixtral 8x22B V0.1 AWQ
Author: lightblue

Karasu Mixtral 8x22B V0.1 AWQ is an open-source language model by lightblue. Features: 140.6b LLM, VRAM: 73.7GB, Context: 64K, MoE, Quantized, Merged, LLM Explorer Score: 0.12.

Merged Model 4-bit Awq Conversational Endpoints compatible Mixtral Moe Quantized Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/lightblue/Karasu-Mixtral-8x22B-v0.1-AWQ

Karasu Mixtral 8x22B V0.1 AWQ Parameters and Internals

Model Type

Causal Language Model

Additional Notes

Quantized using AutoAWQForCausalLM with 4-bit configuration, zero_point enabled, q_group_size of 128, and version GEMM.

LLM Name	Karasu Mixtral 8x22B V0.1 AWQ
Repository 🤗	https://huggingface.co/lightblue/Karasu-Mixtral-8x22B-v0.1-AWQ
Merged Model	Yes
Model Size	140.6b
Required VRAM	73.7 GB
Updated	2026-07-27
Maintainer	lightblue
Model Type	mixtral
Model Files	5.0 GB: 1-of-15 5.0 GB: 2-of-15 5.0 GB: 3-of-15 5.0 GB: 4-of-15 5.0 GB: 5-of-15 5.0 GB: 6-of-15 5.0 GB: 7-of-15 5.0 GB: 8-of-15 5.0 GB: 9-of-15 5.0 GB: 10-of-15 5.0 GB: 11-of-15 5.0 GB: 12-of-15 5.0 GB: 13-of-15 5.0 GB: 14-of-15 3.7 GB: 15-of-15
AWQ Quantization	Yes
Quantization Type	awq
Model Architecture	MixtralForCausalLM
Context Length	65536
Model Max Length	65536
Transformers Version	4.38.2
Tokenizer Class	LlamaTokenizer
Padding Token	</s>
Vocabulary Size	32000
Torch Data Type	float16

Rank the Karasu Mixtral 8x22B V0.1 AWQ capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Karasu Mixtral 8x22B V0.1 AWQ

Best Alternatives	Context / RAM	Downloads	Likes
...ixtral 8x22B Instruct V0.1 AWQ	64K / 73.7 GB	6689	13
...olphin 2.9.2 Mixtral 8x22b AWQ	64K / 73.7 GB	51	0
Mixtral 8x22B V0.1 AWQ	64K / 73.7 GB	358	36
WizardLM 2 8x22B AWQ	64K / 73.7 GB	75	13
Zephyr Orpo 141B A35b V0.1 AWQ	64K / 73.7 GB	4	3
Mixtral 8x22B Instruct V0.1	64K / 221.4 GB	71589	756
Zephyr Orpo 141B A35b V0.1	64K / 207.2 GB	66	270
WizardLM 2 8x22B	64K / 216.8 GB	9656	415
Mixtral 8x22B V0.1	64K / 221.6 GB	5069	239
...ixtral 8x22B Instruct V0.1 FP8	64K / 140.9 GB	11259	0

Note: green Score (e.g. "73.2") means that the model is better than lightblue/Karasu-Mixtral-8x22B-v0.1-AWQ.