Qwen_Qwen3-8B-GGUF Open-Source Text Generation Model - Free for Various Text Generation Tasks

Qwen Qwen3 8B GGUF

Developed by bartowski

Quantized version of Qwen3-8B, quantized using the imatrix option of llama.cpp, suitable for text generation tasks.

Large Language Model Open Source License:Apache-2.0 #Multi-level quantization selection #Low memory usage #Localized deployment

Downloads 23.88k

Release Time : 4/28/2025

Model Overview

Based on the quantized version of Qwen/Qwen3-8B, supports multiple quantization types, suitable for text generation tasks, and can run in LM Studio or llama.cpp.

Model Features

Multiple quantization types

Provides multiple quantization versions from BF16 to Q2_K to meet different hardware and performance requirements.

High-quality recommendation

It is recommended to use high-quality quantization versions such as Q6_K_L, Q6_K, and Q5_K_L, approaching perfect performance.

Supports ARM/AVX devices

Supports online repackaging of ARM and AVX devices to optimize hardware performance.

imatrix quantization

Quantize using the imatrix option to improve model performance.

Model Capabilities

Text generation

Multi-round dialogue

Context understanding

Use Cases

Text generation

Dialogue system

Can be used to build an intelligent dialogue system, supporting multi-round dialogue.

Content creation

Generate high-quality text content, such as articles, stories, etc.

🚀 Llamacpp imatrix Quantizations of Qwen3-8B by Qwen

This project focuses on the quantization of the Qwen3-8B model. It utilizes the llama.cpp library to generate various quantized versions of the model, offering users a range of options based on their specific requirements such as performance, quality, and available hardware resources.

🚀 Quick Start

Using LM Studio: You can easily run the quantized models in LM Studio.
Using llama.cpp: Run the models directly with llama.cpp or any other llama.cpp - based project.

✨ Features

Multiple Quantization Types: Offers a wide variety of quantized versions, including BF16, Q8_0, Q6_K_L, etc., each with different file sizes and qualities.
Optimized for Different Hardware: Some quantizations are optimized for ARM and AVX machines, and there are also options for online repacking of weights to improve performance.
Prompt Format Support: Defines a specific prompt format for interacting with the model.

📦 Installation

Downloading using huggingface - cli

Click to view download instructions

First, make sure you have huggingface - cli installed:

pip install -U "huggingface_hub[cli]"

Then, you can target the specific file you want:

huggingface-cli download bartowski/Qwen_Qwen3-8B-GGUF --include "Qwen_Qwen3-8B-Q4_K_M.gguf" --local-dir ./

If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download bartowski/Qwen_Qwen3-8B-GGUF --include "Qwen_Qwen3-8B-Q8_0/*" --local-dir ./

You can either specify a new local - dir (Qwen_Qwen3-8B-Q8_0) or download them all in place (./)

💻 Usage Examples

Basic Usage

The prompt format for interacting with the model is as follows:

<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

📚 Documentation

Model Information

Property	Details
Quantized By	bartowski
Pipeline Tag	text - generation
License	apache - 2.0
Base Model	Qwen/Qwen3 - 8B
Base Model Relation	quantized

Download Options

You can download a specific file from the following table:

Filename	Quant type	File Size	Split	Description
Qwen3-8B-bf16.gguf	bf16	16.39GB	false	Full BF16 weights.
Qwen3-8B-Q8_0.gguf	Q8_0	8.71GB	false	Extremely high quality, generally unneeded but max available quant.
Qwen3-8B-Q6_K_L.gguf	Q6_K_L	7.03GB	false	Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
Qwen3-8B-Q6_K.gguf	Q6_K	6.73GB	false	Very high quality, near perfect, recommended.
Qwen3-8B-Q5_K_L.gguf	Q5_K_L	6.24GB	false	Uses Q8_0 for embed and output weights. High quality, recommended.
Qwen3-8B-Q5_K_M.gguf	Q5_K_M	5.85GB	false	High quality, recommended.
Qwen3-8B-Q5_K_S.gguf	Q5_K_S	5.72GB	false	High quality, recommended.
Qwen3-8B-Q4_K_L.gguf	Q4_K_L	5.49GB	false	Uses Q8_0 for embed and output weights. Good quality, recommended.
Qwen3-8B-Q4_1.gguf	Q4_1	5.25GB	false	Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon.
Qwen3-8B-Q4_K_M.gguf	Q4_K_M	5.03GB	false	Good quality, default size for most use cases, recommended.
Qwen3-8B-Q3_K_XL.gguf	Q3_K_XL	4.98GB	false	Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
Qwen3-8B-Q4_K_S.gguf	Q4_K_S	4.80GB	false	Slightly lower quality with more space savings, recommended.
Qwen3-8B-Q4_0.gguf	Q4_0	4.79GB	false	Legacy format, offers online repacking for ARM and AVX CPU inference.
Qwen3-8B-IQ4_NL.gguf	IQ4_NL	4.79GB	false	Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference.
Qwen3-8B-IQ4_XS.gguf	IQ4_XS	4.56GB	false	Decent quality, smaller than Q4_K_S with similar performance, recommended.
Qwen3-8B-Q3_K_L.gguf	Q3_K_L	4.43GB	false	Lower quality but usable, good for low RAM availability.
Qwen3-8B-Q3_K_M.gguf	Q3_K_M	4.12GB	false	Low quality.
Qwen3-8B-IQ3_M.gguf	IQ3_M	3.90GB	false	Medium - low quality, new method with decent performance comparable to Q3_K_M.
Qwen3-8B-Q2_K_L.gguf	Q2_K_L	3.89GB	false	Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
Qwen3-8B-Q3_K_S.gguf	Q3_K_S	3.77GB	false	Low quality, not recommended.
Qwen3-8B-IQ3_XS.gguf	IQ3_XS	3.63GB	false	Lower quality, new method with decent performance, slightly better than Q3_K_S.
Qwen3-8B-IQ3_XXS.gguf	IQ3_XXS	3.37GB	false	Lower quality, new method with decent performance, comparable to Q3 quants.
Qwen3-8B-Q2_K.gguf	Q2_K	3.28GB	false	Very low quality but surprisingly usable.
Qwen3-8B-IQ2_M.gguf	IQ2_M	3.05GB	false	Relatively low quality, uses SOTA techniques to be surprisingly usable.

Embed/output weights

Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.

ARM/AVX information

Previously, you would download Q4_0_4_4/4_8/8_8, and these would have their weights interleaved in memory in order to improve performance on ARM and AVX machines by loading up more data in one pass.

Now, however, there is something called "online repacking" for weights. Details in this PR. If you use Q4_0 and your hardware would benefit from repacking weights, it will do it automatically on the fly.

As of llama.cpp build b4282 you will not be able to run the Q4_0_X_X files and will instead need to use Q4_0.

Additionally, if you want to get slightly better quality, you can use IQ4_NL thanks to this PR which will also repack the weights for ARM, though only the 4_4 for now. The loading time may be slower but it will result in an overall speed increase.

Click to view Q4_0_X_X information (deprecated

I'm keeping this section to show the potential theoretical uplift in performance from using the Q4_0 with online repacking.

Click to view benchmarks on an AVX2 system (EPYC7702)

model	size	params	backend	threads	test	t/s	% (vs Q4_0)
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	pp512	204.03 ± 1.03	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	pp1024	282.92 ± 0.19	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	pp2048	259.49 ± 0.44	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	tg128	39.12 ± 0.27	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	tg256	39.31 ± 0.69	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	tg512	40.52 ± 0.03	100%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	pp512	301.02 ± 1.74	147%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	pp1024	287.23 ± 0.20	101%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	pp2048	262.77 ± 1.81	101%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	tg128	18.80 ± 0.99	48%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	tg256	24.46 ± 3.04	83%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	tg512	36.32 ± 3.59	90%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	pp512	271.71 ± 3.53	133%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	pp1024	279.86 ± 45.63	100%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	pp2048	320.77 ± 5.00	124%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	tg128	43.51 ± 0.05	111%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	tg256	43.35 ± 0.09	110%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	tg512	42.60 ± 0.31	105%

Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation

Which file should I choose?

Click here for details

A great write - up with charts showing various performances is provided by Artefact2 here

The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.

If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1 - 2GB smaller than your GPU's total VRAM.

If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1 - 2GB smaller than that total.

Next, you'll need to decide if you want to use an 'I - quant' or a 'K - quant'.

If you don't want to think too much, grab one of the K - quants. These are in format 'QX_K_X', like Q5_K_M.

If you want to get more into the weeds, you can check out this extremely useful feature chart:

[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature - matrix)

But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I - quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.

These I - quants can also be used on CPU, but will be slower than their K - quant equivalent, so speed vs performance is a trade - off you'll have to decide.

🔧 Technical Details

Quantization Method: Using llama.cpp release b5200 for quantization. All quants are made using the imatrix option with a dataset from here.
Online Repacking: For ARM and AVX machines, there is an "online repacking" feature for weights. Details can be found in this PR.

📄 License

This project is licensed under the apache - 2.0 license.

Credits

Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset.
Thank you ZeroWw for the inspiration to experiment with embed/output.
Thank you to LM Studio for sponsoring my work.

Want to support my work? Visit my ko - fi page here: https://ko - fi.com/bartowski

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご