Qwen Qwen2.5 VL 32B Instruct GGUF

🚀 Llamacpp imatrix Quantizations of Qwen2.5-VL-32B-Instruct by Qwen

This project focuses on the quantizations of Qwen2.5-VL-32B-Instruct using the llama.cpp framework. It provides various quantized models for different performance and quality requirements, enabling efficient deployment on diverse hardware platforms.

🚀 Quick Start

• Run in LM Studio: You can run the quantized models in LM Studio.
• Run with llama.cpp: Directly use llama.cpp or any other llama.cpp - based project to run the models.

✨ Features

• Multiple Quantization Types: Offers a wide range of quantization types, such as bf16, Q8_0, Q6_K_L, etc., to meet different needs for performance and quality.
• Online Repacking: Supports "online repacking" for weights, which can automatically optimize performance on ARM and AVX machines.
• Optimized for Different Hardware: Some quantizations are specifically optimized for ARM and AVX hardware to improve performance.

📦 Installation

Install huggingface-cli

First, make sure you have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Download Specific Files

You can target the specific file you want:

huggingface-cli download bartowski/Qwen_Qwen2.5-VL-32B-Instruct-GGUF --include "Qwen_Qwen2.5-VL-32B-Instruct-Q4_K_M.gguf" --local-dir ./

Download Split Files

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_Qwen2.5-VL-32B-Instruct-GGUF --include "Qwen_Qwen2.5-VL-32B-Instruct-Q8_0/*" --local-dir ./

You can either specify a new local - dir (Qwen_Qwen2.5-VL-32B-Instruct-Q8_0) or download them all in place (./)

💻 Usage Examples

Prompt Format

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

📚 Documentation

Model Information

Downloadable Files

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.

Which File to Choose

🔧 Technical Details

Online Repacking for Weights

The concept of "online repacking" for weights is introduced to improve the performance of ARM and AVX machines. When using Q4_0, if the hardware benefits from weight repacking, it will be done automatically.

Performance Comparison

Benchmarks on an AVX2 system (EPYC7702) show the performance differences between different models and quantization types. For example, Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation compared to Q4_0.

📄 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.

If you want to support my work, visit my ko - fi page here: https://ko-fi.com/bartowski