README.md

metadata

base_model: Qwen/Qwen2.5-32B-Instruct
inference: false
language:
  - en
library_name: gguf
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/main/LICENSE
pipeline_tag: text-generation
quantized_by: legraphista
tags:
  - chat
  - quantized
  - GGUF
  - quantization
  - imat
  - imatrix
  - static
  - 16bit
  - 8bit
  - 6bit
  - 5bit
  - 4bit
  - 3bit
  - 2bit
  - 1bit

Qwen2.5-32B-Instruct-IMat-GGUF

Llama.cpp imatrix quantization of Qwen/Qwen2.5-32B-Instruct

Original Model: Qwen/Qwen2.5-32B-Instruct
Original dtype: BF16 (bfloat16)
Quantized by: llama.cpp b3787
IMatrix dataset: here

• Files
  • IMatrix
  • Common Quants
  • All Quants
• Downloading using huggingface-cli
• Inference
  • Simple chat template
  • Chat template with system prompt
  • Llama.cpp
• FAQ
  • Why is the IMatrix not applied everywhere?
  • How do I merge a split GGUF?

---

Files

IMatrix

Status: ✅ Available
Link: here

Common Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
Qwen2.5-32B-Instruct.Q8_0.gguf	Q8_0	34.82GB	✅ Available	⚪ Static	📦 No
Qwen2.5-32B-Instruct.Q6_K.gguf	Q6_K	26.89GB	✅ Available	⚪ Static	📦 No
Qwen2.5-32B-Instruct.Q4_K.gguf	Q4_K	19.85GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q3_K.gguf	Q3_K	15.94GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q2_K.gguf	Q2_K	12.31GB	✅ Available	🟢 IMatrix	📦 No

All Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
Qwen2.5-32B-Instruct.BF16/*	BF16	65.54GB	✅ Available	⚪ Static	✂ Yes
Qwen2.5-32B-Instruct.FP16/*	F16	65.54GB	✅ Available	⚪ Static	✂ Yes
Qwen2.5-32B-Instruct.Q8_0.gguf	Q8_0	34.82GB	✅ Available	⚪ Static	📦 No
Qwen2.5-32B-Instruct.Q6_K.gguf	Q6_K	26.89GB	✅ Available	⚪ Static	📦 No
Qwen2.5-32B-Instruct.Q5_K.gguf	Q5_K	23.26GB	✅ Available	⚪ Static	📦 No
Qwen2.5-32B-Instruct.Q5_K_S.gguf	Q5_K_S	22.64GB	✅ Available	⚪ Static	📦 No
Qwen2.5-32B-Instruct.Q4_K.gguf	Q4_K	19.85GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q4_K_S.gguf	Q4_K_S	18.78GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ4_NL.gguf	IQ4_NL	18.68GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ4_XS.gguf	IQ4_XS	17.69GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q3_K.gguf	Q3_K	15.94GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q3_K_L.gguf	Q3_K_L	17.25GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q3_K_S.gguf	Q3_K_S	14.39GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ3_M.gguf	IQ3_M	14.81GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ3_S.gguf	IQ3_S	14.44GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ3_XS.gguf	IQ3_XS	13.71GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ3_XXS.gguf	IQ3_XXS	12.84GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q2_K.gguf	Q2_K	12.31GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.Q2_K_S.gguf	Q2_K_S	11.49GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ2_M.gguf	IQ2_M	11.26GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ2_S.gguf	IQ2_S	10.39GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ2_XS.gguf	IQ2_XS	9.96GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ2_XXS.gguf	IQ2_XXS	9.03GB	✅ Available	🟢 IMatrix	📦 No
Qwen2.5-32B-Instruct.IQ1_M	IQ1_M	-	⏳ Processing	🟢 IMatrix	-
Qwen2.5-32B-Instruct.IQ1_S	IQ1_S	-	⏳ Processing	🟢 IMatrix	-

Downloading using huggingface-cli

If you do not have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Download the specific file you want:

huggingface-cli download legraphista/Qwen2.5-32B-Instruct-IMat-GGUF --include "Qwen2.5-32B-Instruct.Q8_0.gguf" --local-dir ./

If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download legraphista/Qwen2.5-32B-Instruct-IMat-GGUF --include "Qwen2.5-32B-Instruct.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

---

Inference

Simple chat template

<|im_start|>system
You are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>
<|im_start|>user
{user_prompt}<|im_end|>
<|im_start|>assistant
{assistant_response}<|im_end|>
<|im_start|>user
{next_user_prompt}<|im_end|>

Chat template with system prompt

<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{user_prompt}<|im_end|>
<|im_start|>assistant
{assistant_response}<|im_end|>
<|im_start|>user
{next_user_prompt}<|im_end|>

Llama.cpp

llama.cpp/main -m Qwen2.5-32B-Instruct.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"

---

FAQ

Why is the IMatrix not applied everywhere?

According to this investigation, it appears that lower quantizations are the only ones that benefit from the imatrix input (as per hellaswag results).

How do I merge a split GGUF?

1. Make sure you have gguf-split available
   • To get hold of gguf-split, navigate to https://github.com/ggerganov/llama.cpp/releases
   • Download the appropriate zip for your system from the latest release
   • Unzip the archive and you should be able to find gguf-split
2. Locate your GGUF chunks folder (ex: Qwen2.5-32B-Instruct.Q8_0)
3. Run gguf-split --merge Qwen2.5-32B-Instruct.Q8_0/Qwen2.5-32B-Instruct.Q8_0-00001-of-XXXXX.gguf Qwen2.5-32B-Instruct.Q8_0.gguf
   • Make sure to point gguf-split to the first chunk of the split.

---

Got a suggestion? Ping me @legraphista!