legraphista's picture
Update README.md
9cd49e1 verified
metadata
base_model: Qwen/Qwen2-Math-72B-Instruct
inference: false
language:
  - en
library_name: gguf
license: other
license_link: https://huggingface.co/Qwen/Qwen2-Math-72B-Instruct/blob/main/LICENSE
license_name: tongyi-qianwen
pipeline_tag: text-generation
quantized_by: legraphista
tags:
  - chat
  - quantized
  - GGUF
  - quantization
  - imat
  - imatrix
  - static
  - 16bit
  - 8bit
  - 6bit
  - 5bit
  - 4bit
  - 3bit
  - 2bit
  - 1bit

Qwen2-Math-72B-Instruct-IMat-GGUF

Llama.cpp imatrix quantization of Qwen/Qwen2-Math-72B-Instruct

Original Model: Qwen/Qwen2-Math-72B-Instruct
Original dtype: BF16 (bfloat16)
Quantized by: llama.cpp b3547
IMatrix dataset: here


Files

IMatrix

Status: βœ… Available
Link: here

Common Quants

Filename Quant type File Size Status Uses IMatrix Is Split
Qwen2-Math-72B-Instruct.Q8_0/* Q8_0 77.26GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q6_K/* Q6_K 64.35GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q4_K/* Q4_K 47.42GB βœ… Available 🟒 IMatrix βœ‚ Yes
Qwen2-Math-72B-Instruct.Q3_K.gguf Q3_K 37.70GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.Q2_K.gguf Q2_K 29.81GB βœ… Available 🟒 IMatrix πŸ“¦ No

All Quants

Filename Quant type File Size Status Uses IMatrix Is Split
Qwen2-Math-72B-Instruct.BF16/* BF16 145.42GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.FP16/* F16 145.42GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q8_0/* Q8_0 77.26GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q6_K/* Q6_K 64.35GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q5_K/* Q5_K 54.45GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q5_K_S/* Q5_K_S 51.38GB βœ… Available βšͺ Static βœ‚ Yes
Qwen2-Math-72B-Instruct.Q4_K/* Q4_K 47.42GB βœ… Available 🟒 IMatrix βœ‚ Yes
Qwen2-Math-72B-Instruct.Q4_K_S.gguf Q4_K_S 43.89GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ4_NL.gguf IQ4_NL 41.32GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ4_XS.gguf IQ4_XS 39.71GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.Q3_K.gguf Q3_K 37.70GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.Q3_K_L.gguf Q3_K_L 39.51GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.Q3_K_S.gguf Q3_K_S 34.49GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ3_M.gguf IQ3_M 35.50GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ3_S.gguf IQ3_S 34.49GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ3_XS.gguf IQ3_XS 32.84GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ3_XXS.gguf IQ3_XXS 31.85GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.Q2_K.gguf Q2_K 29.81GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.Q2_K_S.gguf Q2_K_S 29.57GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ2_M.gguf IQ2_M 29.34GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ2_S.gguf IQ2_S 27.94GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ2_XS.gguf IQ2_XS 27.06GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ2_XXS.gguf IQ2_XXS 25.49GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ1_M.gguf IQ1_M 23.74GB βœ… Available 🟒 IMatrix πŸ“¦ No
Qwen2-Math-72B-Instruct.IQ1_S IQ1_S - ❌ Error 🟒 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-Math-72B-Instruct-IMat-GGUF --include "Qwen2-Math-72B-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-Math-72B-Instruct-IMat-GGUF --include "Qwen2-Math-72B-Instruct.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

Inference

Simple chat template

<|im_start|>system
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-Math-72B-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
  2. Locate your GGUF chunks folder (ex: Qwen2-Math-72B-Instruct.Q8_0)
  3. Run gguf-split --merge Qwen2-Math-72B-Instruct.Q8_0/Qwen2-Math-72B-Instruct.Q8_0-00001-of-XXXXX.gguf Qwen2-Math-72B-Instruct.Q8_0.gguf
    • Make sure to point gguf-split to the first chunk of the split.

Got a suggestion? Ping me @legraphista!