Yuan2-M32-hf-IMat-GGUF

Llama.cpp imatrix quantization of IEITYuan/Yuan2-M32-hf

Original Model: IEITYuan/Yuan2-M32-hf
Original dtype: BF16 (bfloat16)
Quantized by: https://github.com/chong000/3rd_party/tree/main
IMatrix dataset: here

• Files
  • IMatrix
  • Common Quants
  • All Quants
• Downloading using huggingface-cli
• Inference
  • 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
Yuan2-M32-hf.Q8_0.gguf	Q8_0	42.93GB	✅ Available	⚪ Static	📦 No
Yuan2-M32-hf.Q6_K.gguf	Q6_K	33.23GB	✅ Available	⚪ Static	📦 No
Yuan2-M32-hf.Q4_K.gguf	Q4_K	24.68GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q3_K.gguf	Q3_K	19.54GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q2_K.gguf	Q2_K	15.02GB	✅ Available	🟢 IMatrix	📦 No

All Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
Yuan2-M32-hf.FP16/*	F16	80.12GB	✅ Available	⚪ Static	✂ Yes
Yuan2-M32-hf.Q8_0.gguf	Q8_0	42.93GB	✅ Available	⚪ Static	📦 No
Yuan2-M32-hf.Q6_K.gguf	Q6_K	33.23GB	✅ Available	⚪ Static	📦 No
Yuan2-M32-hf.Q5_K.gguf	Q5_K	28.82GB	✅ Available	⚪ Static	📦 No
Yuan2-M32-hf.Q5_K_S.gguf	Q5_K_S	27.96GB	✅ Available	⚪ Static	📦 No
Yuan2-M32-hf.Q4_K.gguf	Q4_K	24.68GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q4_K_S.gguf	Q4_K_S	23.19GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.IQ4_NL.gguf	IQ4_NL	22.99GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q3_K.gguf	Q3_K	19.54GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q3_K_L.gguf	Q3_K_L	21.14GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q3_K_S.gguf	Q3_K_S	17.71GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.IQ3_XXS.gguf	IQ3_XXS	15.91GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q2_K.gguf	Q2_K	15.02GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.Q2_K_S.gguf	Q2_K_S	14.05GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.IQ2_XS.gguf	IQ2_XS	12.21GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.IQ2_XXS.gguf	IQ2_XXS	11.04GB	✅ Available	🟢 IMatrix	📦 No
Yuan2-M32-hf.IQ1_S.gguf	IQ1_S	8.70GB	✅ Available	🟢 IMatrix	📦 No

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/Yuan2-M32-hf-IMat-GGUF --include "Yuan2-M32-hf.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/Yuan2-M32-hf-IMat-GGUF --include "Yuan2-M32-hf.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

---

Inference

Llama.cpp

llama.cpp/main -m Yuan2-M32-hf.Q8_0.gguf --color -i -p "prompt here"

---

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: Yuan2-M32-hf.Q8_0)
3. Run gguf-split --merge Yuan2-M32-hf.Q8_0/Yuan2-M32-hf.Q8_0-00001-of-XXXXX.gguf Yuan2-M32-hf.Q8_0.gguf
   • Make sure to point gguf-split to the first chunk of the split.

---

Got a suggestion? Ping me @legraphista!