README.md

metadata

base_model: THUDM/glm-4-9b-chat
inference: false
language:
  - zh
  - en
library_name: gguf
license: other
license_link: https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/LICENSE
license_name: glm-4
pipeline_tag: text-generation
quantized_by: legraphista
tags:
  - glm
  - chatglm
  - thudm
  - quantized
  - GGUF
  - quantization
  - static
  - 16bit
  - 8bit
  - 6bit
  - 5bit
  - 4bit
  - 3bit
  - 2bit

glm-4-9b-chat-GGUF

Llama.cpp static quantization of THUDM/glm-4-9b-chat

Original Model: THUDM/glm-4-9b-chat
Original dtype: BF16 (bfloat16)
Quantized by: https://github.com/ggerganov/llama.cpp/pull/6999
IMatrix dataset: here

• Files
  • 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

Common Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
glm-4-9b-chat.Q8_0.gguf	Q8_0	9.99GB	✅ Available	⚪ Static	📦 No
glm-4-9b-chat.Q6_K.gguf	Q6_K	8.26GB	✅ Available	⚪ Static	📦 No
glm-4-9b-chat.Q4_K.gguf	Q4_K	6.25GB	✅ Available	⚪ Static	📦 No
glm-4-9b-chat.Q3_K	Q3_K	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q2_K	Q2_K	-	⏳ Processing	⚪ Static	-

All Quants

Filename	Quant type	File Size	Status	Uses IMatrix	Is Split
glm-4-9b-chat.BF16	BF16	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.FP16	F16	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q8_0.gguf	Q8_0	9.99GB	✅ Available	⚪ Static	📦 No
glm-4-9b-chat.Q6_K.gguf	Q6_K	8.26GB	✅ Available	⚪ Static	📦 No
glm-4-9b-chat.Q5_K	Q5_K	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q5_K_S	Q5_K_S	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q4_K.gguf	Q4_K	6.25GB	✅ Available	⚪ Static	📦 No
glm-4-9b-chat.Q4_K_S	Q4_K_S	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.IQ4_NL	IQ4_NL	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.IQ4_XS	IQ4_XS	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q3_K	Q3_K	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q3_K_L	Q3_K_L	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q3_K_S	Q3_K_S	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.IQ3_M	IQ3_M	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.IQ3_S	IQ3_S	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.IQ3_XS	IQ3_XS	-	⏳ Processing	⚪ Static	-
glm-4-9b-chat.Q2_K	Q2_K	-	⏳ Processing	⚪ Static	-

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/glm-4-9b-chat-GGUF --include "glm-4-9b-chat.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/glm-4-9b-chat-GGUF --include "glm-4-9b-chat.Q8_0/*" --local-dir ./
# see FAQ for merging GGUF's

---

Inference

Simple chat template

[gMASK]<sop><|user|>
{user_prompt}<|assistant|>
{assistant_response}<|user|>
{next_user_prompt}

Chat template with system prompt

[gMASK]<sop><|system|>
{system_prompt}<|user|>
{user_prompt}<|assistant|>
{assistant_response}<|user|>
{next_user_prompt}

Llama.cpp

llama.cpp/main -m glm-4-9b-chat.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: glm-4-9b-chat.Q8_0)
3. Run gguf-split --merge glm-4-9b-chat.Q8_0/glm-4-9b-chat.Q8_0-00001-of-XXXXX.gguf glm-4-9b-chat.Q8_0.gguf
   • Make sure to point gguf-split to the first chunk of the split.

---

Got a suggestion? Ping me @legraphista!