Llamacpp imatrix Quantizations of codegeex4-all-9b

Using llama.cpp release b3333 for quantization.

Original model: https://huggingface.co./THUDM/codegeex4-all-9b

All quants made using imatrix option with dataset from here

Prompt format

[gMASK] <sop> <|system|>
{system_prompt} <|user|>
{prompt} <|assistant|>

Download a file (not the whole branch) from below:

Filename Quant type File Size Description
codegeex4-all-9b-Q8_0.gguf Q8_0 9.99GB Extremely high quality, generally unneeded but max available quant.
codegeex4-all-9b-Q6_K_L.gguf Q6_K_L 8.56GB Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
codegeex4-all-9b-Q6_K.gguf Q6_K 8.26GB Very high quality, near perfect, recommended.
codegeex4-all-9b-Q5_K_L.gguf Q5_K_L 7.52GB Uses Q8_0 for embed and output weights. High quality, recommended.
codegeex4-all-9b-Q5_K_M.gguf Q5_K_M 7.14GB High quality, recommended.
codegeex4-all-9b-Q5_K_S.gguf Q5_K_S 6.69GB High quality, recommended.
codegeex4-all-9b-Q4_K_L.gguf Q4_K_L 6.71GB Uses Q8_0 for embed and output weights. Good quality, uses about 4.83 bits per weight, recommended.
codegeex4-all-9b-Q4_K_M.gguf Q4_K_M 6.25GB Good quality, uses about 4.83 bits per weight, recommended.
codegeex4-all-9b-Q4_K_S.gguf Q4_K_S 5.75GB Slightly lower quality with more space savings, recommended.
codegeex4-all-9b-IQ4_XS.gguf IQ4_XS 5.25GB Decent quality, smaller than Q4_K_S with similar performance, recommended.
codegeex4-all-9b-Q3_K_XL.gguf Q3_K_XL 5.82GB Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
codegeex4-all-9b-Q3_K_L.gguf Q3_K_L 5.28GB Lower quality but usable, good for low RAM availability.
codegeex4-all-9b-Q3_K_M.gguf Q3_K_M 5.06GB Even lower quality.
codegeex4-all-9b-IQ3_M.gguf IQ3_M 4.81GB Medium-low quality, new method with decent performance comparable to Q3_K_M.
codegeex4-all-9b-Q3_K_S.gguf Q3_K_S 4.58GB Low quality, not recommended.
codegeex4-all-9b-IQ3_XS.gguf IQ3_XS 4.42GB Lower quality, new method with decent performance, slightly better than Q3_K_S.
codegeex4-all-9b-IQ3_XXS.gguf IQ3_XXS 4.25GB Lower quality, new method with decent performance, comparable to Q3 quants.
codegeex4-all-9b-Q2_K_L.gguf Q2_K_L 4.59GB Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
codegeex4-all-9b-Q2_K.gguf Q2_K 3.99GB Very low quality but surprisingly usable.
codegeex4-all-9b-IQ2_M.gguf IQ2_M 3.93GB Very low quality, uses SOTA techniques to also be surprisingly usable.
codegeex4-all-9b-IQ2_S.gguf IQ2_S 3.76GB Very low quality, uses SOTA techniques to be usable.
codegeex4-all-9b-IQ2_XS.gguf IQ2_XS 3.61GB Very low quality, uses SOTA techniques to be usable.

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

Downloading using huggingface-cli

First, make sure you have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Then, you can target the specific file you want:

huggingface-cli download bartowski/codegeex4-all-9b-GGUF --include "codegeex4-all-9b-Q4_K_M.gguf" --local-dir ./

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/codegeex4-all-9b-GGUF --include "codegeex4-all-9b-Q8_0.gguf/*" --local-dir codegeex4-all-9b-Q8_0

You can either specify a new local-dir (codegeex4-all-9b-Q8_0) or download them all in place (./)

Which file should I choose?

A great write up with charts showing various performances is provided by Artefact2 here

The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.

If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.

If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.

Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.

If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.

If you want to get more into the weeds, you can check out this extremely useful feature chart:

llama.cpp feature matrix

But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.

These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.

The I-quants are not compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.

Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski

Downloads last month
1,114
GGUF
Model size
9.4B params
Architecture
chatglm

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

32-bit

Inference Examples
Inference API (serverless) has been turned off for this model.