bartowski's picture
Add base_model (#2)
0e6abe1 verified
metadata
inference: false
license: other
license_name: mnpl
license_link: https://mistral.ai/licences/MNPL-0.1.md
tags:
  - code
language:
  - code
quantized_by: bartowski
pipeline_tag: text-generation
base_model: mistralai/Codestral-22B-v0.1

Llamacpp imatrix Quantizations of Codestral-22B-v0.1

Using llama.cpp release b3024 for quantization.

Original model: https://huggingface.co/mistralai/Codestral-22B-v0.1

All quants made using imatrix option with dataset from here

Prompt format

<s> [INST] <<SYS>>
{system_prompt}
<</SYS>>

{prompt} [/INST]  </s>

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

Filename Quant type File Size Description
Codestral-22B-v0.1-Q8_0.gguf Q8_0 23.64GB Extremely high quality, generally unneeded but max available quant.
Codestral-22B-v0.1-Q6_K.gguf Q6_K 18.25GB Very high quality, near perfect, recommended.
Codestral-22B-v0.1-Q5_K_M.gguf Q5_K_M 15.72GB High quality, recommended.
Codestral-22B-v0.1-Q5_K_S.gguf Q5_K_S 15.32GB High quality, recommended.
Codestral-22B-v0.1-Q4_K_M.gguf Q4_K_M 13.34GB Good quality, uses about 4.83 bits per weight, recommended.
Codestral-22B-v0.1-Q4_K_S.gguf Q4_K_S 12.66GB Slightly lower quality with more space savings, recommended.
Codestral-22B-v0.1-IQ4_XS.gguf IQ4_XS 11.93GB Decent quality, smaller than Q4_K_S with similar performance, recommended.
Codestral-22B-v0.1-Q3_K_L.gguf Q3_K_L 11.73GB Lower quality but usable, good for low RAM availability.
Codestral-22B-v0.1-Q3_K_M.gguf Q3_K_M 10.75GB Even lower quality.
Codestral-22B-v0.1-IQ3_M.gguf IQ3_M 10.06GB Medium-low quality, new method with decent performance comparable to Q3_K_M.
Codestral-22B-v0.1-Q3_K_S.gguf Q3_K_S 9.64GB Low quality, not recommended.
Codestral-22B-v0.1-IQ3_XS.gguf IQ3_XS 9.17GB Lower quality, new method with decent performance, slightly better than Q3_K_S.
Codestral-22B-v0.1-IQ3_XXS.gguf IQ3_XXS 8.59GB Lower quality, new method with decent performance, comparable to Q3 quants.
Codestral-22B-v0.1-Q2_K.gguf Q2_K 8.27GB Very low quality but surprisingly usable.
Codestral-22B-v0.1-IQ2_M.gguf IQ2_M 7.61GB Very low quality, uses SOTA techniques to also be surprisingly usable.
Codestral-22B-v0.1-IQ2_S.gguf IQ2_S 7.03GB Very low quality, uses SOTA techniques to be usable.
Codestral-22B-v0.1-IQ2_XS.gguf IQ2_XS 6.64GB Very low quality, uses SOTA techniques to be usable.

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/Codestral-22B-v0.1-GGUF --include "Codestral-22B-v0.1-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/Codestral-22B-v0.1-GGUF --include "Codestral-22B-v0.1-Q8_0.gguf/*" --local-dir Codestral-22B-v0.1-Q8_0

You can either specify a new local-dir (Codestral-22B-v0.1-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