Llamacpp imatrix Quantizations of Llama3-8B-Instruct-Replete-Adapted

Using llama.cpp release b3291 for quantization.

Original model: https://huggingface.co/Replete-AI/Llama3-8B-Instruct-Replete-Adapted

All quants made using imatrix option with dataset from here

Prompt format

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

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

Filename	Quant type	File Size	Description
Llama3-8B-Instruct-Replete-Adapted-Q8_0.gguf	Q8_0	8.54GB	Extremely high quality, generally unneeded but max available quant.
Llama3-8B-Instruct-Replete-Adapted-Q6_K_L.gguf	Q6_K_L	6.85GB	Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q6_K.gguf	Q6_K	6.59GB	Very high quality, near perfect, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q5_K_L.gguf	Q5_K_L	6.05GB	Uses Q8_0 for embed and output weights. High quality, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q5_K_M.gguf	Q5_K_M	5.73GB	High quality, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q5_K_S.gguf	Q5_K_S	5.59GB	High quality, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q4_K_L.gguf	Q4_K_L	5.31GB	Uses Q8_0 for embed and output weights. Good quality, uses about 4.83 bits per weight, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q4_K_M.gguf	Q4_K_M	4.92GB	Good quality, uses about 4.83 bits per weight, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q4_K_S.gguf	Q4_K_S	4.69GB	Slightly lower quality with more space savings, recommended.
Llama3-8B-Instruct-Replete-Adapted-IQ4_XS.gguf	IQ4_XS	4.44GB	Decent quality, smaller than Q4_K_S with similar performance, recommended.
Llama3-8B-Instruct-Replete-Adapted-Q3_K_XL.gguf	Q3_K_XL	4.78GB	Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
Llama3-8B-Instruct-Replete-Adapted-Q3_K_L.gguf	Q3_K_L	4.32GB	Lower quality but usable, good for low RAM availability.
Llama3-8B-Instruct-Replete-Adapted-Q3_K_M.gguf	Q3_K_M	4.01GB	Even lower quality.
Llama3-8B-Instruct-Replete-Adapted-IQ3_M.gguf	IQ3_M	3.78GB	Medium-low quality, new method with decent performance comparable to Q3_K_M.
Llama3-8B-Instruct-Replete-Adapted-Q3_K_S.gguf	Q3_K_S	3.66GB	Low quality, not recommended.
Llama3-8B-Instruct-Replete-Adapted-IQ3_XS.gguf	IQ3_XS	3.51GB	Lower quality, new method with decent performance, slightly better than Q3_K_S.
Llama3-8B-Instruct-Replete-Adapted-IQ3_XXS.gguf	IQ3_XXS	3.27GB	Lower quality, new method with decent performance, comparable to Q3 quants.
Llama3-8B-Instruct-Replete-Adapted-Q2_K_L.gguf	Q2_K_L	3.69GB	Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
Llama3-8B-Instruct-Replete-Adapted-Q2_K.gguf	Q2_K	3.17GB	Very low quality but surprisingly usable.
Llama3-8B-Instruct-Replete-Adapted-IQ2_M.gguf	IQ2_M	2.94GB	Very low quality, uses SOTA techniques to also be surprisingly usable.
Llama3-8B-Instruct-Replete-Adapted-IQ2_S.gguf	IQ2_S	2.75GB	Very low quality, uses SOTA techniques to be usable.
Llama3-8B-Instruct-Replete-Adapted-IQ2_XS.gguf	IQ2_XS	2.60GB	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/Llama3-8B-Instruct-Replete-Adapted-GGUF --include "Llama3-8B-Instruct-Replete-Adapted-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/Llama3-8B-Instruct-Replete-Adapted-GGUF --include "Llama3-8B-Instruct-Replete-Adapted-Q8_0.gguf/*" --local-dir Llama3-8B-Instruct-Replete-Adapted-Q8_0

You can either specify a new local-dir (Llama3-8B-Instruct-Replete-Adapted-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