Understanding AI model files

If you want to understand which model files to use for LLM text generation a great article is written by Bartowski [1] for his Nemotron LLama trained model version. We summarizes it's contents to give you an idea on which model size is best for your solution.

Before you dive into understanding the model file names, please readh our what is the difference between I-quant and K-quant models.

Typical model file names

Filename Quant type File Size Split Description
Llama-3.1-70B-Q8_0.gguf Q8_0 74.98GB true Extremely high quality, generally unneeded but max available quant.
Llama-3.1-70B-Q6_K.gguf Q6_K 57.89GB true Very high quality, near perfect, recommended.
Llama-3.1-70B-Q5_K_L.gguf Q5_K_L 50.60GB true Uses Q8_0 for embed and output weights. High quality, recommended.
Llama-3.1-70B-Q5_K_M.gguf Q5_K_M 49.95GB true High quality, recommended.
Llama-3.1-70B-Q5_K_S.gguf Q5_K_S 48.66GB false High quality, recommended.
Llama-3.1-70B-Q4_K_L.gguf Q4_K_L 43.30GB false Uses Q8_0 for embed and output weights. Good quality, recommended.
Llama-3.1-70B-Q4_K_M.gguf Q4_K_M 42.52GB false Good quality, default size for must use cases, recommended.
Llama-3.1-70B-Q4_K_S.gguf Q4_K_S 40.35GB false Slightly lower quality with more space savings, recommended.
Llama-3.1-70B-Q4_0.gguf Q4_0 40.12GB false Legacy format, generally not worth using over similarly sized formats
Llama-3.1-70B-Q3_K_XL.gguf Q3_K_XL 38.06GB false Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
Llama-3.1-70B-IQ4_XS.gguf IQ4_XS 37.90GB false Decent quality, smaller than Q4_K_S with similar performance, recommended.
Llama-3.1-70B-Q3_K_L.gguf Q3_K_L 37.14GB false Lower quality but usable, good for low RAM availability.
Llama-3.1-70B-Q3_K_M.gguf Q3_K_M 34.27GB false Low quality.
Llama-3.1-70B-IQ3_M.gguf IQ3_M 31.94GB false Medium-low quality, new method with decent performance comparable to Q3_K_M.
Llama-3.1-70B-Q3_K_S.gguf Q3_K_S 30.91GB false Low quality, not recommended.
Llama-3.1-70B-IQ3_XXS.gguf IQ3_XXS 27.47GB false Lower quality, new method with decent performance, comparable to Q3 quants.
Llama-3.1-70B-Q2_K_L.gguf Q2_K_L 27.40GB false Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
Llama-3.1-70B-Q2_K.gguf Q2_K 26.38GB false Very low quality but surprisingly usable.
Llama-3.1-70B-IQ2_M.gguf IQ2_M 24.12GB false Relatively low quality, uses SOTA techniques to be surprisingly usable.
Llama-3.1-70B-IQ2_XS.gguf IQ2_XS 21.14GB false Low quality, uses SOTA techniques to be usable.
Llama-3.1-70B-IQ2_XXS.gguf IQ2_XXS 19.10GB false Very low quality, uses SOTA techniques to be usable.
Llama-3.1-70B-IQ1_M.gguf IQ1_M 16.75GB false Extremely low quality, not recommended.

Q4_0_X_X

These are NOT for Metal (Apple) offloading, only ARM chips. If you're using an ARM chip, the Q4_0_X_X quants will have a substantial speedup. Check out Q4_0_4_4 speed comparisons on the original pull request To check which one would work best for your ARM chip, you can check AArch64 SoC features.

Which file should I choose?

If you want to decide which models to use, 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'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.

References:

[1] https://huggingface.co/bartowski/Llama-3.1-70B-GGUF

More information