bartowski's picture
Update metadata with huggingface_hub
383cf38 verified
|
raw
history blame
13.6 kB
metadata
quantized_by: bartowski
pipeline_tag: text-generation
datasets:
  - allenai/tulu-3-sft-mixture
base_model: allenai/Llama-3.1-Tulu-3-8B-SFT
license: llama3.1
language:
  - en

Llamacpp imatrix Quantizations of Llama-3.1-Tulu-3-8B-SFT

Using llama.cpp release b4132 for quantization.

Original model: https://huggingface.co/allenai/Llama-3.1-Tulu-3-8B-SFT

All quants made using imatrix option with dataset from here

Run them in LM Studio

Prompt format

<|system|>
{system_prompt}
<|user|>
{prompt}
<|assistant|>

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

Filename Quant type File Size Split Description
Llama-3.1-Tulu-3-8B-SFT-f16.gguf f16 16.07GB false Full F16 weights.
Llama-3.1-Tulu-3-8B-SFT-Q8_0.gguf Q8_0 8.54GB false Extremely high quality, generally unneeded but max available quant.
Llama-3.1-Tulu-3-8B-SFT-Q6_K_L.gguf Q6_K_L 6.85GB false Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q6_K.gguf Q6_K 6.60GB false Very high quality, near perfect, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q5_K_L.gguf Q5_K_L 6.06GB false Uses Q8_0 for embed and output weights. High quality, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q5_K_M.gguf Q5_K_M 5.73GB false High quality, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q5_K_S.gguf Q5_K_S 5.60GB false High quality, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q4_K_L.gguf Q4_K_L 5.31GB false Uses Q8_0 for embed and output weights. Good quality, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q4_K_M.gguf Q4_K_M 4.92GB false Good quality, default size for most use cases, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q3_K_XL.gguf Q3_K_XL 4.78GB false Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
Llama-3.1-Tulu-3-8B-SFT-Q4_K_S.gguf Q4_K_S 4.69GB false Slightly lower quality with more space savings, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q4_0.gguf Q4_0 4.68GB false Legacy format, generally not worth using over similarly sized formats
Llama-3.1-Tulu-3-8B-SFT-Q4_0_8_8.gguf Q4_0_8_8 4.66GB false Optimized for ARM and AVX inference. Requires 'sve' support for ARM (see details below). Don't use on Mac.
Llama-3.1-Tulu-3-8B-SFT-Q4_0_4_8.gguf Q4_0_4_8 4.66GB false Optimized for ARM inference. Requires 'i8mm' support (see details below). Don't use on Mac.
Llama-3.1-Tulu-3-8B-SFT-Q4_0_4_4.gguf Q4_0_4_4 4.66GB false Optimized for ARM inference. Should work well on all ARM chips, not for use with GPUs. Don't use on Mac.
Llama-3.1-Tulu-3-8B-SFT-IQ4_XS.gguf IQ4_XS 4.45GB false Decent quality, smaller than Q4_K_S with similar performance, recommended.
Llama-3.1-Tulu-3-8B-SFT-Q3_K_L.gguf Q3_K_L 4.32GB false Lower quality but usable, good for low RAM availability.
Llama-3.1-Tulu-3-8B-SFT-Q3_K_M.gguf Q3_K_M 4.02GB false Low quality.
Llama-3.1-Tulu-3-8B-SFT-IQ3_M.gguf IQ3_M 3.78GB false Medium-low quality, new method with decent performance comparable to Q3_K_M.
Llama-3.1-Tulu-3-8B-SFT-Q2_K_L.gguf Q2_K_L 3.69GB false Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
Llama-3.1-Tulu-3-8B-SFT-Q3_K_S.gguf Q3_K_S 3.66GB false Low quality, not recommended.
Llama-3.1-Tulu-3-8B-SFT-IQ3_XS.gguf IQ3_XS 3.52GB false Lower quality, new method with decent performance, slightly better than Q3_K_S.
Llama-3.1-Tulu-3-8B-SFT-Q2_K.gguf Q2_K 3.18GB false Very low quality but surprisingly usable.
Llama-3.1-Tulu-3-8B-SFT-IQ2_M.gguf IQ2_M 2.95GB false Relatively low quality, uses SOTA techniques to be surprisingly usable.

Embed/output weights

Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.

Downloading using huggingface-cli

Click to view download instructions

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/Llama-3.1-Tulu-3-8B-SFT-GGUF --include "Llama-3.1-Tulu-3-8B-SFT-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/Llama-3.1-Tulu-3-8B-SFT-GGUF --include "Llama-3.1-Tulu-3-8B-SFT-Q8_0/*" --local-dir ./

You can either specify a new local-dir (Llama-3.1-Tulu-3-8B-SFT-Q8_0) or download them all in place (./)

Q4_0_X_X information

Click to view Q4_0_X_X information These are *NOT* for Metal (Apple) or GPU (nvidia/AMD/intel) offloading, only ARM chips (and certain AVX2/AVX512 CPUs).

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 (thanks EloyOn!).

If you're using a CPU that supports AVX2 or AVX512 (typically server CPUs and AMD's latest Zen5 CPUs) and are not offloading to a GPU, the Q4_0_8_8 may offer a nice speed as well:

Click to view benchmarks on an AVX2 system (EPYC7702)
model size params backend threads test t/s % (vs Q4_0)
qwen2 3B Q4_0 1.70 GiB 3.09 B CPU 64 pp512 204.03 ± 1.03 100%
qwen2 3B Q4_0 1.70 GiB 3.09 B CPU 64 pp1024 282.92 ± 0.19 100%
qwen2 3B Q4_0 1.70 GiB 3.09 B CPU 64 pp2048 259.49 ± 0.44 100%
qwen2 3B Q4_0 1.70 GiB 3.09 B CPU 64 tg128 39.12 ± 0.27 100%
qwen2 3B Q4_0 1.70 GiB 3.09 B CPU 64 tg256 39.31 ± 0.69 100%
qwen2 3B Q4_0 1.70 GiB 3.09 B CPU 64 tg512 40.52 ± 0.03 100%
qwen2 3B Q4_K_M 1.79 GiB 3.09 B CPU 64 pp512 301.02 ± 1.74 147%
qwen2 3B Q4_K_M 1.79 GiB 3.09 B CPU 64 pp1024 287.23 ± 0.20 101%
qwen2 3B Q4_K_M 1.79 GiB 3.09 B CPU 64 pp2048 262.77 ± 1.81 101%
qwen2 3B Q4_K_M 1.79 GiB 3.09 B CPU 64 tg128 18.80 ± 0.99 48%
qwen2 3B Q4_K_M 1.79 GiB 3.09 B CPU 64 tg256 24.46 ± 3.04 83%
qwen2 3B Q4_K_M 1.79 GiB 3.09 B CPU 64 tg512 36.32 ± 3.59 90%
qwen2 3B Q4_0_8_8 1.69 GiB 3.09 B CPU 64 pp512 271.71 ± 3.53 133%
qwen2 3B Q4_0_8_8 1.69 GiB 3.09 B CPU 64 pp1024 279.86 ± 45.63 100%
qwen2 3B Q4_0_8_8 1.69 GiB 3.09 B CPU 64 pp2048 320.77 ± 5.00 124%
qwen2 3B Q4_0_8_8 1.69 GiB 3.09 B CPU 64 tg128 43.51 ± 0.05 111%
qwen2 3B Q4_0_8_8 1.69 GiB 3.09 B CPU 64 tg256 43.35 ± 0.09 110%
qwen2 3B Q4_0_8_8 1.69 GiB 3.09 B CPU 64 tg512 42.60 ± 0.31 105%

Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation

Which file should I choose?

Click here for details

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.

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.

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