README.md · bartowski/Hermes-3-Llama-3.1-405B-GGUF at 69eba52b31f63bfd9b750b04778b847e22bc50bf

metadata

base_model: NousResearch/Hermes-3-Llama-3.1-405B
language:
  - en
license: llama3
pipeline_tag: text-generation
tags:
  - Llama-3
  - instruct
  - finetune
  - chatml
  - gpt4
  - synthetic data
  - distillation
  - function calling
  - json mode
  - axolotl
  - roleplaying
  - chat
quantized_by: bartowski
widget:
  - example_title: Hermes 3
    messages:
      - role: system
        content: >-
          You are a sentient, superintelligent artificial general intelligence,
          here to teach and assist me.
      - role: user
        content: >-
          Write a short story about Goku discovering kirby has teamed up with
          Majin Buu to destroy the world.
model-index:
  - name: Hermes-3-Llama-3.1-405B
    results: []

Llamacpp imatrix Quantizations of Hermes-3-Llama-3.1-405B

Using llama.cpp release b3600 for quantization.

Original model: https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-405B

All quants made using imatrix option with dataset from here

Imatrix was measured from Q8 instead of full weights, this seems to have not caused any issues, will post some info after further research

Run them in LM Studio

Prompt format

<|begin_of_text|><|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

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

Filename	Quant type	File Size	Split	Description
Hermes-3-Llama-3.1-405B-Q8_0.gguf	Q8_0	431.24GB	true	Extremely high quality, generally unneeded but max available quant.
Hermes-3-Llama-3.1-405B-Q6_K.gguf	Q6_K	332.95GB	true	Very high quality, near perfect, recommended.
Hermes-3-Llama-3.1-405B-Q5_K_M.gguf	Q5_K_M	286.65GB	true	High quality, recommended.
Hermes-3-Llama-3.1-405B-Q4_K_L.gguf	Q4_K_L	244.63GB	true	Uses Q8_0 for embed and output weights. Good quality, recommended.
Hermes-3-Llama-3.1-405B-Q4_K_M.gguf	Q4_K_M	243.07GB	true	Good quality, default size for must use cases, recommended.
Hermes-3-Llama-3.1-405B-Q4_K_S.gguf	Q4_K_S	230.50GB	true	Slightly lower quality with more space savings, recommended.
Hermes-3-Llama-3.1-405B-IQ4_XS.gguf	IQ4_XS	216.57GB	true	Decent quality, smaller than Q4_K_S with similar performance, recommended.
Hermes-3-Llama-3.1-405B-Q3_K_XL.gguf	Q3_K_XL	214.68GB	true	Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
Hermes-3-Llama-3.1-405B-Q3_K_L.gguf	Q3_K_L	212.84GB	true	Lower quality but usable, good for low RAM availability.
Hermes-3-Llama-3.1-405B-Q3_K_M.gguf	Q3_K_M	195.37GB	true	Low quality.
Hermes-3-Llama-3.1-405B-IQ3_M.gguf	IQ3_M	181.74GB	true	Medium-low quality, new method with decent performance comparable to Q3_K_M.
Hermes-3-Llama-3.1-405B-Q3_K_S.gguf	Q3_K_S	175.23GB	true	Low quality, not recommended.
Hermes-3-Llama-3.1-405B-IQ3_XXS.gguf	IQ3_XXS	155.85GB	true	Lower quality, new method with decent performance, comparable to Q3 quants.
Hermes-3-Llama-3.1-405B-Q2_K_L.gguf	Q2_K_L	151.37GB	true	Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
Hermes-3-Llama-3.1-405B-Q2_K.gguf	Q2_K	149.32GB	true	Very low quality but surprisingly usable.
Hermes-3-Llama-3.1-405B-IQ2_M.gguf	IQ2_M	136.67GB	true	Relatively low quality, uses SOTA techniques to be surprisingly usable.
Hermes-3-Llama-3.1-405B-IQ2_XS.gguf	IQ2_XS	119.35GB	true	Low quality, uses SOTA techniques to be usable.
Hermes-3-Llama-3.1-405B-IQ2_XXS.gguf	IQ2_XXS	107.27GB	true	Very low quality, uses SOTA techniques to be usable.
Hermes-3-Llama-3.1-405B-IQ1_M.gguf	IQ1_M	93.50GB	true	Extremely low quality, not recommended.

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.

Some say that this improves the quality, others don't notice any difference. If you use these models PLEASE COMMENT with your findings. I would like feedback that these are actually used and useful so I don't keep uploading quants no one is using.

Thanks!

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/Hermes-3-Llama-3.1-405B-GGUF --include "Hermes-3-Llama-3.1-405B-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/Hermes-3-Llama-3.1-405B-GGUF --include "Hermes-3-Llama-3.1-405B-Q8_0/*" --local-dir ./

You can either specify a new local-dir (Hermes-3-Llama-3.1-405B-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