Multiple GPTQ parameter permutations are provided; see Provided Files below for details of the options provided, their parameters, and the software used to create them.
These files were quantised using hardware kindly provided by Massed Compute.
This may not be a complete list; if you know of others, please let me know!
Provided files, and GPTQ parameters
Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
Each separate quant is in a different branch. See below for instructions on fetching from different branches.
Most GPTQ files are made with AutoGPTQ. Mistral models are currently made with Transformers.
Explanation of GPTQ parameters
Bits: The bit size of the quantised model.
GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
Act Order: True or False. Also known as desc_act. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now.
Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
GPTQ dataset: The calibration dataset used during quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ calibration dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s).
Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences.
ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama and Mistral models in 4-bit.
If you remove the --local-dir-use-symlinks False parameter, the files will instead be stored in the central Hugging Face cache directory (default location on Linux is: ~/.cache/huggingface), and symlinks will be added to the specified --local-dir, pointing to their real location in the cache. This allows for interrupted downloads to be resumed, and allows you to quickly clone the repo to multiple places on disk without triggering a download again. The downside, and the reason why I don't list that as the default option, is that the files are then hidden away in a cache folder and it's harder to know where your disk space is being used, and to clear it up if/when you want to remove a download model.
The cache location can be changed with the HF_HOME environment variable, and/or the --cache-dir parameter to huggingface-cli.
Note that using Git with HF repos is strongly discouraged. It will be much slower than using huggingface-hub, and will use twice as much disk space as it has to store the model files twice (it stores every byte both in the intended target folder, and again in the .git folder as a blob.)
Python code example: inference from this GPTQ model
Install the necessary packages
Requires: Transformers 4.33.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
shell
1pip3 install --upgrade transformers optimum
2# If using PyTorch 2.1 + CUDA 12.x:3pip3 install --upgrade auto-gptq
4# or, if using PyTorch 2.1 + CUDA 11.x:5pip3 install --upgrade auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
If you are using PyTorch 2.0, you will need to install AutoGPTQ from source. Likewise if you have problems with the pre-built wheels, you should try building from source:
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
Patreon special mentions: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
Original model card: Disco Research's DiscoLM 70B
EM Logo
DiscoLM 70b
DiscoLM 70b is a 70b model based on Laion's LeoLM 70b which underwent additional continued pretraining for 65b tokens of German
text, strengthening it's multilingual capabilities while retaining (and partially improving) English capabilities.
This was then further finetuned on a combination of some the most popular open-source instruction sets.
DiscoLM 70b is a DiscoResearch project and was trained by Björn Plüster.
The model was trained with compute provided by HessianAI - we are very grateful for their support; please check out their wesbite and projects!
This models is still an early Alpha and we can't guarantee that there isn't any contamination.
However, the average of 71.24 would earn the #2 spot on the HF leaderboard at the time of writing.
Metric
Value
ARC (25-shot)
68.77
HellaSwag (10-shot)
85.41
MMLU (5-shot)
68.64
TruthfulQA (0-shot)
57.69
Winogrande (5-shot)
83.27
GSM8k (5-shot)
63.68
Avg.
71.24
We use Language Model Evaluation Harness to run the benchmark tests above, using the same version as the HuggingFace LLM Leaderboard.
FastEval
Metric
Value
GSM8K
70.6
Math
17.8
BBH
63.4
MMLU
64.7
Avg.
48.87
Screenshot of the current (sadly no longer maintained) FastEval CoT leaderboard:
Screenshot of the current FastEval MT Bench leaderboard:
FastEval Leaderboard
Prompt Format
This model follows the ChatML format:
<|im_start|>system
You are DiscoLM, a helpful assistant.
<|im_end|>
<|im_start|>user
Please tell me possible reasons to call a research collective "Disco Research"<|im_end|>
<|im_start|>assistant
This formatting is also available via a pre-defined Transformers chat template, which means that lists of messages can be formatted for you with the apply_chat_template() method:
python
1chat =[2{"role":"system","content":"You are DiscoLM, a helpful assistant."},3{"role":"user","content":"Please tell me possible reasons to call a research collective Disco Research"}4]5tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
If you use tokenize=True and return_tensors="pt" instead, then you will get a tokenized and formatted conversation ready to pass to model.generate().
Dataset
The dataset curation for DiscoLM 70b followed a "brute force"/"PoC" approach.
The following datasets were used for training DiscoLM 70b:
DiscoResearch is an aspiring open research community. Disco should be a place where researchers from many communities can come together to combine their expertise and create innovative and groundbreaking LLMs. Come join our Discord, share your opinions and ideas, and advance open LLM research with us!
Acknowledgements
Disco 70b is a DiscoResearch project and was trained by Björn Plüster. Jan Harries helped with technical adivce, logistics and the Model Card.
AutoMeta also provided helpful technical advice and rounded up his connections to select a set of high-quality datasets.
The model was trained with compute provided by HessianAI - many thanks in particular to Patrick Schramowski for his support.
We are standing on the shoulders of giants; many thanks in no particular order to Laion for LeoLM 70b
(especially to Christoph Schuhmann who got us all connected),
TheBloke for providing quantized versions, winglian for Axolotl which was used to train the model and the SlimOrca dataset, garage-bAInd, Teknium, Migel Tissera, MetaMath for their great datasets (please contact us if we forgot to mention you here!).
The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model.
This model should only be used for research purposes. The original Llama2 license and all restrictions of datasets used to train this model apply.
