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: Michael Levine, 阿明, Trailburnt, Nikolai Manek, John Detwiler, Randy H, Will Dee, Sebastain Graf, NimbleBox.ai, Eugene Pentland, Emad Mostaque, Ai Maven, Jim Angel, Jeff Scroggin, Michael Davis, Manuel Alberto Morcote, Stephen Murray, Robert, Justin Joy, Luke @flexchar, Brandon Frisco, Elijah Stavena, S_X, Dan Guido, Undi ., Komninos Chatzipapas, Shadi, theTransient, Lone Striker, Raven Klaugh, jjj, Cap'n Zoog, Michel-Marie MAUDET (LINAGORA), Matthew Berman, David, Fen Risland, Omer Bin Jawed, Luke Pendergrass, Kalila, OG, Erik Bjäreholt, Rooh Singh, Joseph William Delisle, Dan Lewis, TL, John Villwock, AzureBlack, Brad, Pedro Madruga, Caitlyn Gatomon, K, jinyuan sun, Mano Prime, Alex, Jeffrey Morgan, Alicia Loh, Illia Dulskyi, Chadd, transmissions 11, fincy, Rainer Wilmers, ReadyPlayerEmma, knownsqashed, Mandus, biorpg, Deo Leter, Brandon Phillips, SuperWojo, Sean Connelly, Iucharbius, Jack West, Harry Royden McLaughlin, Nicholas, terasurfer, Vitor Caleffi, Duane Dunston, Johann-Peter Hartmann, David Ziegler, Olakabola, Ken Nordquist, Trenton Dambrowitz, Tom X Nguyen, Vadim, Ajan Kanaga, Leonard Tan, Clay Pascal, Alexandros Triantafyllidis, JM33133, Xule, vamX, ya boyyy, subjectnull, Talal Aujan, Alps Aficionado, wassieverse, Ari Malik, James Bentley, Woland, Spencer Kim, Michael Dempsey, Fred von Graf, Elle, zynix, William Richards, Stanislav Ovsiannikov, Edmond Seymore, Jonathan Leane, Martin Kemka, usrbinkat, Enrico Ros
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
Original model card: Southern university of science and technology's SUS Chat 34B
🐷SUS-Chat: Instruction tuning done right
News
2023-12-05: SUS-Chat is ranked 2nd in Open LLM
leaderboard
and surpassed all models under 70B.
2023-12-01: SUS-Chat-34B is now avaliable on HuggingFace🤗.
Inrtoduction
Figure 1: DALL·E 2023-12-01 11.03.28 - An imposing, majestic wild boar combined with elements of a futuristic transformer robot. The boar itself should be intricately blended with these tra
SUS-Chat is a 34B bilingual Chinese-English dialogue model, jointly
released by the Southern University of Science and Technology and
International Digital Economy Academy. The SUS-Chat-34B model has
been fine-tuned on millions of high-quality, multilingual instruction
data. While maintaining the strong language capabilities of the base
model, the SUS-Chat-34B model has improved the model’s response to human
instructions through high-quality instruction fine-tuning and excels at
imitating human thought processes through chains of thought. It
introduces inter-instruction attention sharing in long texts, expanding
the window size from 4K to 8K, significantly enhancing the usability of
multi-round dialogues.
It has surpassed all models of the same size in almost all benchmark
tests and is better suited to meet the practical needs of complex
multilingual tasks. Compared to larger models, SUS-Chat-34B remains
highly competitive and achieved state-of-the-art performance in our
comprehensive evaluations.
SUS-Chat powerfully demonstrates that through the right instruction
fine-tuning, academic institutions can achieve better performance
without increasing model parameters, using open-source datasets and
models. This bridges the gap between academia and industry in large
language models and opens new possibilities for collaboration between
academic and industrial sectors.
Performance
To better evaluate the performance of the SUS-Chat-34B model, we
conducted assessments across multiple benchmark tests and have
open-sourced the evaluation framework
TLEM to facilitate
replication and comparison by other researchers.
In TLEM, we utilized various benchmark tests including MMLU, CMMLU,
C-Eval, BBH, GSM-8K, and MATH, focusing on measuring the model’s
knowledge and thinking capabilities. In these metrics, the SUS-Chat-34B
model achieved state-of-the-art performance. Additionally, we
incorporated
lm-eval to test
SUS-Chat and similar models on winogrande, hellaswag, arc, and
truthful-qa, assessing the model’s common-sense reasoning ability and
susceptibility to illusions.
Overall, the SUS-Chat-34B model significantly outperformed models of
similar scale and achieved the most advanced comprehensive performance.
model
mmlu-chat
cmmlu-chat
ceval-chat
gsm8k
BBH
MATH
winogrande
arc
hellaswag
truthfulqa
average
GPT-4
83
71
69.9
91.4
86.7
45.8
87.5
94.5
91.4
nan
80.1333
SUS-Chat-34B
77.35
78.68
82.42
80.06
67.62
28.8
81.22
81.54
83.79
57.47
71.895
Qwen-72B-Chat
74.52
77.02
77.22
76.57
72.63
35.9
80.58
81.29
87.02
50.64
71.339
DeepSeek-67B-Chat
69.43
48.51
59.7
74.45
69.73
29.56
76.09
82.1
86.06
56.37
65.2
OrionStar-34B
68.51
66.88
65.13
54.36
62.88
12.8
77.27
80.19
84.54
53.24
62.58
Yi-34B-Chat
66.96
55.16
77.16
63.76
61.54
10.02
76.64
70.66
82.29
54.57
61.876
Figure 2: Benchmark
Usage
SUS-Chat-34B is a standard LLaMA model and should be seamlessly
compatible with the LLaMA ecosystem. We provide the following example to
demonstrate how it can be used for multi-turn dialogues.
python
1from transformers import AutoModelForCausalLM, AutoTokenizer
234defchat_template(messages):5 history =""6for message in messages:7match message:8case{"role":"user","content": message}:9 history +=f"### Human: {message}\n\n### Assistant: "10case{"role":"assistant","content": message}:11 history += message
12return history
131415model_path ="SUSTech/SUS-Chat-34B"1617tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)18model = AutoModelForCausalLM.from_pretrained(19 model_path, device_map="auto", torch_dtype="auto"20).eval()2122messages =[{"role":"user","content":"hi"}]2324input_ids = tokenizer.encode(chat_template(messages), return_tensors="pt").to("cuda")25output_ids = model.generate(input_ids.to("cuda"))26response = tokenizer.decode(27 output_ids[0][input_ids.shape[1]:], skip_special_tokens=True28)2930messages.append({"role":"assistant","content": response})3132# Second round3334messages.append({"role":"user","content":"What is the capital of China?"})3536input_ids = tokenizer.encode(chat_template(messages), return_tensors="pt").to("cuda")37output_ids = model.generate(input_ids.to("cuda"))38response = tokenizer.decode(39 output_ids[0][input_ids.shape[1]:], skip_special_tokens=True40)4142messages.append({"role":"assistant","content": response})
Limitations
SUS-Chat has only undergone supervised fine-tuning and has not yet been
trained on human preference learning. As a result, it may produce
unreasonable responses in some situations and exacerbate existing issues
in language models, including hallucinations, non-determinism, and
cumulative errors. To achieve better performance for downstream tasks,
we recommend adjusting the generation configuration parameters
accordingly.
Disclaimer
During the training process, we used data compliance check algorithms to
ensure the compliance of the training model as much as possible. Due to
the complexity of the data and the diverse use cases of language models,
we cannot guarantee that the model will produce correct and reasonable
outputs in all scenarios. Please be aware that there is still a risk of
the model generating problematic outputs. We will not be responsible for
any risks or issues arising from misuse, misguidance, illegal use, and
related misinformation, as well as data security issues related to the
model.
License
This model is developed entirely for academic research and free
commercial use, but it must adhere to the
license
from 01-ai.
