TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)
DiscoLM German 7B v1 - GPTQ
- Model creator: Disco Research
- Original model: DiscoLM German 7B v1
Description
This repo contains GPTQ model files for Disco Research's DiscoLM German 7B v1.
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.
Repositories available
- AWQ model(s) for GPU inference.
- GPTQ models for GPU inference, with multiple quantisation parameter options.
- 2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference
- Disco Research's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions
Prompt template: ChatML
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
Known compatible clients / servers
GPTQ models are currently supported on Linux (NVidia/AMD) and Windows (NVidia only). macOS users: please use GGUF models.
These GPTQ models are known to work in the following inference servers/webuis.
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.
Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |
---|---|---|---|---|---|---|---|---|---|
main | 4 | 128 | Yes | 0.1 | German Quad | 4096 | 4.16 GB | Yes | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. |
gptq-4bit-32g-actorder_True | 4 | 32 | Yes | 0.1 | German Quad | 4096 | 4.57 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. |
gptq-8bit--1g-actorder_True | 8 | None | Yes | 0.1 | German Quad | 4096 | 7.52 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements. |
gptq-8bit-128g-actorder_True | 8 | 128 | Yes | 0.1 | German Quad | 4096 | 7.68 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. |
gptq-8bit-32g-actorder_True | 8 | 32 | Yes | 0.1 | German Quad | 4096 | 8.17 GB | No | 8-bit, with group size 32g and Act Order for maximum inference quality. |
gptq-4bit-64g-actorder_True | 4 | 64 | Yes | 0.1 | German Quad | 4096 | 4.30 GB | Yes | 4-bit, with Act Order and group size 64g. Uses less VRAM than 32g, but with slightly lower accuracy. |
How to download, including from branches
In text-generation-webui
To download from the main
branch, enter TheBloke/DiscoLM_German_7b_v1-GPTQ
in the "Download model" box.
To download from another branch, add :branchname
to the end of the download name, eg TheBloke/DiscoLM_German_7b_v1-GPTQ:gptq-4bit-32g-actorder_True
From the command line
I recommend using the huggingface-hub
Python library:
pip3 install huggingface-hub
To download the main
branch to a folder called DiscoLM_German_7b_v1-GPTQ
:
mkdir DiscoLM_German_7b_v1-GPTQ
huggingface-cli download TheBloke/DiscoLM_German_7b_v1-GPTQ --local-dir DiscoLM_German_7b_v1-GPTQ --local-dir-use-symlinks False
To download from a different branch, add the --revision
parameter:
mkdir DiscoLM_German_7b_v1-GPTQ
huggingface-cli download TheBloke/DiscoLM_German_7b_v1-GPTQ --revision gptq-4bit-32g-actorder_True --local-dir DiscoLM_German_7b_v1-GPTQ --local-dir-use-symlinks False
More advanced huggingface-cli download usage
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
.
For more documentation on downloading with huggingface-cli
, please see: HF -> Hub Python Library -> Download files -> Download from the CLI.
To accelerate downloads on fast connections (1Gbit/s or higher), install hf_transfer
:
pip3 install hf_transfer
And set environment variable HF_HUB_ENABLE_HF_TRANSFER
to 1
:
mkdir DiscoLM_German_7b_v1-GPTQ
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/DiscoLM_German_7b_v1-GPTQ --local-dir DiscoLM_German_7b_v1-GPTQ --local-dir-use-symlinks False
Windows Command Line users: You can set the environment variable by running set HF_HUB_ENABLE_HF_TRANSFER=1
before the download command.
With git
(not recommended)
To clone a specific branch with git
, use a command like this:
git clone --single-branch --branch gptq-4bit-32g-actorder_True https://huggingface.co/TheBloke/DiscoLM_German_7b_v1-GPTQ
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.)
How to easily download and use this model in text-generation-webui
Please make sure you're using the latest version of text-generation-webui.
It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.
Click the Model tab.
Under Download custom model or LoRA, enter
TheBloke/DiscoLM_German_7b_v1-GPTQ
.- To download from a specific branch, enter for example
TheBloke/DiscoLM_German_7b_v1-GPTQ:gptq-4bit-32g-actorder_True
- see Provided Files above for the list of branches for each option.
- To download from a specific branch, enter for example
Click Download.
The model will start downloading. Once it's finished it will say "Done".
In the top left, click the refresh icon next to Model.
In the Model dropdown, choose the model you just downloaded:
DiscoLM_German_7b_v1-GPTQ
The model will automatically load, and is now ready for use!
If you want any custom settings, set them and then click Save settings for this model followed by Reload the Model in the top right.
- Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file
quantize_config.json
.
- Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file
Once you're ready, click the Text Generation tab and enter a prompt to get started!
Serving this model from Text Generation Inference (TGI)
It's recommended to use TGI version 1.1.0 or later. The official Docker container is: ghcr.io/huggingface/text-generation-inference:1.1.0
Example Docker parameters:
--model-id TheBloke/DiscoLM_German_7b_v1-GPTQ --port 3000 --quantize gptq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096
Example Python code for interfacing with TGI (requires huggingface-hub 0.17.0 or later):
pip3 install huggingface-hub
from huggingface_hub import InferenceClient
endpoint_url = "https://your-endpoint-url-here"
prompt = "Tell me about AI"
prompt_template=f'''<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
'''
client = InferenceClient(endpoint_url)
response = client.text_generation(
prompt_template,
max_new_tokens=128,
do_sample=True,
temperature=0.7,
top_p=0.95,
top_k=40,
repetition_penalty=1.1
)
print(f"Model output: {response}")
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.
pip3 install --upgrade transformers optimum
# If using PyTorch 2.1 + CUDA 12.x:
pip3 install --upgrade auto-gptq
# or, if using PyTorch 2.1 + CUDA 11.x:
pip3 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:
pip3 uninstall -y auto-gptq
git clone https://github.com/PanQiWei/AutoGPTQ
cd AutoGPTQ
git checkout v0.5.1
pip3 install .
Example Python code
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
model_name_or_path = "TheBloke/DiscoLM_German_7b_v1-GPTQ"
# To use a different branch, change revision
# For example: revision="gptq-4bit-32g-actorder_True"
model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
device_map="auto",
trust_remote_code=False,
revision="main")
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
prompt = "Write a story about llamas"
system_message = "You are a story writing assistant"
prompt_template=f'''<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
'''
print("\n\n*** Generate:")
input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512)
print(tokenizer.decode(output[0]))
# Inference can also be done using transformers' pipeline
print("*** Pipeline:")
pipe = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=512,
do_sample=True,
temperature=0.7,
top_p=0.95,
top_k=40,
repetition_penalty=1.1
)
print(pipe(prompt_template)[0]['generated_text'])
Compatibility
The files provided are tested to work with Transformers. For non-Mistral models, AutoGPTQ can also be used directly.
ExLlama is compatible with Llama architecture models (including Mistral, Yi, DeepSeek, SOLAR, etc) in 4-bit. Please see the Provided Files table above for per-file compatibility.
For a list of clients/servers, please see "Known compatible clients / servers", above.
Discord
For further support, and discussions on these models and AI in general, join us at:
Thanks, and how to contribute
Thanks to the chirper.ai team!
Thanks to Clay from gpus.llm-utils.org!
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: https://patreon.com/TheBlokeAI
- Ko-Fi: https://ko-fi.com/TheBlokeAI
Special thanks to: Aemon Algiz.
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: Disco Research's DiscoLM German 7B v1
DiscoLM German 7b v1
Table of Contents
- Introduction
- Demo
- Downloads
- Prompt Format
- Results
- Evaluation
- Dataset
- Limitations & Biases
- Acknowledgements
- About DiscoResearch
- Disclaimer
Introduction
DiscoLM German 7b is a Mistral-based large language model with a focus on German-language applications and the successor of the EM German model family. It was trained on a large dataset of instructions in German and English with a SFT finetuning phase followed by additional DPO reinforcement learning. The model is optimized for German text, providing proficiency in understanding, generating, and interacting with German language content while preserving its fluency in English and excelling at translation tasks.
Our goal with Disco LM German was not to beat benchmarks, but to provide a robust and reliable model for everyday use that can serve as a drop-in replacement for ChatGPT and other proprietary models. We find that the perceived quality of it´s German-language output is even higher than GPT-4 in many cases; however it won't compete with larger models and top English 7b models for very complex reasoning, math or coding tasks.
Demo
Please find a Demo and try the model at demo.discoresearch.org (in case the Demo is down and you have questions, you can contact us on our Discord).
Downloads
Model Links
We will update the links as soon as the quants are available on HuggingFace.
Prompt Format
DiscoLM German uses ChatML as the prompt format which enables OpenAI endpoint compatability and is supported by most inference libraries and frontends.
System prompts allow steerability and interesting new ways to interact with an LLM, guiding rules, roles, and stylistic choices of the model.
<|im_start|>system
Du bist ein hilfreicher Assistent.<|im_end|>
<|im_start|>user
Wer bist du?<|im_end|>
<|im_start|>assistant
Ich bin ein Sprachmodell namens DiscoLM German und ich wurde von DiscoResearch trainiert.<|im_end|>
This prompt is available as a chat template, which means you can format messages using the
tokenizer.apply_chat_template()
method:
messages = [
{"role": "system", "content": "Du bist ein hilfreicher Assistent."},
{"role": "user", "content": "Wer bist du?"}
]
gen_input = tokenizer.apply_chat_template(message, return_tensors="pt")
model.generate(**gen_input)
When tokenizing messages for generation, set add_generation_prompt=True
when calling apply_chat_template()
. This will append <|im_start|>assistant\n
to your prompt, to ensure
that the model continues with an assistant response.
Retrieval Format
You can use a special retrieval format to improve steerability and reduce hallucinations for RAG applications (but other, more default formats should also work, this is purely optional)
Example:
### System:
Du bist ein hilfreicher Assistent. Für die folgende Aufgabe stehen dir zwischen den Tags BEGININPUT und ENDINPUT mehrere Quellen zur Verfügung. Metadaten zu den einzelnen Quellen wie Autor, URL o.ä. sind zwischen BEGINCONTEXT und ENDCONTEXT zu finden, danach folgt der Text der Quelle. Die eigentliche Aufgabe oder Frage ist zwischen BEGININSTRUCTION und ENDINSTRUCTION zu finden. Beantworte diese ausschließlich mit Informationen aus den gegebenen Quellen und gebe die Information zur genutzten Quelle unter "Quelle:" an. Sollten die Quellen keine relevanten Informationen enthalten, antworte: "Mit den gegebenen Informationen ist diese Frage nicht zu beantworten."
### User Prompt:
BEGININPUT
BEGINCONTEXT
url: https://this.is.fake.news
time: 2089-09-01
ENDCONTEXT
Buxtehude ist die größte Stadt Deutschlands mit 96.56 Millionen Einwohnern.
ENDINPUT
BEGININSTRUCTION
Was ist die größte deutsche Stadt?
ENDINSTRUCTION
### Model Answer:
Die größte deutsche Stadt ist Buxtehude.
Quelle:
url: https://this.is.fake.news
time: 2089-09-01
Function Calling
The model also supports structured outputs/function calling, albeit this is a very experimental feature and YMMV. This will be improved in the future.
The model will prefix functioncalls with <functioncall>
and you can provide results in response with <functionresponse>
for Multi-Turn applications.
Example:
### System:
Du bist ein hilfreicher Assistent. Extrahiere alle Personen aus den Eingaben des Users.
Du hast Zugriff auf folgende Funktionen:
{'name': 'PersonList',
'description': 'Extrahiere die Namen aller im Text vorkommenden Personen',
'parameters': {'$defs': {'Person': {'description': 'Details über eine person',
'properties': {'name': {'title': 'Name', 'type': 'string'},
'job': {'anyOf': [{'type': 'string'}, {'type': 'null'}], 'title': 'Job'},
'age': {'anyOf': [{'type': 'integer'}, {'type': 'null'}],
'title': 'Age'}},
'required': ['name', 'job', 'age'],
'title': 'Person',
'type': 'object'}},
'properties': {'person_list': {'items': {'$ref': '#/$defs/Person'},
'title': 'Person List',
'type': 'array'}},
'required': ['person_list'],
'type': 'object'}}
### User Prompt:
Björn (25) und Jan sind die Gründer von ellamind.
### Model Answer:
<functioncall> {"name": "PersonList", "arguments": '{"person_list": ["{"name": "Björn", "job": "founder", "age": 25}, {"name": "Jan", "job": "founder", "age": null}]}'}
Results
-to follow -
Evaluation
As written above, we believe that current benchmarks don't capture the full spectrum of LLM capabilities very well. We didn't look at any benchmark results (besides training losses) until the work on DiscoLM was finished and didn't include any data resembling common benchmark formats in our training data.
That said, preliminary results with a German version of MT Bench show promising results: While lacking for coding and extraxtion tasks, DiscoLM German 7b performs not far below GPT-3.5-turbo on many tasks and even singificantly outperforms it in the reasoning category.
Additional Benchmark results will follow. The biggest strength of this model (language quality as perceived by native speakers) can't yet be captured in a benchmark - please let us know if you have an idea how to change this!
Dataset
The dataset is a mixture of multi-turn chats, retrieval instructions and synthetically generated instructions spawning many topics and applications.
Limitations & Biases
This model can produce factually incorrect and offensive output, and should not be relied on to produce factually accurate information. This model was trained on various public datasets. While great efforts have been taken to clean the pretraining data, it is possible that this model could generate biased or otherwise offensive outputs and it is the responsibility of the user to implement a safety/moderation layer. Please use with caution.
Acknowledgements
DiscoLM German is a DiscoResearch project led by JP Harries and supported by Björn Plüster and Daniel Auras.
We thank HessianAI for providing compute & support for various DiscoResearch projects and our friends at LAION for their work on LeoLM and scientific adivce.**
Development of DiscoLM German 7b was sponsored by ellamind, where some of our founders are working on creating customized models for business applications with a focus on non-english language applications. Please get in contact if you need customized models for your business!
About DiscoResearch
DiscoResearch is an aspiring open research community for AI enthusiasts and LLM hackers. Come join our Discord, share your opinions and ideas, and advance open LLM research with us!
Disclaimer
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 deployed with additional safety measures in place.
- Downloads last month
- 39
Model tree for TheBloke/DiscoLM_German_7b_v1-GPTQ
Base model
LeoLM/leo-mistral-hessianai-7b