TheBloke/leo-hessianai-13B-chat-AWQ

Chat & support: TheBloke's Discord server

Want to contribute? TheBloke's Patreon page

TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)

---

Leo Hessianai 13B Chat - AWQ

• Model creator: LAION LeoLM
• Original model: Leo Hessianai 13B Chat

Description

This repo contains AWQ model files for LAION LeoLM's Leo Hessianai 13B Chat.

About AWQ

AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference.

It is also now supported by continuous batching server vLLM, allowing use of Llama AWQ models for high-throughput concurrent inference in multi-user server scenarios.

As of September 25th 2023, preliminary Llama-only AWQ support has also been added to Huggingface Text Generation Inference (TGI).

Note that, at the time of writing, overall throughput is still lower than running vLLM or TGI with unquantised models, however using AWQ enables using much smaller GPUs which can lead to easier deployment and overall cost savings. For example, a 70B model can be run on 1 x 48GB GPU instead of 2 x 80GB.

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
• LAION LeoLM'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

Provided files, and AWQ parameters

For my first release of AWQ models, I am releasing 128g models only. I will consider adding 32g as well if there is interest, and once I have done perplexity and evaluation comparisons, but at this time 32g models are still not fully tested with AutoAWQ and vLLM.

Models are released as sharded safetensors files.

Branch	Bits	GS	AWQ Dataset	Seq Len	Size
main	4	128	German Quad	8192	7.25 GB

Serving this model from vLLM

Documentation on installing and using vLLM can be found here.

• When using vLLM as a server, pass the --quantization awq parameter, for example:

python3 python -m vllm.entrypoints.api_server --model TheBloke/leo-hessianai-13B-chat-AWQ --quantization awq --dtype half

Note: at the time of writing, vLLM has not yet done a new release with support for the quantization parameter.

If you try the code below and get an error about quantization being unrecognised, please install vLLM from Github source.

When using vLLM from Python code, pass the quantization=awq parameter, for example:

from vllm import LLM, SamplingParams

prompts = [
    "Hello, my name is",
    "The president of the United States is",
    "The capital of France is",
    "The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)

llm = LLM(model="TheBloke/leo-hessianai-13B-chat-AWQ", quantization="awq", dtype="half")

outputs = llm.generate(prompts, sampling_params)

# Print the outputs.
for output in outputs:
    prompt = output.prompt
    generated_text = output.outputs[0].text
    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")

Serving this model from TGI

TGI merged support for AWQ on September 25th, 2023. At the time of writing you need to use the :latest Docker container: ghcr.io/huggingface/text-generation-inference:latest

Add the parameter --quantize awq for AWQ support.

Example parameters:

--model-id TheBloke/leo-hessianai-13B-chat-AWQ --port 3000 --quantize awq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096

How to use this AWQ model from Python code

Install the necessary packages

Requires: AutoAWQ 0.0.2 or later

pip3 install autoawq

If you have problems installing AutoAWQ using the pre-built wheels, install it from source instead:

pip3 uninstall -y autoawq
git clone https://github.com/casper-hansen/AutoAWQ
cd AutoAWQ
pip3 install .

You can then try the following example code

from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer

model_name_or_path = "TheBloke/leo-hessianai-13B-chat-AWQ"

# Load model
model = AutoAWQForCausalLM.from_quantized(model_name_or_path, fuse_layers=True,
                                          trust_remote_code=False, safetensors=True)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=False)

prompt = "Tell me about AI"
prompt_template=f'''<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

'''

print("\n\n*** Generate:")

tokens = tokenizer(
    prompt_template,
    return_tensors='pt'
).input_ids.cuda()

# Generate output
generation_output = model.generate(
    tokens,
    do_sample=True,
    temperature=0.7,
    top_p=0.95,
    top_k=40,
    max_new_tokens=512
)

print("Output: ", tokenizer.decode(generation_output[0]))

"""
# Inference should be possible with transformers pipeline as well in future
# But currently this is not yet supported by AutoAWQ (correct as of September 25th 2023)
from transformers import 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:

• AutoAWQ
• vLLM
• Huggingface Text Generation Inference (TGI)

TGI merged AWQ support on September 25th, 2023: TGI PR #1054. Use the :latest Docker container until the next TGI release is made.

Discord

For further support, and discussions on these models and AI in general, join us at:

TheBloke AI's Discord server

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: Alicia Loh, Stephen Murray, K, Ajan Kanaga, RoA, Magnesian, Deo Leter, Olakabola, Eugene Pentland, zynix, Deep Realms, Raymond Fosdick, Elijah Stavena, Iucharbius, Erik Bjäreholt, Luis Javier Navarrete Lozano, Nicholas, theTransient, John Detwiler, alfie_i, knownsqashed, Mano Prime, Willem Michiel, Enrico Ros, LangChain4j, OG, Michael Dempsey, Pierre Kircher, Pedro Madruga, James Bentley, Thomas Belote, Luke @flexchar, Leonard Tan, Johann-Peter Hartmann, Illia Dulskyi, Fen Risland, Chadd, S_X, Jeff Scroggin, Ken Nordquist, Sean Connelly, Artur Olbinski, Swaroop Kallakuri, Jack West, Ai Maven, David Ziegler, Russ Johnson, transmissions 11, John Villwock, Alps Aficionado, Clay Pascal, Viktor Bowallius, Subspace Studios, Rainer Wilmers, Trenton Dambrowitz, vamX, Michael Levine, 준교 김, Brandon Frisco, Kalila, Trailburnt, Randy H, Talal Aujan, Nathan Dryer, Vadim, 阿明, ReadyPlayerEmma, Tiffany J. Kim, George Stoitzev, Spencer Kim, Jerry Meng, Gabriel Tamborski, Cory Kujawski, Jeffrey Morgan, Spiking Neurons AB, Edmond Seymore, Alexandros Triantafyllidis, Lone Striker, Cap'n Zoog, Nikolai Manek, danny, ya boyyy, Derek Yates, usrbinkat, Mandus, TL, Nathan LeClaire, subjectnull, Imad Khwaja, webtim, Raven Klaugh, Asp the Wyvern, Gabriel Puliatti, Caitlyn Gatomon, Joseph William Delisle, Jonathan Leane, Luke Pendergrass, SuperWojo, Sebastain Graf, Will Dee, Fred von Graf, Andrey, Dan Guido, Daniel P. Andersen, Nitin Borwankar, Elle, Vitor Caleffi, biorpg, jjj, NimbleBox.ai, Pieter, Matthew Berman, terasurfer, Michael Davis, Alex, Stanislav Ovsiannikov

Thank you to all my generous patrons and donaters!

And thank you again to a16z for their generous grant.

Original model card: LAION LeoLM's Leo Hessianai 13B Chat

LAION LeoLM: Linguistically Enhanced Open Language Model

Meet LeoLM, the first open and commercially available German Foundation Language Model built on Llama-2. Our models extend Llama-2's capabilities into German through continued pretraining on a large corpus of German-language and mostly locality specific text. Thanks to a compute grant at HessianAI's new supercomputer 42, we release two foundation models trained with 8k context length, LeoLM/leo-hessianai-7b and LeoLM/leo-hessianai-13b under the Llama-2 community license (70b also coming soon! 👀). With this release, we hope to bring a new wave of opportunities to German open-source and commercial LLM research and accelerate adoption. Read our blog post or our paper (preprint coming soon) for more details!

A project by Björn Plüster and Christoph Schuhmann in collaboration with LAION and HessianAI.

LeoLM Chat

LeoLM/leo-hessianai-13b-chat is a German chat model built on our foundation model LeoLM/leo-hessianai-13b and finetuned on a selection of German instruction datasets. The model performs exceptionally well on writing, explanation and discussion tasks but struggles somewhat with math and advanced reasoning. See our MT-Bench-DE scores:

{
    "first_turn": 6.525,
    "second_turn": 5.15,
    "categories": {
        "writing": 6.925,
        "roleplay": 6.7,
        "reasoning": 4.55,
        "math": 3.25,
        "coding": 3.45,
        "extraction": 5.4,
        "stem": 7.55,
        "humanities": 8.875
    },
    "average": 5.8375
}

Model Details

• Finetuned from: LeoLM/leo-hessianai-13b
• Model type: Causal decoder-only transformer language model
• Language: English and German
• Demo: Web Demo
• License: LLAMA 2 COMMUNITY LICENSE AGREEMENT
• Contact: LAION Discord or Björn Plüster

Use in 🤗Transformers

First install direct dependencies:

pip install transformers torch sentencepiece

If you want faster inference using flash-attention2, you need to install these dependencies:

pip install packaging ninja
pip install flash-attn==v2.1.1 --no-build-isolation
pip install git+https://github.com/HazyResearch/flash-attention.git@v2.1.1#subdirectory=csrc/rotary

Then load the model in transformers:

from transformers import pipeline
import torch

system_prompt = """<|im_start|>system
Dies ist eine Unterhaltung zwischen einem intelligenten, hilfsbereitem KI-Assistenten und einem Nutzer.
Der Assistent gibt ausführliche, hilfreiche und ehrliche Antworten.<|im_end|>

"""
prompt_format = "<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
prompt = "Erkläre mir wie die Fahrradwegesituation in Hamburg ist."

generator = pipeline(model="LeoLM/leo-hessianai-13b-chat", device="cuda", torch_dtype=torch.float16, trust_remote_code=True) # True for flash-attn2 else False
print(generator(prompt_format.format(prompt=prompt), do_sample=True, top_p=0.95, max_length=8192))

"Hamburg verfügt über ein umfangreiches Netzwerk an Radwegen, das ein effizientes und angenehmes Radfahren in der ganzen Stadt ermöglicht. Die Radwege variieren in Bezug auf ihre Qualität, wobei einige erstklassig mit eigens für Radfahrer reservierten Spuren sind, während andere einfache Fahrradsymbole auf dem Boden haben, die anzeigen, dass Radfahrer abwechselnd mit dem Autoverkehr auf der Straße fahren sollten. Einige Nebenstraßen haben auch spezielle Fahrradspuren, wobei einige mit Bordsteinabsenkungen zur Seite der Autospuren markiert sind. Zusätzlich haben viele Hauptstraßen, insbesondere in Nebenstraßen, fahrradfreundliche Abstände zwischen den geparkten Autos und dem Gehweg, was ein bequemes Fahren auf der Straße ermöglicht. Der Bau von Radschnellwegen, die schnelles und effizientes Radfahren in und aus der Stadt ermöglichen, ist im Gange und wird in den kommenden Jahren fortgesetzt. Insgesamt sind die Radwege in Hamburg weitläufig und gut ausgeschildert, was es zu einem angenehmen Ort macht, um mit dem Fahrrad zu fahren."

Prompting / Prompt Template

Prompt dialogue template (ChatML format):

"""
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
"""

The model input can contain multiple conversation turns between user and assistant, e.g.

<|im_start|>user
{prompt 1}<|im_end|>
<|im_start|>assistant
{reply 1}<|im_end|>
<|im_start|>user
{prompt 2}<|im_end|>
<|im_start|>assistant
(...)

Ethical Considerations and Limitations

LeoLM has been tested in English and German, and has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, the potential outputs of LeoLM/leo-hessianai-13b-chat cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of LeoLM/leo-hessianai-13b-chat, developers should perform safety testing and tuning tailored to their specific applications of the model.

Please see Meta's Responsible Use Guide.

Finetuning Details

Hyperparameter	Value
Num epochs	3
Examples per epoch	131214
Global batch size	256
Learning rate	3e-5
Warmup steps	100
LR scheduler	Cosine
Adam betas	(0.9, 0.95)

Dataset Details

## Stats for 'Subset of OpenAssistant/OASST-DE' (3534 samples (100.0%))
-----------------
  Accepted: 3534/3534 (100.0%)
  Accepted tokens: 2259302
  Skipped: 0 (0.0%)
  Min tokens per sample: 29
  Max tokens per sample: 2484
  Avg tokens per sample: 639.3044708545557
-----------------

## Stats for 'Subset of FreedomIntelligence/evol-instruct-deutsch' (57841 samples (100.0%))
-----------------
  Accepted: 57841/57841 (100.0%)
  Accepted tokens: 42958192
  Skipped: 0 (0.0%)
  Min tokens per sample: 33
  Max tokens per sample: 5507
  Avg tokens per sample: 742.6944900675991
-----------------

## Stats for 'Subset of FreedomIntelligence/alpaca-gpt4-deutsch' (48969 samples (100.0%))
-----------------
  Accepted: 48969/48969 (100.0%)
  Accepted tokens: 13372005
  Skipped: 0 (0.0%)
  Min tokens per sample: 19
  Max tokens per sample: 1359
  Avg tokens per sample: 273.07082031489307
-----------------

## Stats for 'Subset of LeoLM/OpenSchnabeltier' (21314 samples (100.0%))
-----------------
  Accepted: 21314/21314 (100.0%)
  Accepted tokens: 8134690
  Skipped: 0 (0.0%)
  Min tokens per sample: 25
  Max tokens per sample: 1202
  Avg tokens per sample: 381.65947264708643
-----------------

## Stats for 'Subset of LeoLM/German_Poems' (490 samples (100.0%))
-----------------
  Accepted: 490/490 (100.0%)
  Accepted tokens: 618642
  Skipped: 0 (0.0%)
  Min tokens per sample: 747
  Max tokens per sample: 1678
  Avg tokens per sample: 1262.534693877551
-----------------

## Stats for 'Subset of LeoLM/German_Songs' (392 samples (100.0%))
-----------------
  Accepted: 392/392 (100.0%)
  Accepted tokens: 187897
  Skipped: 0 (0.0%)
  Min tokens per sample: 231
  Max tokens per sample: 826
  Avg tokens per sample: 479.3290816326531
-----------------

## Stats for 'total' (132540 samples (100.0%))
-----------------
  Accepted: 132540/132540 (100.0%)
  Accepted tokens: 67530728
  Skipped: 0 (0.0%)
  Min tokens per sample: 19
  Max tokens per sample: 5507
  Avg tokens per sample: 509.51205673758864
-----------------