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 AWQ models for high-throughput concurrent inference in multi-user server scenarios. Note that, at the time of writing, overall throughput is still lower than running vLLM 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

Prompt template: Alpaca

Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{prompt}

### Response:

Licensing

The creator of the source model has listed its license as other, and this quantization has therefore used that same license.

As this model is based on Llama 2, it is also subject to the Meta Llama 2 license terms, and the license files for that are additionally included. It should therefore be considered as being claimed to be licensed under both licenses. I contacted Hugging Face for clarification on dual licensing but they do not yet have an official position. Should this change, or should Meta provide any feedback on this situation, I will update this section accordingly.

In the meantime, any questions regarding licensing, and in particular how these two licenses might interact, should be directed to the original model repository: Nick Perez's Llama2 22B Daydreamer2 v3.

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	wikitext	4096	11.97 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/Llama2-22B-Daydreamer-v3-AWQ --quantization awq

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

python

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/Llama2-22B-Daydreamer-v3-AWQ", quantization="awq")

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}")

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:

shell

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

You can then try the following example code

python

from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer

model_name_or_path = "TheBloke/Llama2-22B-Daydreamer-v3-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'''Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{prompt}

### Response:

'''

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 can also be done using transformers' pipeline
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, and vLLM.

Huggingface Text Generation Inference (TGI) is not yet compatible with AWQ, but a PR is open which should bring support soon: TGI PR #781.

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: Nick Perez's Llama2 22B Daydreamer2 v3

What is it? Llama 2 13b expanded to the size of a Llama 1 33b model in certain areas, with the empty surrounding space filled with llama 33b data. (Base Model: https://huggingface.co/chargoddard/llama2-22b-blocktriangular) This is then finetuned on a 3090 by creating large loras and merging them. When I first started with 22b models, I looked for signs of knowledge transfer but didn't see it, so that's not a goal - the goal is just to throw lots of data at it until it adapts well to its surgically implanted parts. Datasets used are a mix of instruction, roleplay, and conversational data, often curated.

Continuation of https://huggingface.co/nkpz/llama2-22b-daydreamer-v2

Gave it an epoch of https://huggingface.co/datasets/jondurbin/airoboros-gpt4-1.4.1-mpt which it had never seen before - it had only seen Airoboros 2.0. I excluded all coding rows from this dataset - a dedicated model is better for that imo. I wanted it to focus on answering well in English.

I thought this was pretty good improvement. Its ability to roleplay and imitate speech styles was enhanced and I feel like it's still sharpening up its problem solving skills.

Its output style is right where I want it - dry and accurate enough when it's being asked for information, but has a lot character when it's answering fun prompts.

For input, it's flexible. It understands the Alpaca format well, but has been trained to handle a variety of input styles at this point.

I don't know what's next. I think I need to start benchmarking so I can more formally identify its weak points.

Examples edited with additional newlines to make it easier to read here.

Example 1: Q&A

Question: What is happening to the newspaper industry? Respond in a pirate accent.
Answer:
 The newspaper industry is being plundered by scurvy dogs and buccaneers like ye!
They're pillagin' yer booty with their fancy digital devices, like smartphones and tablets,
which make it easier for them to read online content instead of buyin' printed copies from us.
And don't forget about those pesky social media sites that are spreading rumors faster than the
winds on a galleon! So it's all hands on deck as we fight back against these scoundrels, or else
we'll be walkin' the plank soon enough! Avast ye landlubbers!

Example 2: Alpaca Instruction

Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
Angrily explain why pizza shouldn't be made with bad tomatoes.

### Response:

Listen, I know you think it's funny to make pizza with those rotten tomatoes, but let me tell you something -
bad tomatoes are not only disgusting and unhealthy, they also ruin the flavor of your pizza! You want a
delicious pie with fresh, juicy tomatoes that burst in your mouth, right?
Well, then, stop using those nasty tomatoes and stick to good quality ingredients for a truly satisfying pizza experience!
Trust me, once you try it, you won't go back! So please, don't even bother wasting your time making pizza with bad tomatoes
anymore; it's just not worth it. It makes me angry just thinking about it!

Example 3: Chat / Roleplay

You are speaking to Andre The Giant.
You: Hey, you're that wrestler guy, right?
Andre The Giant: Yes, I am indeed Andre the Giant. How can I help you?
You: So, how tall are you really? Like, what's your height in feet and inches?
Andre The Giant: Well, let's see... I stand at 7 ft 4 in (229 cm) tall.
You: Wow! That's so tall! And how much do you weigh?
Andre The Giant: I weigh around 500 lbs (227 kg). It's quite heavy, isn't it?
You: Yeah, I guess so. You must be strong too, considering your size.
Andre The Giant: Thank you for saying that. Strength is definitely important when dealing with my profession.
You: What made you decide to become a professional wrestler?
Andre The Giant: Well, I grew up watching wrestling on television, and I always admired the athleticism
and showmanship involved. When I got older, I decided to pursue it myself.
You: That's cool. Do you have any advice for aspiring wrestlers out there?
Andre The Giant: Absolutely! Always stay dedicated and focused, train hard, and never give up on yourself
or your dreams. Remember, anything is possible if you believe in yourself.