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

Prompt template: Adventure

[Genres: Science Fiction]
[Tags: humor, old school, sci fi]
[Mode: Adventure]
[Description: A puzzle about committing acts of financial skulduggery and exploiting ridiculous magical items.]
[Misc: Writing era: 1993]
[Intro]
It is the year 2045. You are a young man in his twenties living in New York City. Your father was an inventor who died when you were very small; your mother raised you alone for many years until she remarried. Now you live with your stepfather, but he doesn't care much for you and has never given you any money to help support yourself. You have no job and little hope of getting one because of your lack of experience. However, you do have some unusual abilities that could be put to good use if only you knew how...

> {prompt}

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	32768	4.15 GB

Serving this model from vLLM

Documentation on installing and using vLLM can be found here.

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

If you try the vLLM examples below and get an error about quantization being unrecognised, or other AWQ-related issues, please install vLLM from Github source.

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

python3 python -m vllm.entrypoints.api_server --model TheBloke/Dans-AdventurousWinds-7B-AWQ --quantization awq --dtype half

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/Dans-AdventurousWinds-7B-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 Text Generation Inference (TGI)

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/Dans-AdventurousWinds-7B-AWQ --port 3000 --quantize awq --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

python

from huggingface_hub import InferenceClient

endpoint_url = "https://your-endpoint-url-here"

prompt = "Tell me about AI"
prompt_template=f'''[Genres: Science Fiction]
[Tags: humor, old school, sci fi]
[Mode: Adventure]
[Description: A puzzle about committing acts of financial skulduggery and exploiting ridiculous magical items.]
[Misc: Writing era: 1993]
[Intro]
It is the year 2045. You are a young man in his twenties living in New York City. Your father was an inventor who died when you were very small; your mother raised you alone for many years until she remarried. Now you live with your stepfather, but he doesn't care much for you and has never given you any money to help support yourself. You have no job and little hope of getting one because of your lack of experience. However, you do have some unusual abilities that could be put to good use if only you knew how...

> {prompt}

'''

client = InferenceClient(endpoint_url)
response = client.text_generation(prompt,
                                  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}")

How to use this AWQ model from Python code

Install the necessary packages

Requires: AutoAWQ 0.1.1 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/Dans-AdventurousWinds-7B-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'''[Genres: Science Fiction]
[Tags: humor, old school, sci fi]
[Mode: Adventure]
[Description: A puzzle about committing acts of financial skulduggery and exploiting ridiculous magical items.]
[Misc: Writing era: 1993]
[Intro]
It is the year 2045. You are a young man in his twenties living in New York City. Your father was an inventor who died when you were very small; your mother raised you alone for many years until she remarried. Now you live with your stepfather, but he doesn't care much for you and has never given you any money to help support yourself. You have no job and little hope of getting one because of your lack of experience. However, you do have some unusual abilities that could be put to good use if only you knew how...

> {prompt}

'''

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:

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

Thank you to all my generous patrons and donaters!

And thank you again to a16z for their generous grant.

Original model card: PocketDoc Labs's Dans AdventurousWinds 7B

What is the model for?

This model is proficient in crafting text-based adventure games. It can both concise replies and more expansive, novel-like descriptions. The ability to alternate between these two response styles can be triggered by a distinct system message.

What's in the sausage?

This model was trained on Mistral-7b using a custom version of the floyd and cys datasets from the skein text adventure dataset. The dataset has been improved over what was used in my previous model "Dans-CreepingSenseOfDoom" and this time presented in the more traditional adventure mode format.

Prompt format and examples:

Format description:

Traditional adventure format with a description of the genre, tags, and other misc information at the top followed by an optional introductory message. User messages are prefixed with two new lines and "> ", model messages are one new line after the last user message or the description message if an introductory message is desired.

Examples of shorter responses:

[Genres: Science Fiction]
[Tags: humor, old school, sci fi]
[Mode: Adventure]
[Description: A puzzle about committing acts of financial skulduggery and exploiting ridiculous magical items.]
[Misc: Writing era: 1993]
[Intro]
It is the year 2045. You are a young man in his twenties living in New York City. Your father was an inventor who died when you were very small; your mother raised you alone for many years until she remarried. Now you live with your stepfather, but he doesn't care much for you and has never given you any money to help support yourself. You have no job and little hope of getting one because of your lack of experience. However, you do have some unusual abilities that could be put to good use if only you knew how...

> Look around
You are standing on the sidewalk outside your apartment building. The street is busy with traffic and people going about their daily business. Across the street from you stands a tall office tower where your stepfather works as an accountant. To your left is a convenience store where you can buy food and drinks.

> You go into the store
The doorbell rings as you enter the store. An elderly woman behind the counter looks up at you expectantly. "Hello," she says cheerfully, "what can I get for you?"

> You tell her you'll browse
"All right then," she replies brightly, "just take your time."

> You ask if she's hiring
"I'm sorry," she says sadly, "but we don't need anyone new just now."

> You leave without buying anything
As you walk out of the store, the bell jingles merrily.

> You go see your stepfather
When you arrive at work, your stepfather greets you warmly. He invites you into his office and offers you a seat. "So," he begins eagerly, "how would you like to make some extra cash? There's this guy named Mr. Jones who owes my company thousands of dollars. If we could somehow convince him to pay us back, it would really help our bottom line."

He pauses expectantly, waiting for your response.

> You ask him whats involved
"Well," he continues excitedly, "it turns out that Mr. Jones has a very valuable antique watch that he keeps locked away in a safe deposit box at the bank. We think that if we could somehow break into the box and steal the watch, we could sell it for enough money to cover his debt."

Your heart races as you realize the potential danger of such a scheme. But before you can say anything, your stepfather cuts off your thoughts by continuing, "Of course, there's always the chance that something could go wrong and we might end up losing everything..."

Examples of longer novel like responses:

[Genres: grimdark fantasy] [Tags: Fantasy, Grimdark, Serious, Villain Protagonist, War] [Description: This story is definitely more "story" than "game", however there are two distinct paths that can be followed, each with their own pitfalls and additional choices. This story can get long, and assuming you're surviving, years do pass in it and you and your environment will change.] [First message intro] [Mode: Interactive Storyteller]
You have been a soldier for as long as you remember. You don't know who you were before the war began, but you were someone else. Someone less angry. More forgiving. Perhaps even kinder.

It all started when you were very young. The King of this land was attacked by an army from another kingdom. His forces were outmatched, his city overrun. He had no choice but to flee into the forest, taking what few soldiers he could muster along with him.

As the king fled deeper into the woods, his army fell back, forming a protective wall around them. They fought valiantly against the invaders, giving the king time to escape, but they eventually succumbed to overwhelming numbers. When the fighting finally ended, only one man remained alive; you.

The king found you hiding behind some bushes, covered in blood and dirt. He took pity on you and offered you a place at court if you would swear allegiance to him. Without hesitation, you accepted. And so began your journey down the path of darkness...

> You meet with the king to discuss the plans
A week has passed since the battle and you have just met with the king to discuss future plans. It seems like things are finally starting to settle down after such chaos. However, there is still much work to be done before we can truly consider ourselves safe again.

The first order of business is rebuilding our army. We need men willing to fight for us once more. Men who understand what needs to be done and aren't afraid to die trying. I believe I have found such warriors among those loyal subjects who survived the initial attack. If we train them properly, then perhaps they might prove useful against our enemies next time around.

Next, we must find ways to improve our defenses. There are many areas where improvements could be made. For instance, we should construct better walls around our cities. These walls must not only keep intruders out, but also allow us easy access inside should anyone try to breach them. In addition, we must strengthen our gates so that no matter how strong our enemy becomes, he cannot break through without great effort. Finally, we must fortify our castles and other strategic locations throughout the country. By doing these things, we will ensure that nobody ever attacks us again!

> You ask what he desires your involvement to be
After hearing about his plans, you feel compelled to help him achieve victory. After all, you owe everything to him. But before agreeing to anything, you want to make sure that he understands exactly what you expect from him. So you decide to approach him directly.

"Your Majesty," you begin, "I am honored beyond measure to serve under your command. However, before committing myself fully to this cause, I wish to clarify certain matters."

He nods slowly. "Very well. What would you like me to explain?"

"Well," you say, "first off, I want to know why you chose me above everyone else? Why did you pick me specifically?"

He smiles softly. "Because I saw potential within you. Potential which others lacked. Potentials which could prove valuable during times of crisis. Therefore, I felt confident enough to entrust my fate to you."

"But what makes you think I possess such abilities?" you ask curiously.

"Oh, nothing special really," he replies casually. "Just raw determination combined with intelligence and resourcefulness. Those qualities alone are enough to carry us forward toward victory!"

Some quick and dirty training details:

Built with Axolotl
Sequence length: 4096
# of epochs: 3
Training time: 1 hour
Hardware: 1x RTX 3090
Training type: QLoRA
PEFT R/A: 32/32

Credits:

Skein Text Adventure Data:

Thank you to the Kobold AI community for curating the Skein dataset, which is pivotal to this model's capabilities.