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:

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	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/13B-Legerdemain-L2-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/13B-Legerdemain-L2-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/13B-Legerdemain-L2-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: CalderaAI's 13B Legerdemain L2

13B-Legerdemain-L2

13B-Legerdemain-L2 is the first model merge of its kind in a series of LLaMaV2 models mixed using a custom script built in-house by CalderaAI called Model-REVOLVER. M-REVOLVER is also the first in a series of custom scripts based on the concept of mixtuning - not only does the end user have contol over which models are mixed and their percentages on a per-layer basis, we tackle the problem of overcomplexity that arises from such a level of control; this model is the first of its series.

The Model-REVOLVER Process Designed by CalderaAI

M-REVOLVER (Rapid Evolution Via Optimized-List Viewer Evaluated Response) Per-layer merging between parent models is a nebulous inexact science, and therefore impractical to most users despite the raw power it offers. We propose an entirely new approach that gives the user a clear looking glass into the impact vastly different layer merge configurations between selected parent models of their choice will have on the potential offspring model - especially its inherited behaviors. We've developed solution MK.1 - A cyclic random pattern search in place that determines all layer merge ratios, combines test models, infers prompt completions, and deletes a prototype after data collection is saved. When the cyclic system has completed its entire run, nothing is left but the telemetry collected along with the cycle and layer merge ratios from every single prototype merge. This data is then used to empower the user to choose which offspring is most fit to their desired outcome. This final step is only initiated when all necessary data has been aggregated from all assembled-tested-erased prototypes sampled in the search space.

From here, the user is provided five 300 token prompt completions from each and every offspring contender that was created and tested during the cyclic process. The user simply browses each prototype's series of responses and selects their desired outcome model by entering the cycle number associated with the prompt completions they feel best suits their vision. That model is then instantly repatriated into the official offspring of its parent models and tokenizer files found to be most relevant are instantly auto-copied from the parent model dir to the offspring.

That's it - the user instantly has a complete model based on the behavior they decided on, suggested from one of many potentials; all with their own unique trait inheritence thanks to layer merge auto randomization inside an ordered system. One more thing - the user not only selects how many cycles to run, the user can edit prompts.txt which the system reads as a single prompt - this means if the user desires to use any multiline instruct format to observe all potential model outcomes from instruct, or desires simply their own prompt, it's up to them.. simply works.

Link to GitHub for M-REVOLVER are at the end of the model card. More advanced MergeTech toolsets and merge techniques are currently under internal testing and development by Caldera.

13B-Legerdemain-L2 Use

13B-Legerdemain-L2 is capable of following Alpaca instructions however it seems far more receptive to the by-the-book method as seen here:

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

### Instruction:
{instruction}

### Response:
{New Line}

The primary model of choice for this model was a story-only model called Holodeck by KoboldAI. Traits preserved seem to be detailed descriptiveness, verbosity, and characters with personality. The two other models selected were 13B-Nous-Hermes by NousResearch and 13B-orca-8k-3319 by OpenAssistant. I began the process by providing an incredibly obscene prompt and simply ignored each and every guardrail or censorship laden prompt completion and accepted the offensive ones in turn - intent wasn't to be crass but trigger censorship parts of the network to test if it's possible to completely undermine them. Second pass with offspring model and Orca was a simple milquetoast prompt to gauge vocabulary, word flow, and intelligence as I selected the most fit in that category. Result model seems a bit of a curiosity - different samplers and even a different UI (as I went from TGUI to KoboldAI) seem to uncover different facets of behavior. Godlike preset with Alpaca Instruct in TGUI worked fine. In KoboldAI some tweaking was necessary to get the same experience. If you choose to test this model, have fun - it's got a mind of its own.

Model-REVOLVER Git:

https://github.com/Digitous/ModelREVOLVER