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: Vicuna

A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: {prompt} 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	wikitext	4096	3.89 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/Xwin-LM-7B-V0.1-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/Xwin-LM-7B-V0.1-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}")

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/Xwin-LM-7B-V0.1-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'''A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: {prompt} 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 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: Xwin-LM's Xwin-LM 7B V0.1

Xwin-LM: Powerful, Stable, and Reproducible LLM Alignment

Step up your LLM alignment with Xwin-LM!

Xwin-LM aims to develop and open-source alignment technologies for large language models, including supervised fine-tuning (SFT), reward models (RM), reject sampling, reinforcement learning from human feedback (RLHF), etc. Our first release, built-upon on the Llama2 base models, ranked TOP-1 on AlpacaEval. Notably, it's the first to surpass GPT-4 on this benchmark. The project will be continuously updated.

News

💥 [Sep, 2023] We released Xwin-LM-70B-V0.1, which has achieved a win-rate against Davinci-003 of 95.57% on AlpacaEval benchmark, ranking as TOP-1 on AlpacaEval. It was the FIRST model surpassing GPT-4 on AlpacaEval. Also note its winrate v.s. GPT-4 is 60.61.
🔍 [Sep, 2023] RLHF plays crucial role in the strong performance of Xwin-LM-V0.1 release!
💥 [Sep, 2023] We released Xwin-LM-13B-V0.1, which has achieved 91.76% win-rate on AlpacaEval, ranking as top-1 among all 13B models.
💥 [Sep, 2023] We released Xwin-LM-7B-V0.1, which has achieved 87.82% win-rate on AlpacaEval, ranking as top-1 among all 7B models.

Model Card

Model	Checkpoint	Report	License
Xwin-LM-7B-V0.1	🤗 HF Link	📃Coming soon (Stay tuned)	Llama 2 License
Xwin-LM-13B-V0.1	🤗 HF Link		Llama 2 License
Xwin-LM-70B-V0.1	🤗 HF Link		Llama 2 License

Benchmarks

Xwin-LM performance on AlpacaEval.

The table below displays the performance of Xwin-LM on AlpacaEval, where evaluates its win-rate against Text-Davinci-003 across 805 questions. To provide a comprehensive evaluation, we present, for the first time, the win-rate against ChatGPT and GPT-4 as well. Our Xwin-LM model family establish a new state-of-the-art performance across all metrics. Notably, Xwin-LM-70B-V0.1 has eclipsed GPT-4 for the first time, achieving an impressive win-rate of 95.57% to Text-Davinci-003 and 60.61% to GPT-4.

Model	AlpacaEval (winrate %)	AlpacaEval (winrate %)	AlpacaEval (winrate %)
	v.s. Text-Davinci-003	v.s. ChatGPT	v.s. GPT4
Xwin-LM-70B-V0.1	95.57	87.50	60.61
GPT-4	95.28	84.66	50.00
WizardLM-70B-V1.0	92.91	80.19	46.70
Llama-2-70B-Chat	92.66	81.97	51.19
Xwin-LM-13B-V0.1	91.76	81.79	55.30
ChatGPT	89.37	50.00	16.60
WizardLM-13B-V1.2	89.17	75.75	41.91
Xwin-LM-7B-V0.1	87.35	76.40	47.57
Llama-2-13B-Chat	81.09	64.22	30.92

Case study on AlpacaEval

Prompts	GTP4 response	Xwin-LM response
I've put a plate on top of an apple, and take the plate from kitchen to dinning room. Where is the apple now	The apple is now in the dining room, as it was underneath the plate when you moved the plate from the kitchen to the dining room.	The apple is still in the kitchen, as you have only moved the plate. The apple is under the plate, so if you want to bring the apple to the dining room, you should either move the apple directly or put the apple on the plate and then move the plate to the dining room.
Answer the following question. How do you say "good evening" in French.	In French, "good evening" is said as "bonsoir."	In French, "good evening" is translated to "bonsoir." This greeting is used in the evening and nighttime hours, typically after 6 p.m. It is a combination of "bon," which means "good," and "soir," which means "evening." To pronounce it, say "bone-swahr."

Xwin-LM performance on NLP foundation tasks.

The following table provides a comparison of Xwin-LMs with other LLMs on NLP foundation tasks in Open LLM Leaderboard.

Model	MMLU 5-shot	ARC 25-shot	TruthfulQA 0-shot	HellaSwag 10-shot	Average
Text-davinci-003	56.9	85.2	59.3	82.2	70.9
Vicuna-13b 1.1	51.3	53.0	51.8	80.1	59.1
Guanaco 30B	57.6	63.7	50.7	85.1	64.3
WizardLM-7B 1.0	42.7	51.6	44.7	77.7	54.2
WizardLM-13B 1.0	52.3	57.2	50.5	81.0	60.2
WizardLM-30B 1.0	58.8	62.5	52.4	83.3	64.2
Llama-2-7B-Chat	48.3	52.9	45.6	78.6	56.4
Llama-2-13B-Chat	54.6	59.0	44.1	81.9	59.9
Llama-2-70B-Chat	63.9	64.6	52.8	85.9	66.8
Xwin-LM-7B-V0.1	49.7	56.2	48.1	79.5	58.4
Xwin-LM-13B-V0.1	56.6	62.4	45.5	83.0	61.9
Xwin-LM-70B-V0.1	69.6	70.5	60.1	87.1	71.8

Inference

Conversation templates

To obtain desired results, please strictly follow the conversation templates when utilizing our model for inference. Our model adopts the prompt format established by Vicuna and is equipped to support multi-turn conversations.

A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: Hi! ASSISTANT: Hello.</s>USER: Who are you? ASSISTANT: I am Xwin-LM.</s>......

HuggingFace Example

python

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("Xwin-LM/Xwin-LM-7B-V0.1")
tokenizer = AutoTokenizer.from_pretrained("Xwin-LM/Xwin-LM-7B-V0.1")
(
    prompt := "A chat between a curious user and an artificial intelligence assistant. "
            "The assistant gives helpful, detailed, and polite answers to the user's questions. "
            "USER: Hello, can you help me? "
            "ASSISTANT:"
)
inputs = tokenizer(prompt, return_tensors="pt")
samples = model.generate(**inputs, max_new_tokens=4096, temperature=0.7)
output = tokenizer.decode(samples[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
print(output) 
# Of course! I'm here to help. Please feel free to ask your question or describe the issue you're having, and I'll do my best to assist you.

vllm Example

Because Xwin-LM is based on Llama2, it also offers support for rapid inference using vllm. Please refer to vllm for detailed installation instructions.

python

from vllm import LLM, SamplingParams
(
    prompt := "A chat between a curious user and an artificial intelligence assistant. "
            "The assistant gives helpful, detailed, and polite answers to the user's questions. "
            "USER: Hello, can you help me? "
            "ASSISTANT:"
)
sampling_params = SamplingParams(temperature=0.7, max_tokens=4096)
llm = LLM(model="Xwin-LM/Xwin-LM-7B-V0.1")
outputs = llm.generate([prompt,], sampling_params)

for output in outputs:
    prompt = output.prompt
    generated_text = output.outputs[0].text
    print(generated_text)

TODO

Release the source code
Release more capabilities, such as math, reasoning, and etc.

Citation

Please consider citing our work if you use the data or code in this repo.

@software{xwin-lm,
  title = {Xwin-LM},
  author = {Xwin-LM Team},
  url = {https://github.com/Xwin-LM/Xwin-LM},
  version = {pre-release},
  year = {2023},
  month = {9},
}

Acknowledgements

Thanks to Llama 2, FastChat, AlpacaFarm, and vllm.