TheBloke/juanako-7B-v1-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)

---

Juanako 7B V1 - AWQ

• Model creator: FBL
• Original model: Juanako 7B V1

Description

This repo contains AWQ model files for FBL's Juanako 7B V1.

These files were quantised using hardware kindly provided by Massed Compute.

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 with equivalent or better quality compared to the most commonly used GPTQ settings.

It is supported by:

• Text Generation Webui - using Loader: AutoAWQ
• vLLM - Llama and Mistral models only
• Hugging Face Text Generation Inference (TGI)
• Transformers version 4.35.0 and later, from any code or client that supports Transformers
• AutoAWQ - for use from Python code

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
• FBL'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

I currently release 128g GEMM models only. The addition of group_size 32 models, and GEMV kernel models, is being actively considered.

Models are released as sharded safetensors files.

Branch	Bits	GS	AWQ Dataset	Seq Len	Size
main	4	128	VMware Open Instruct	4096	4.15 GB

How to easily download and use this model in text-generation-webui

Please make sure you're using the latest version of text-generation-webui.

It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.

1. Click the Model tab.
2. Under Download custom model or LoRA, enter TheBloke/juanako-7B-v1-AWQ.
3. Click Download.
4. The model will start downloading. Once it's finished it will say "Done".
5. In the top left, click the refresh icon next to Model.
6. In the Model dropdown, choose the model you just downloaded: juanako-7B-v1-AWQ
7. Select Loader: AutoAWQ.
8. Click Load, and the model will load and is now ready for use.
9. If you want any custom settings, set them and then click Save settings for this model followed by Reload the Model in the top right.
10. Once you're ready, click the Text Generation tab and enter a prompt to get started!

Multi-user inference server: vLLM

Documentation on installing and using vLLM can be found here.

• Please ensure you are using vLLM version 0.2 or later.
• When using vLLM as a server, pass the --quantization awq parameter.

For example:

python3 -m vllm.entrypoints.api_server --model TheBloke/juanako-7B-v1-AWQ --quantization awq --dtype auto

• When using vLLM from Python code, again set quantization=awq.

For example:

from vllm import LLM, SamplingParams

prompts = [
    "Tell me about AI",
    "Write a story about llamas",
    "What is 291 - 150?",
    "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
]
prompt_template=f'''<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
'''

prompts = [prompt_template.format(prompt=prompt) for prompt in prompts]

sampling_params = SamplingParams(temperature=0.8, top_p=0.95)

llm = LLM(model="TheBloke/juanako-7B-v1-AWQ", quantization="awq", dtype="auto")

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}")

Multi-user inference server: Hugging Face 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/juanako-7B-v1-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

from huggingface_hub import InferenceClient

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

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

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)

Inference from Python code using Transformers

Install the necessary packages

• Requires: Transformers 4.35.0 or later.
• Requires: AutoAWQ 0.1.6 or later.

pip3 install --upgrade "autoawq>=0.1.6" "transformers>=4.35.0"

Note that if you are using PyTorch 2.0.1, the above AutoAWQ command will automatically upgrade you to PyTorch 2.1.0.

If you are using CUDA 11.8 and wish to continue using PyTorch 2.0.1, instead run this command:

pip3 install https://github.com/casper-hansen/AutoAWQ/releases/download/v0.1.6/autoawq-0.1.6+cu118-cp310-cp310-linux_x86_64.whl

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 .

Transformers example code (requires Transformers 4.35.0 and later)

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

model_name_or_path = "TheBloke/juanako-7B-v1-AWQ"

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
model = AutoModelForCausalLM.from_pretrained(
    model_name_or_path,
    low_cpu_mem_usage=True,
    device_map="cuda:0"
)

# Using the text streamer to stream output one token at a time
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

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

# Convert prompt to tokens
tokens = tokenizer(
    prompt_template,
    return_tensors='pt'
).input_ids.cuda()

generation_params = {
    "do_sample": True,
    "temperature": 0.7,
    "top_p": 0.95,
    "top_k": 40,
    "max_new_tokens": 512,
    "repetition_penalty": 1.1
}

# Generate streamed output, visible one token at a time
generation_output = model.generate(
    tokens,
    streamer=streamer,
    **generation_params
)

# Generation without a streamer, which will include the prompt in the output
generation_output = model.generate(
    tokens,
    **generation_params
)

# Get the tokens from the output, decode them, print them
token_output = generation_output[0]
text_output = tokenizer.decode(token_output)
print("model.generate output: ", text_output)

# Inference is also possible via Transformers' pipeline
from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    **generation_params
)

pipe_output = pipe(prompt_template)[0]['generated_text']
print("pipeline output: ", pipe_output)

Compatibility

The files provided are tested to work with:

• text-generation-webui using Loader: AutoAWQ.
• vLLM version 0.2.0 and later.
• Hugging Face Text Generation Inference (TGI) version 1.1.0 and later.
• Transformers version 4.35.0 and later.
• AutoAWQ version 0.1.1 and later.

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: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius

Thank you to all my generous patrons and donaters!

And thank you again to a16z for their generous grant.

Original model card: FBL's Juanako 7B V1

juanako-7b-v1

This model is a fine-tuned version of fblgit/zephyr-lora-dpo-b1 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.4594
• Rewards/chosen: -1.1095
• Rewards/rejected: -2.3132
• Rewards/accuracies: 0.7964
• Rewards/margins: 1.2037
• Logps/rejected: -220.0052
• Logps/chosen: -217.5506
• Logits/rejected: -2.5535
• Logits/chosen: -2.7973

** Please feel free to run more tests and commit the results. Also if you are interested to participate in UNA's paper research or GPU sponsorship **

Model description

It seems to outperforms the original Zephyr in most of the tasks.

I trained Juanako with the same datasets and trainer from alignment-handbook/zephyr-7b-sft-lora

• 1 epoch on DPO with transformers-UNA, the result is fblgit/zephyr-lora-dpo-b1 after merge using FastChat converter.
• finally 1 epoch on DPO with transformers-UNA to fblgit/zephyr-lora-dpo-b1.

Some other experiments were performed as well to test transformers-UNA capabilities on diverse scenarios and models.

This is a complete version of the model, the result of converting LoRa's

Intended uses & limitations

Research purposes.

Training and evaluation data

alignment-handbook DPO with UNA on top of the SFT lora.

Evaluation lm-evaluation-harness

GSM8K

hf (pretrained=/root/juanako-7b-v1-beta,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 3, batch_size: 4

Tasks	Version	Filter	Metric	Value		Stderr
gsm8k	Yaml	get-answer	exact_match	0.4556	±	0.0137

0-Shot

hf (pretrained=fblgit/juanako-7b-v1,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 0, batch_size: 8

Tasks	Version	Filter	Metric	Value		Stderr
arc_challenge	Yaml	none	acc	0.5691	±	0.0145
		none	acc_norm	0.6041	±	0.0143
arc_easy	Yaml	none	acc	0.8363	±	0.0076
		none	acc_norm	0.8161	±	0.0079
hellaswag	Yaml	none	acc	0.6554	±	0.0047
		none	acc_norm	0.8411	±	0.0036
boolq	Yaml	none	acc	0.8355	±	0.0065
lambada	N/A	none	perplexity	3.3607	±	0.1398
		none	acc	0.7309	±	0.0137
piqa	Yaml	none	acc	0.8194	±	0.0090
		none	acc_norm	0.8335	±	0.0087
sciq	Yaml	none	acc	0.9480	±	0.0070
		none	acc_norm	0.8960	±	0.0097
truthfulqa	N/A	none	bleu_max	26.0803	±	0.6528
- truthfulqa_mc1	Yaml	none	acc	0.4198	±	0.0173
- truthfulqa_mc2	Yaml	none	acc	0.5847	±	0.0153
winogrande	Yaml	none	acc	0.7609	±	0.0120

1-Shot

hf (pretrained=fblgit/juanako-7b-v1,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 1, batch_size: 8

Tasks	Version	Filter	Metric	Value		Stderr
arc_challenge	Yaml	none	acc	0.6084	±	0.0143
		none	acc_norm	0.6357	±	0.0141
arc_easy	Yaml	none	acc	0.8645	±	0.0070
		none	acc_norm	0.8645	±	0.0070
hellaswag	Yaml	none	acc	0.6475	±	0.0048
		none	acc_norm	0.8372	±	0.0037
boolq	Yaml	none	acc	0.8609	±	0.0061
lambada	N/A	none	perplexity	3.5484	±	0.1034
		none	acc	0.7207	±	0.0107
piqa	Yaml	none	acc	0.8259	±	0.0088
		none	acc_norm	0.8384	±	0.0086
sciq	Yaml	none	acc	0.9730	±	0.0051
		none	acc_norm	0.9740	±	0.0050
truthfulqa	N/A	none	bleu_max	18.9814	±	0.4805
		none	acc	0.4856	±	0.0521
- truthfulqa_mc1	Yaml	none	acc	0.4333	±	0.0173
- truthfulqa_mc2	Yaml	none	acc	0.5903	±	0.0153
winogrande	Yaml	none	acc	0.7609	±	0.0120

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 12
• gradient_accumulation_steps: 16
• total_train_batch_size: 192
• total_eval_batch_size: 12
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.01
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.4966	0.15	50	0.4893	-1.1759	-2.2914	0.7485	1.1155	-219.7872	-218.2148	-2.5450	-2.7884
0.4522	0.31	100	0.4808	-0.8099	-1.8893	0.7784	1.0794	-215.7659	-214.5544	-2.5644	-2.8095
0.5048	0.46	150	0.4706	-1.0526	-2.1412	0.7725	1.0887	-218.2852	-216.9814	-2.5638	-2.8089
0.4853	0.62	200	0.4640	-1.0787	-2.2821	0.7725	1.2034	-219.6941	-217.2426	-2.5460	-2.7891
0.4639	0.77	250	0.4636	-1.2348	-2.4583	0.8084	1.2235	-221.4559	-218.8034	-2.5533	-2.7970
0.4634	0.93	300	0.4601	-1.1370	-2.3243	0.7964	1.1873	-220.1163	-217.8257	-2.5540	-2.7977
-	1.00	300	0.4594	-1.1095	-2.3132	0.7964	1.2037	-220.0052	-217.5506	-2.5535	-2.7973

Framework versions

• Transformers 4.35.0-UNA
• Pytorch 2.1.0
• Datasets 2.14.6
• Tokenizers 0.14.1

MMLU Results

1-Shot

hf (pretrained=fblgit/juanako-7b-v1,load_in_4bit=False,dtype=float16), limit: None, num_fewshot: 1, batch_size: 1

Tasks	Version	Filter	Metric	Value		Stderr
mmlu	N/A	none	acc	0.6085	±	0.1321
- humanities	N/A	none	acc	0.5405	±	0.1478
- formal_logic	Yaml	none	acc	0.4206	±	0.0442
- high_school_european_history	Yaml	none	acc	0.7576	±	0.0335
- high_school_us_history	Yaml	none	acc	0.8186	±	0.0270
- high_school_world_history	Yaml	none	acc	0.7890	±	0.0266
- international_law	Yaml	none	acc	0.7438	±	0.0398
- jurisprudence	Yaml	none	acc	0.8056	±	0.0383
- logical_fallacies	Yaml	none	acc	0.7791	±	0.0326
- moral_disputes	Yaml	none	acc	0.7023	±	0.0246
- moral_scenarios	Yaml	none	acc	0.2145	±	0.0137
- philosophy	Yaml	none	acc	0.7074	±	0.0258
- prehistory	Yaml	none	acc	0.7377	±	0.0245
- professional_law	Yaml	none	acc	0.4361	±	0.0127
- world_religions	Yaml	none	acc	0.8421	±	0.0280
- other	N/A	none	acc	0.6894	±	0.1091
- business_ethics	Yaml	none	acc	0.5600	±	0.0499
- clinical_knowledge	Yaml	none	acc	0.6981	±	0.0283
- college_medicine	Yaml	none	acc	0.6185	±	0.0370
- global_facts	Yaml	none	acc	0.3300	±	0.0473
- human_aging	Yaml	none	acc	0.6726	±	0.0315
- management	Yaml	none	acc	0.8058	±	0.0392
- marketing	Yaml	none	acc	0.8419	±	0.0239
- medical_genetics	Yaml	none	acc	0.7200	±	0.0451
- miscellaneous	Yaml	none	acc	0.8033	±	0.0142
- nutrition	Yaml	none	acc	0.7288	±	0.0255
- professional_accounting	Yaml	none	acc	0.4929	±	0.0298
- professional_medicine	Yaml	none	acc	0.6801	±	0.0283
- virology	Yaml	none	acc	0.5000	±	0.0389
- social_sciences	N/A	none	acc	0.7195	±	0.0676
- econometrics	Yaml	none	acc	0.5000	±	0.0470
- high_school_geography	Yaml	none	acc	0.7879	±	0.0291
- high_school_government_and_politics	Yaml	none	acc	0.8601	±	0.0250
- high_school_macroeconomics	Yaml	none	acc	0.6231	±	0.0246
- high_school_microeconomics	Yaml	none	acc	0.6471	±	0.0310
- high_school_psychology	Yaml	none	acc	0.8000	±	0.0171
- human_sexuality	Yaml	none	acc	0.7557	±	0.0377
- professional_psychology	Yaml	none	acc	0.6552	±	0.0192
- public_relations	Yaml	none	acc	0.6636	±	0.0453
- security_studies	Yaml	none	acc	0.7184	±	0.0288
- sociology	Yaml	none	acc	0.8358	±	0.0262
- us_foreign_policy	Yaml	none	acc	0.8500	±	0.0359
- stem	N/A	none	acc	0.5217	±	0.1149
- abstract_algebra	Yaml	none	acc	0.3000	±	0.0461
- anatomy	Yaml	none	acc	0.6222	±	0.0419
- astronomy	Yaml	none	acc	0.6711	±	0.0382
- college_biology	Yaml	none	acc	0.7361	±	0.0369
- college_chemistry	Yaml	none	acc	0.4400	±	0.0499
- college_computer_science	Yaml	none	acc	0.5000	±	0.0503
- college_mathematics	Yaml	none	acc	0.3100	±	0.0465
- college_physics	Yaml	none	acc	0.4902	±	0.0497
- computer_security	Yaml	none	acc	0.7100	±	0.0456
- conceptual_physics	Yaml	none	acc	0.5362	±	0.0326
- electrical_engineering	Yaml	none	acc	0.5862	±	0.0410
- elementary_mathematics	Yaml	none	acc	0.4365	±	0.0255
- high_school_biology	Yaml	none	acc	0.7129	±	0.0257
- high_school_chemistry	Yaml	none	acc	0.5074	±	0.0352
- high_school_computer_science	Yaml	none	acc	0.6500	±	0.0479
- high_school_mathematics	Yaml	none	acc	0.3259	±	0.0286
- high_school_physics	Yaml	none	acc	0.3709	±	0.0394
- high_school_statistics	Yaml	none	acc	0.5139	±	0.0341
- machine_learning	Yaml	none	acc	0.5089	±	0.0475

Groups	Version	Filter	Metric	Value		Stderr
mmlu	N/A	none	acc	0.6085	±	0.1321
- humanities	N/A	none	acc	0.5405	±	0.1478
- other	N/A	none	acc	0.6894	±	0.1091
- social_sciences	N/A	none	acc	0.7195	±	0.0676
- stem	N/A	none	acc	0.5217	±	0.1149