TheBloke/bagel-dpo-7B-v0.1-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)

---

Bagel DPO 7B v0.1 - AWQ

• Model creator: Jon Durbin
• Original model: Bagel DPO 7B v0.1

Description

This repo contains AWQ model files for Jon Durbin's Bagel DPO 7B v0.1.

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.

AWQ models are currently supported on Linux and Windows, with NVidia GPUs only. macOS users: please use GGUF models instead.

It is supported by:

• Text Generation Webui - using Loader: AutoAWQ
• vLLM - version 0.2.2 or later for support for all model types.
• 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
• Jon Durbin's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions

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

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/bagel-dpo-7B-v0.1-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: bagel-dpo-7B-v0.1-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/bagel-dpo-7B-v0.1-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'''Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{prompt}

### Response:
'''

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

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

llm = LLM(model="TheBloke/bagel-dpo-7B-v0.1-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/bagel-dpo-7B-v0.1-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'''Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{prompt}

### Response:
'''

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/bagel-dpo-7B-v0.1-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'''Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{prompt}

### Response:
'''

# 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: Michael Levine, 阿明, Trailburnt, Nikolai Manek, John Detwiler, Randy H, Will Dee, Sebastain Graf, NimbleBox.ai, Eugene Pentland, Emad Mostaque, Ai Maven, Jim Angel, Jeff Scroggin, Michael Davis, Manuel Alberto Morcote, Stephen Murray, Robert, Justin Joy, Luke @flexchar, Brandon Frisco, Elijah Stavena, S_X, Dan Guido, Undi ., Komninos Chatzipapas, Shadi, theTransient, Lone Striker, Raven Klaugh, jjj, Cap'n Zoog, Michel-Marie MAUDET (LINAGORA), Matthew Berman, David, Fen Risland, Omer Bin Jawed, Luke Pendergrass, Kalila, OG, Erik Bjäreholt, Rooh Singh, Joseph William Delisle, Dan Lewis, TL, John Villwock, AzureBlack, Brad, Pedro Madruga, Caitlyn Gatomon, K, jinyuan sun, Mano Prime, Alex, Jeffrey Morgan, Alicia Loh, Illia Dulskyi, Chadd, transmissions 11, fincy, Rainer Wilmers, ReadyPlayerEmma, knownsqashed, Mandus, biorpg, Deo Leter, Brandon Phillips, SuperWojo, Sean Connelly, Iucharbius, Jack West, Harry Royden McLaughlin, Nicholas, terasurfer, Vitor Caleffi, Duane Dunston, Johann-Peter Hartmann, David Ziegler, Olakabola, Ken Nordquist, Trenton Dambrowitz, Tom X Nguyen, Vadim, Ajan Kanaga, Leonard Tan, Clay Pascal, Alexandros Triantafyllidis, JM33133, Xule, vamX, ya boyyy, subjectnull, Talal Aujan, Alps Aficionado, wassieverse, Ari Malik, James Bentley, Woland, Spencer Kim, Michael Dempsey, Fred von Graf, Elle, zynix, William Richards, Stanislav Ovsiannikov, Edmond Seymore, Jonathan Leane, Martin Kemka, usrbinkat, Enrico Ros

Thank you to all my generous patrons and donaters!

And thank you again to a16z for their generous grant.

Original model card: Jon Durbin's Bagel DPO 7B v0.1

A bagel, with everything

Overview

This is the DPO'd version of https://huggingface.co/jondurbin/bagel-7b-v0.1

If you are getting too many AALLM or other refusals, even with explicitly human system prompts, you may want to try the non-DPO version.

Benchmarks

I ran these against the latest main branch of lm-evaluation-harness (and opencompass/FastChat for agieval and mt-bench), since batch size/etc effects score for some benchmarks.

model	arc_challenge	boolq	gsm8k	hellaswag	mmlu	openbookqa	piqa	truthful_qa	winogrande
bagel	0.6715	0.8813	0.5618	0.8397	0.6408	0.51	0.8406	0.6275	0.7561
openhermes-2.5	0.6476	0.8835	0.4852	0.8414	0.6347	0.498	0.8400	0.5295	0.7443

MT-Bench:

########## First turn ##########
                      score
model         turn         
bagel-7b-v0.1 1     7.60625

########## Second turn ##########
                      score
model         turn         
bagel-7b-v0.1 2     7.00625

########## Average ##########
                 score
model                 
bagel-7b-v0.1  7.30625

Data selection.

The first step in the process is creating a dataset. In this case, we're actually creating a composite dataset, consisting of both supervised fine-tuning data (SFT) and direct preference optimization (DPO) data.

All instruction data, that is, data that is not plain text (like project Gutenberg and items from Cinematika) or DPO, is converted into ShareGPT format so it's easier to work with.

See the corresponding code in bagel/data_sources/*.py for full implementation for each data source.

Deduplication is done by creating a uuid v5 of the instruction/text, then only adding items not previously seen (where datasets are loaded in order of the confidence score I assign them). This means that if an instruction is in data source "Foo" with confidence 4 as well as in data source "Bar" with confidence score 2, only the entry from "Foo" will be taken.

SFT data sources

Yes, you will see benchmark names in the list, but this only uses the train splits, and a decontamination by cosine similarity is performed at the end as a sanity check

• ai2_arc
  • Abstraction and reasoning dataset, useful in measuring "intelligence" to a certain extent.
• airoboros
  • Variety of categories of synthetic instructions generated by gpt-4.
• apps
  • Python coding dataset with 10k problems.
• belebele
  • Multi-lingual reading comprehension dataset.
• boolq
  • Corpus of yes/no questions (which can be surprisingly difficult for AI to answer apparently?)
• cinematika (instruction and plain text)
  • RP-style data synthesized from movie scripts so the model isn't quite as boring as it otherwise would be.
• drop
  • More reading comprehension.
• gutenberg (plain text)
  • Books/plain text, again to make the model less boring, only a handful of examples supported by chapterize
• lmsys_chat_1m (only gpt-4 items, also used for DPO)
  • Chats collected by the lmsys chat arena, containing a wide variety of chats with various models.
• mathinstruct
  • Composite dataset with a variety of math-related tasks and problem/question formats.
• mmlu
  • Massive Multitask Language Understanding - a wide variety of questions about various subject matters.
• natural_instructions
  • Millions of instructions from 1600+ task categories (sampled down substantially, stratified by task type)
• openbookqa
  • Question answering dataset.
• piqa
  • Phyiscal interaction question answering.
• python_alpaca
  • Python instruction response pairs, validated as functional.
• rosetta_code
  • Code problems and solutions in a variety of programming languages taken from rosettacode.org.
• slimorca
  • Collection of ~500k gpt-4 verified chats from OpenOrca.
• spider
  • SQL-targeted dataset.
• squad_v2
  • Contextual question answering (RAG).
• synthia
  • GPT-4 generated data using advanced prompting from Migel Tissera.
• winogrande
  • Fill in the blank style prompts.

DPO data sources

• airoboros 3.1 vs airoboros 2.2.1
  • The creative/writing tasks from airoboros-2.2.1 were re-generated using gpt4-0314 and a custom prompt to get longer, more creative, less clichè responses for airoboros 3.1, so we can use the shorter/boring version as the "rejected" value and the rerolled response as "chosen"
• helpsteer
  • Really neat dataset provided by the folks at NVidia with human annotation across a variety of metrics. Only items with the highest "correctness" value were used for DPO here, with the highest scoring output as "chosen" and random lower scoring value as "rejected"
• orca_dpo_pairs
  • Another interesting dataset by Intel, which provides various DPO pairs generated from prompts included in the SlimOrca dataset.
• toxic-dpo
  • highly toxic and potentially illegal content! De-censorship, for academic and lawful purposes only, of course. Generated by llama-2-70b via prompt engineering.
• truthy
  • DPO pairs meant to increase truthfulness of the model, e.g. common misconceptions, differentiate between AI assistants and roleplayed human in terms of corporeal awareness/locality/etc.
• ultrafeedback
  • One of the bits of magic behind the Zephyr model. Only the items with a chosen score of 8 or higher were included.

Only the train splits were used (if a split was provided), and an additional pass of decontamination is performed using approximate nearest neighbor search (via faiss).

Total dataset size

The deduplicated and decontamined list of instructions contains 1,671,822 items:

• 1,602,217 SFT/instructions
• 59,247 DPO pairs
• 1606 with both SFT and DPO data

Keep in mind, this number becomes 4x larger when applying the various prompt formats.

Prompt formatting

In sticking with the theme of the bagel, I didn't want to use a single prompt format, so I used 4 - vicuna, llama-2, alpaca, and chat-ml (sorta). I also didn't want to randomly select a single prompt format for each item (hoping each instruction would generalize more when used in a variety of prompt formats), so each instruction is actually converted into every prompt format.

This means each epoch of our fine-tune is really basically 4 epochs. So, for the fine-tunes, I would recommend only doing 1 epoch (or 0.75 epochs). I am testing with a single epoch using a relatively low learning rate.

Alpaca (sort of)

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

### Instruction:
{system prompt, if provided}
{instruction}

### Response:

The main difference here is that because of the dataset formatting and variety of data sources, it would have been much to tedious to add an ### Input: block, so the inputs are just in the instruction section.

Vicuna

{system prompt, if provided, randomly defaulting to "A chat between a user and an unbiased, uncensored assistant."}
USER: {instruction}
ASSISTANT: 

ChatML (sort of)

I don't really understand the point of having special tokens for <|im_start|> and <|im_end|>, because in practice they just act as BOS and EOS tokens (but, please correct me if I'm wrong).

So, instead of:

{bos}<|im_start|>{role}
{text}
<|im_end|>{eos}

I just changed it to:

{bos}{role}
{text}
{eos}

In practice, this would mean tokenization code like such:

tokenizer = AutoTokenizer.from_pretrained('mistralai/mistral-7b-v0.1')

input_str = f"""system
You are a goat.
{tokenizer.eos_token}
{tokenizer.bos_token}user
Tell me how to fry an egg.
{tokenizer.eos_token}
{tokenizer.bos_token}assistant
"""

inputs = tokenizer(input_str, return_tensors="pt")

If you really want to use <|im_start|> and <|im_end|>, just update your tokenizer_config.json to use <|im_start|> instead of <s> and <|im_end|> instead of </s> and when tokenizing. And if you still don't like what I've done to this chat-ml-ish format, feel free to cry into your pillow or fork the code and do a new fine-tune.

Llama-2 chat

[INST] <<SYS>>
{system}
<</SYS>>

{instruction} [/INST]

Fine tuning

SFT phase

An example for mistral-7b:

Note: I actually used my fork of qlora's train.py for this, but I'm porting it to a minified version here, not tested yet!

More notes: I stopped the SFT phase around 50% because of budget constraints.

export BASE_DIR=/workspace
export WANDB_API_KEY=[redacted]
export WANDB_PROJECT=bagel-7b-v0.1

# Run the pretraining.
accelerate launch bagel/tune/sft.py \
  --model_name_or_path $BASE_DIR/mistral-7b \
  --final_output_dir $BASE_DIR/$WANDB_PROJECT \
  --output_dir $BASE_DIR/$WANDB_PROJECT-workdir \
  --num_train_epochs 1 \
  --logging_steps 1 \
  --save_strategy steps \
  --save_steps 200 \
  --save_total_limit 5 \
  --data_seed 42 \
  --evaluation_strategy steps \
  --eval_dataset_size 0.0006 \
  --eval_steps 200 \
  --max_new_tokens 4096 \
  --dataloader_num_workers 3 \
  --logging_strategy steps \
  --remove_unused_columns False \
  --do_train \
  --full_finetune \
  --bf16 \
  --bits 16 \
  --optim adamw_torch \
  --lr_scheduler_type linear \
  --dataset $BASE_DIR/bagel/bagel-input-output-v0.1.parquet \
  --dataset_format input-output \
  --model_max_len 4096 \
  --per_device_train_batch_size 8 \
  --learning_rate 3.5e-7 \
  --warmup_ratio 0.005 \
  --adam_beta2 0.999 \
  --max_grad_norm 0.3 \
  --weight_decay 0.001 \
  --seed 42 \
  --report_to wandb \
  --gradient_checkpointing True \
  --gradient_accumulation_steps 4 \
  --skip_excess_length False \
  --ddp_find_unused_parameters False \
  --use_flash_attention_2 \
  --deepspeed deepspeed.json

Deepspeed configuration:

{
  "gradient_accumulation_steps": "auto",
  "gradient_clipping": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "bf16": {
    "enabled": true
  },
  "zero_optimization": {
    "stage": 2,
    "contiguous_gradients": true,
    "overlap_comm": true,
    "reduce_scatter": true,
    "reduce_bucket_size": 5e8,
    "allgather_bucket_size": 5e8
  }
}

DPO phase

An example of the DPO phase for mistral-7b (requires first running the SFT):

export BASE_DIR=/mnt/data
export WANDB_API_KEY=[redacted]
export WANDB_PROJECT=bagel-dpo-7b-v0.1

accelerate launch bagel/tune/dpo.py \
  --model_name_or_path bagel-7b-v0.1 \
  --learning_rate 3e-7 \
  --per_device_train_batch_size 2 \
  --gradient_accumulation_steps 4 \
  --max_length 4096 \
  --max_prompt_length 1024 \
  --max_target_length 3092 \
  --num_train_epochs 3 \
  --report_to wandb \
  --gradient_checkpointing true \
  --use_flash_attention_2 true \
  --dataset $BASE_DIR/bagel/bagel-dpo-v0.1.parquet \
  --eval_steps 5 \
  --eval_dataset_size 0.03 \
  --workdir $BASE_DIR/$WANDB_PROJECT-workdir \
  --output_dir $BASE_DIR/$WANDB_PROJECT \
  --deepspeed deepspeed.json \
  --save_steps 25 \
  --save_total_limit 5