RAFT

RAFT is a recipe to adapting LLMs to domain-specific RAG. You can learn more in our release-blogs here and here. RAFT takes an input document from the user and creates a dataset using the document, consisting of synthetically generated { question, answer, documents } triplets. The dataset can then be used to fine-tune models for improved question-answering and retrieval.

The input data from the user can be either a general text document (pdf, json, or txt) for general QA or an API documentation in the API Zoo JSONL format for API calling.

Dev environment with Codespaces

Everything is setup automatically in the dev container, open a terminal into the raft folder:

Note: The raft virtual env will be activated in your shell when entering into the raft folder.

Install Dependencies

Dependencies can be installed using the following command:

pip install -r requirements.txt

Arguments:

• --datapath - if a file, the path at which the document is located. If a folder, the path at which to load all documents
• --output - the path at which to save the dataset
• --output-format - the format of the output dataset. Defaults to hf for HuggingFace. Can be one of hf, completion, chat, eval.
• --output-type - the type of the output dataset file. Defaults to jsonl. Can be one of jsonl, parquet.
• --output-chat-system-prompt - The system prompt to use when the output format is chat. Optional.
• --output-completion-prompt-column - The column (json field name) for the prompt / instruction when using the completion output format. Defaults to prompt.
• --output-completion-completion-column - The column (json field name) for the completion when using the completion output format. Defaults to completion.
• --distractors - the number of distractor documents to include per data point / triplet
• --doctype - the type of the document, must be one of the accepted doctypes
  • currently accepted doctypes: pdf, txt, json, api
  • documents in json format must have a "text" attribute containing the content from which chunks are extracted
  • documents in api format must follow the API json format detailed in the Gorilla API Store
• --p - the percentage of including the oracle documents in the context
• --chunk_size - the size of each chunk in number of tokens
• --questions - the number of data points / triplets to generate per chunk
• --openai_key - your OpenAI key used to make queries to GPT-3.5 or GPT-4
• --embedding-model - The embedding model to use to encode documents chunks. Defaults to text-embedding-ada-002.
• --completion-model - The model to use to generate questions and answers. Defaults to gpt-4.
• --system-prompt-key - The system prompt key to use to generate the dataset. Defaults to gpt. Can by one of gpt, llama.
• --workers - The number of worker threads to use to generate the dataset. Defaults to 2.
• --auto-clean-checkpoints - Whether to auto clean the checkpoints after the dataset is generated. Defaults to false.

Note: The --fast mode flag has been removed, checkpointing is now always active.

Usage

Usage with OpenAI API

Run the following command with your desired arguments to generate the dataset.

python3 raft.py \
  --datapath PATH_TO_DATA \
  --output OUTPUT_PATH \
  --output-format hf \ # or completion or chat
  --distractors 3 \
  --p 1.0 \
  --doctype pdf \
  --chunk_size 512 \
  --questions 5 \
  --openai_key YOUR_OPENAI_KEY

Note: As an alternative to passing the OpenAI key with the --openai_key argument, you also store the standard OpenAI environment variables in a file called .env like so. All standard OpenAI env variables are supported.

# OpenAI
OPENAI_API_KEY=<replace_me>

raft.py does the following:

• Takes a document located at PATH_TO_DATA, breaks it into chunks of size chunk_size tokens if the data is a pdf, json, or txt, or chunks of one API endpoint if the data is an API documentation, as denoted by doctype.
• For each chunk, uses GPT-4 to synthetically generate questions question-answer pairs and adds distractors distractor chunks to each pair, creating {Q, A, D} triplets. Each triplet represents one datapoint in the dataset, where Q is the question/use-case, A is the answer, and D is the relevant chunk + distractor chunks.
• Each data point / triplet also contains other attributes (e.g. metadata), such as id, type, and cot_answer.
• Uses the HuggingFace Dataset API to create a dataset from all triplets and saves it at OUTPUT_PATH in the .arrow and .jsonl formats.

Usage with Azure OpenAI API

Create a file .env like so. All standard Azure OpenAI environment variables are supported.

# Azure OpenAI API
AZURE_OPENAI_ENDPOINT=https://<endpoint_sub_domain>.openai.azure.com/
AZURE_OPENAI_API_KEY=<replace_me>
OPENAI_API_VERSION=2023-05-15

Note: make sure your strip the path from the endpoint and keep just the domain. The full base URL will automatically be built based on the other env variables.

In addition, if you used non default Azure OpenAI deployment names, you'll need to specify them using the following CLI arguments:

--completion-model my-gpt-deployment-name
--embedding-model my-ada-deployment-name

Configuring different endpoints for the completion and embedding models

When using a non OpenAI endpoint, it is often the case that the endpoints for the embedding and completion models are different.

In that situation, it is possible to override default OpenAI and Azure OpenAI environment variables with COMPLETION_ or EMBEDDING_ prefixed environment variables. Here is an example:

# Llama 3 70b Instruct completion model
# Uses an OpenAI v1 compatible endpoint on Azure MaaS

COMPLETION_OPENAI_BASE_URL=https://Meta-Llama-3-70B-Instruct-<replace_me>-serverless.eastus2.inference.ai.azure.com/v1
COMPLETION_OPENAI_API_KEY=<replace_me>

# Ada 2 embedding model
# Uses an Azure OpenAI endpoint

EMBEDDING_AZURE_OPENAI_ENDPOINT=https://<replace_me>.openai.azure.com/
EMBEDDING_AZURE_OPENAI_API_KEY=<replace_me>
EMBEDDING_OPENAI_API_VERSION=<replace_me>

Running the raft.py CLI will look like:

cd raft
python3 raft.py \
    --datapath $PWD/sample_data/UC_Berkeley.pdf \
    --output $PWD/output \
    --distractors 3 \
    --doctype pdf \
    --chunk_size 512 \
    --questions 5 \
    --completion-model Meta-Llama-3-70B-Instruct-<replace_me> \
    --embedding-model text-embedding-ada-002

Note: The --completion-model and --embedding-model in the case of an Azure OpenAI endpoint must be set to the deployment name.

Example Usage

This details the command and process used to generate the example dataset found in ./sample_ds4. The document is a pdf of the Wikipedia page on the United States of America.

python3 raft.py --datapath sample_data/United_States_PDF.pdf --output ./sample_ds4 --distractors 4 --doctype pdf --chunk_size 512 --questions 5 --openai_key OPENAI_KEY

Usage with Completely locally using hugging-face models

This details the command and process used to generate the example dataset found in ./sample_ds4. The document is a pdf of the Wikipedia page on the United States of America. To run the script completely locally use

python3 raft_local.py --datapath sample_data/UC_Berkeley_short.pdf --output ./sample_ds4 --chunk_size 512 --questions 5 --doctype pdf --fast

1. Chunk generation

RAFT takes pdf and divides text into chunks of size 512 tokens. A sample chunk:

 "[CLS] United States of America Flag Coat of arms Motto : \" In God We Trust \" [ 1 ] Other traditional mottos : [ 2 ] \" E pluribus unum \" ( Latin ) \" Out of many, one \" \" Annuit cœptis \" ( Latin ) \" Providence favors our undertakings \" \" Novus ordo seclorum \" ( Latin ) \" New order of the ages \" Anthem : \" The Star - Spangled Banner \" [ 3 ] United States The United States of America ( USA or U. S. A. ), commonly know n as the United States ( US or U. S. ) or America, is a country primarily located in North America, between Canada and Mexico. It is a liberal democracy and republic of 50 federated states, a federal capital district ( Washington, D. C. ), and 326 Indian reservations that overlap with state boundaries. Outside the union of states, it asserts sovereignty over five major unincorporated island territories and various uninhabited islands. [ i ] The country has the world\'s third - largest land area, [ c ] largest maritime exclusive economic zone, and the third - largest population ( over 334 million ). [ j ] The federal government uses a presidential system with three separate branches : legislative, executive, and judicial. American territory was first settled by Paleo - Indians who migrated across the Bering land bridge over 12, 000 years ago. Colonization by the British began in 1607. Thirteen colonies eventually rebelled against the British Crown over taxation and political representation, declaring independence on July 4, 1776. Their victory in the American Revolutionary War ( 1775 – 83 ) resulted in a confederation of states before the U. S. Constitution and Bill of Rights were ratified. The young nation continued to acquire neighbor ing territories and spanned North America by the late 1840s. Longstanding disagreements over slavery led to the secession of the southern Confederate States of America, which were defeated by the remaining Union in the American Civil War ( 1861 – 65 ). Slavery was abolished, but discriminatory laws persisted in the South. By 1900, rapid industrialization established the United States as a great power and the world\'s largest economy. Following the Japanese attack on Pearl Harbor in December 1941, the United States joined the Allies of World War II. After their victory, it competed against the Soviet Union for dominance in nuclear and conventional"

2. Question and answer generation

RAFT then uses GPT-4 to generate 5 questions per chunk as well as the label (answer) for each question. Proceeding with the previous example chunk:

Questions:

['What is the official motto of the United States of America?',
  'How many states are there in the United States of America?',
  'Which territories does the United States claim sovereignty over, outside the union of states?',
  'When did the thirteen colonies declare independence from the British Crown?',
  'What caused the secession of the southern Confederate States of America?']

Answers:

['"In God We Trust"',
 '50 federated states',
 'Five major unincorporated island territories.',
 'July 4, 1776',
 'Disagreements over slavery']

3. Append distractor documents

For each question-answer pair, append 4 randomly selected chunks as distractor documents to form the {Q, A, D} triplet. Proceeding with the current example, a {Q, A, D} triplet, or one datapoint, would look like:

{
  'id': 'seed_task_0', 
  'type': 'general', 
  'question': 'What is the official motto of the United States of America?', 
  'context': {
    'sentences': [
      ["the Gulf of Mexico are prone to hurricanes, ... and enforces the Act. [ 189 ] As of 2022, the U. S",
    "energy from fossil fuel and the largest ... there are 19, 969 airports in the U. S., of which 5, 193 are designated",
    'weaponry, ideology, and international i... and is a permanent member of the UN Security Council. The first documentary evidence of the phrase " United States',
    '[CLS] United States of America Flag Coat of arms ... dominance in nuclear and conventional',
    '##om ic soft pow er. [ 405 ] [ 406 ] Nearly all present ... rights in the United States are advanced by global standards.']
    ],
    'title': [
      ['placeholder_title',
      'placeholder_title',
      'placeholder_title',
      'placeholder_title',
      'placeholder_title']
    ]
  },
  'answer': '"In God We Trust"',
  'cot_answer': None
}

4. Generate and save dataset

RAFT repeats steps 2 and 3 for each chunk and saves the dataset to the path specified by the --output argument.

5. Convert the dataset to the format expected for fine tuning

If you specified the --output-format completion or --output-format chat argument for the raft.py script, you can skip this part.

Otherwise, you need to convert the dataset to the format expected for fine-tuning a completion model in Azure with the following command:

python3 format.py --input output/data-00000-of-00001.arrow --output output.completion.jsonl --output-format completion

Note: the format.py script also has its own help

python3 format.py --help

usage: format.py [-h] --input INPUT [--input-type {arrow,jsonl}] --output OUTPUT --output-format {hf,completion,chat,eval} [--output-type {parquet,jsonl}] [--output-chat-system-prompt OUTPUT_CHAT_SYSTEM_PROMPT] [--output-completion-prompt-column OUTPUT_COMPLETION_PROMPT_COLUMN] [--output-completion-completion-column OUTPUT_COMPLETION_COMPLETION_COLUMN] [--output-completion-stop OUTPUT_COMPLETION_STOP]

options:
  -h, --help            show this help message and exit
  --input INPUT         Input HuggingFace dataset file (default: None)
  --input-type {arrow,jsonl}
                        Format of the input dataset. Defaults to arrow. (default: arrow)
  --output OUTPUT       Output file (default: None)
  --output-format {hf,completion,chat,eval}
                        Format to convert the dataset to (default: None)
  --output-type {parquet,jsonl}
                        Type to export the dataset to. Defaults to jsonl. (default: jsonl)
  --output-chat-system-prompt OUTPUT_CHAT_SYSTEM_PROMPT
                        The system prompt to use when the output format is chat (default: None)
  --output-completion-prompt-column OUTPUT_COMPLETION_PROMPT_COLUMN
                        The prompt column name to use for the completion format (default: prompt)
  --output-completion-completion-column OUTPUT_COMPLETION_COMPLETION_COLUMN
                        The completion column name to use for the completion format (default: completion)
  --output-completion-stop OUTPUT_COMPLETION_STOP
                        The stop keyword to use for the completion format (default: <STOP>)

Note: If fine tuning a chat model, then you need to use --output-format chat and optionally add the --output-chat-system-prompt parameter to configure the system prompt included in the dataset.

6. Finetune your own model on Microsoft AI Studio

Once the dataset is prepared, follow the instructions in azure-ai-studio-ft/howto.md to finetune and deploy your own RAFT model. Make sure to use prompt as input and completion as output when fine tuning a completion model and the messages column as input when fine tuning a chat model.

7. Evaluate RAFT model

After deploying your model in AI Studio, use command to evaluate the RAFT model. Make sure to fill in base_url, api_key and model_name in the eval.py, these can be found in the AI Studio.

python3 eval.py --question-file YOUR_EVAL_FILE.jsonl --answer-file YOUR_ANSWER_FILE

The YOUR_EVAL_FILE.jsonl is in the format where

{
  'instruction': '<DOCUMENT> document1 </DOCUMENT>\n<DOCUMENT> document2 </DOCUMENT> ...\n{question}",
  'gold_answer': '{answer}'
}