Spiralflow Data Chat

Spiralflow Data Chat is a chatbot designed to ingest, index, and interact with data using the Spiralflow framework. It can utilize multiple data sources to generate relevant context and provide accurate responses to user prompts. The chatbot takes into account the relevancy of each data source to ensure accurate and context-aware responses.

Features

• Utilize multiple data sources for context generation.
• Generate accurate and context-aware responses based on the provided context and data sources.

Prerequisites

Before using Spiralflow Data Chat, make sure to have the following prerequisites:

• Python 3.9 or later
• An OpenAI API key with access to GPT-3.5-turbo (or another suitable chat models)

Installation

1. Clone the repository:

bash

git clone https://github.com/Tiger767/spiralflow-data-chat.git
cd spiralflow-data-chat

2. Install the required Python packages:

bash

pip install spiralflow fastapi
conda install -c conda-forge pytorch faiss-cpu -y

Usage

To run the Spiralflow Data Chat chatbot website, run the below command and go to the respective website:

python main.py

To run the chatbot in the terminal, simply execute the chatbot script:

python chatbot.py

You can also use command-line arguments to customize the chatbot's behavior, such as:

• --memory: Memory to use.
• --memory_file: Memory file to use (default: "memory_default.pkl").
• --max_num_query_results: Maximum number of results for context (default: 14).
• --num_query_results: Number of memory queries (default: 20).
• --max_memory_context_tokens: Maximum number of tokens for the full context (default: 1500).
• --memory_score_threshold: Threshold for memory queries. A value of .8 is strict and .7 is loose (default: 0.7).
• --combine_threshold: Threshold for combining memory queries (default: 0.1).
• --summarize_context: Each document in context will be summarized, attempting to extract the relevant parts.
• --openai_chat_model: OpenAI chat model to use (default: "gpt-3.5-turbo").
• --max_num_prompt_tokens: Number of tokens a prompt can contain without having to break it up (default: 2000).
• --max_num_tokens_per_memory: Number of tokens a prompt can contain without having to break it up (default: 500).
• --temperature: Temperature for response generation (default: 0.3).
• --persona: Persona to use for response generation. The persona is one aspect in a response and so may not be stressed (default: "").
• --enable_chat_history: Enables queryable chat history so prompts can refer to previous prompts and responses.
• --max_chat_history_tokens: Number of tokens chat history can have. Any more will be truncated (default: 2000).
• --enable_prompt_response_memory: Enables memory of prompts and responses.
• --verbose: Increase output verbosity.

For example:

python chatbot.py --memory_file memory_default.pkl --temperature 0.1 --enable_chat_history

Ingesting Data with ingest.py

The ingest.py script helps you to ingest data and create memory for the SpiralFlow Data Chat chatbot. You can customize the data ingestion process by specifying the directory containing the data, the chunk size, overlap between chunks, input format, and postfix for the memory.

Usage

1. Replace the contents of the data folder with your data files.
2. Remove the placeholder memory_default.pkl file from the repository.
3. Run the ingest.py script:

python ingest.py

You can customize the data ingestion process using command-line arguments:

• -d, --directory: Directory containing the data (default: "data/").
• -l, --load: Load a previous memory and append to it.
• -pf, --postfix: Postfix for the memory (default: "default").
• -c, --chunk_size: Chunk size for text splitting (default: 240).
• -o, --chunk_overlap_factor: Overlap factor between chunks for text splitting (default: 1/3).
• --dry_run: Perform a dry run, stopping before embeddings are declared.

For example, to ingest data from a directory named my_data with a chunk size of 500, an overlap of 200, and an input format of "text", you can run:

python ingest.py -d my_data -c 500 -o 200

After running the ingest.py script, a new memory file will be generated based on your data. Make sure to update the --memory_file argument in the main.py script to use the new memory file.

License

Spiralflow Data Chat is released under the MIT License.