pxgpt 0.0.0rc0

Your personal, powerful and private GPT.

GPT: Your personal, powerful and private GPT

Features

• Ingest of your own documents and talk to them.
• Store your data locally on your device.
• Choose from a variety of models, including OpenAI.
• Support conversation history and memory.
• Switch between profiles with different settings.
• Support both web interface and command line interface.

Installation

# With all supported models
$ pip install -U pxgpt[all]

# With support for GPT4all only
$ pip install -U pxgpt[gpt4all]

# With support for llama-cpp only
$ pip install -U pxgpt[llama-cpp]

# With support for openai only
$ pip install -U pxgpt[openai]

Usage

Chat from CLI

$ 23:45:06 ❯ pxgpt chat

Welcome to chat via the pxgpt CLI v0.0.0!
Hit 'Ctrl+c' or 'Ctrl+d' to exit.
[2023-07-18 23:45:09,969] INFO Creating LLM (GPT4All)
Found model file at  models/gpt4all/ggml-gpt4all-j-v1.3-groovy.bin
gptj_model_load: loading model from 'models/gpt4all/ggml-gpt4all-j-v1.3-groovy.bin' - please wait ...
gptj_model_load: n_vocab = 50400
gptj_model_load: n_ctx   = 2048
gptj_model_load: n_embd  = 4096
gptj_model_load: n_head  = 16
gptj_model_load: n_layer = 28
gptj_model_load: n_rot   = 64
gptj_model_load: f16     = 2
gptj_model_load: ggml ctx size = 5401.45 MB
gptj_model_load: kv self size  =  896.00 MB
gptj_model_load: ................................... done
gptj_model_load: model size =  3609.38 MB / num tensors = 285
[2023-07-18 23:45:11,741] INFO Entering chat mode
Loaded conversation:  2023-07-16_23-11-27
Use the up/down arrow keys to navigate history.
Use /help to see the list of commands.

>>> Hello?
  Hi there! How may I help you today?

>>> /help
Commands:
  - /help: List the commands
  - /new: Start a new conversation
  - /switch: Switch to a conversation
  - /list: List all conversations
  - /path: Show the path of the current conversation file
  - /delete: Delete a conversation
  - /rename: Rename a conversation
  - /ingest: Ingest documents from the source directory
  - /docs: List ingested and uningested documents in the source directory
  - /exit: Exit the CLI

>>>

Chat from the web interface

$ pxgpt serve
# Open http://localhost:7758 in your browser

Configuration

The configuration files are loaded from the following paths:

• ~/.config/pxgpt/config.yml
• ~/.pxgpt.config.yml
• ./.pxgpt.config.yml

Profiles

Note that you need to define profiles in the configuration file. For example:

openai:  # The profile
    model:
        type: ChatOpenAI

The configuration items are inherited from the default profile. For example:

default:
    credentials:
        openai_api_key: sk-xxxxxxxxxxx

openai:
    model:
        type: ChatOpenAI

Then when you use openai profile, the configurations are expanded as:

openai:
    credentials:
        openai_api_key: sk-xxxxxxxxxxx
    model:
        type: ChatOpenAI

Higher-level configurations override lower-level configurations. For example:

If you define the default profile in ~/.config/pxgpt/config.yml and the openai profile in ./.pxgpt.config.yml, then the openai profile will inherit the default profile, as well.

Configuration items

• log_level: The log level for the logger in your teminal
• history_directory: The directory to store the conversation history
• history_into_memory: Whether to load the conversation history into memory
  • You can turn this off if you are using small models
• credentials: The credentials for the models. For example, for OpenAI, you need to provide the openai_api_key.
• model: Type of the model and arguments for it.
  • type: The type of the model, supported models are: GPT4All, LlamaCpp, ChatOpenAI and OpenAI
  • <other>: The arguments for the model. Passed to langchain llms.
    • For GPT4All, you can pass the arguments listed in here.
    • For LlamaCpp, you can pass the arguments listed in here.
    • For ChatOpenAI, you can pass the arguments listed in here.
    • For OpenAI, you can pass the arguments listed in here.
• qmodel: The arguments for model used to condense questions
  • type: Same as model.type, with and Echo model added, which is useful for models that don't do question condensing very well.
  • <other>: Same as model.<other>.
• ingest: The arguments for the ingestion.
  • source_directory: The directory to ingest documents from.
    • If not provided, we will enter the chat mode.
  • persist_directory: The directory to save the vectorstore database.
    • If not provided, will use <source_directory>/.pxgpt-<model>-db.
  • target_source_chunks: The number of chunks to return against the query.
  • n_workers: The number of workers to use for ingestion.
  • chunk_size and chunk_overlap: The chunk size and overlap for the ingestion.
• embeddings: The arguments for the embeddings.
  • For GPT4All, you can pass the arguments listed in here.
  • For LlamaCpp, you can pass the arguments listed in here.
  • For OpenAI or ChatOpenAI, you can pass the arguments listed in here.

Ingest documents

$ pxgpt ingest  # default profile
$ pxgpt ingest --profile openai-docs
# Will ingest documents under `ingest.source_directory` under `openai-docs` profile

Credits

pxgpt is Inspired by privateGPT, with the addition of openai API support, history and memory support, and a web interface.

TODO

• Support ingestion management (upload/download/delete/ingest documents) from the web interface
• Support profile management (add/remove/modify) from the web interface
• Build a docker image
• Support more models

Q & A

QA.md