h2ogpt System
A h2ogpt System is an open source project for querying and summarizing documents or chatting with private GPT language models locally.
- See: H2O.ai, H2O LLM Eval, Chatbot UI.
References
2023
- https://github.com/h2oai/h2ogpt
- SUMMARY:
- h2ogpt is an open source project for querying and summarizing documents or chatting with private GPT language models locally.
- It supports a variety of models including LLaMa2, Falcon, Vicuna, WizardLM, GPT4ALL.
- Both GPU (using HF and LLaMa.cpp models) and CPU (using HF, LLaMa.cpp, GPT4ALL) modes are available.
- Provides a UI and CLI interface for interacting with the models.
- The UI allows uploading and viewing documents, downloading models, controlling LLaMa.cpp models.
- Has support for Linux, Docker, MacOS and Windows.
- Easy one-click installer available for Windows.
- Allows creating a private offline database of documents like PDFs, Word, Excel etc using accurate embeddings like instructor-large.
- Maintains quality through parallel summarization and over 1000 unit and integration tests.
- Integrates with inference servers like HF Transformers Inference API, Gradio, OpenAI API.
- Provides an OpenAI-compliant Python client API for controlling models programatically.
- Key features include document summarization, chat, handling images, code completion, reasoning abilities.
- Roadmap includes better math / code capabilities, integration in downstream apps, complementing with search APIs, distributed training of larger models.
- Provides instructions and support for fine-tuning models on custom data.
- Has active community on Discord.
- Developed by H2O.ai makers of various other AI platforms like H2O Driverless AI, H2O Wave etc.
- SUMMARY: