alexbakers/llama_ipfs

LLaMA-IPFS Demo

This Streamlit app demonstrates how to use llama-cpp-python with IPFS models using the llama-ipfs integration.

Features

• —Load models directly from IPFS using their CID
• —Input custom context and questions
• —Adjust generation parameters (temperature, max tokens, etc.)
• —Get instant answers from an IPFS-hosted model

How It Works

The llama-ipfs package patches the llama-cpp-python library to:

1. 1.Recognize ipfs:// URIs as valid model identifiers
2. 2.Download model files from IPFS nodes or gateways
3. 3.Cache models locally for faster loading in subsequent runs

Model Used

This demo uses a GPT-2 (117M) model hosted on IPFS with CID: bafybeie7quk74kmqg34nl2ewdwmsrlvvt6heayien364gtu2x6g2qpznhq

Usage

1. 1.Enter or modify the context in the text area
2. 2.Type your question
3. 3.Adjust generation parameters if desired
4. 4.Click "Generate Answer" to get a response

Note: The first run may take longer as the model is downloaded from IPFS.

Requirements

• —streamlit
• —llama-cpp-python
• —huggingface-hub
• —llama-ipfs