TL;DR

GPT4All now supports connecting to and using remote models from Groq, OpenAI, and Mistral, expanding access to powerful language models.

New

• Remote Model Support: Easily configure and use models hosted by Groq, OpenAI, and Mistral (cloud-based AI services).
• Granite Model: Added support for the new Granite model.

Fixes Worth Knowing

• Whitespace: Improved output formatting for DeepSeek-R1 models.
• Stability: Resolved several crash issues.

Before You Upgrade

Ensure your GPU meets the CUDA 5.0 compatibility requirement if you intend to use the CUDA backend.

TL;DR

GPT4All now includes a beta web search feature (internet access) powered by the Llama 3.1 model, allowing it to answer questions using current information.

Breaking

• Model Renaming: Rename Meta-Llama-3.1-8B-Instruct-128k-Q4_0.gguf to Meta-Llama-3.1-8B-Instruct-Q4_0.gguf.
• Prompt/System Template Update: Update prompt and system templates with provided code snippets.

New

• Web Search: Access the internet to provide more informed responses. (Requires setup - see link in release notes)

Fixes Worth Knowing

• Improved web search context injection and source display.
• Fixed RoPE scaling issue with Llama 3.1.

Before You Upgrade

1. Rename the specified model file.
2. Update your prompt and system templates with the provided code. See the wiki for detailed instructions: https://github.com/nomic-ai/gpt4all/wiki/Web-Search-Beta-Release

TL;DR

GPT4All now supports NVIDIA GPUs (CUDA) via llama.cpp, significantly accelerating prompt processing and generation on compatible hardware.

Breaking

• Unsupported models (Mamba, Persimmon, PLaMo) have been removed from the allowed list.

New

• CUDA Support: Utilize NVIDIA GPUs for faster performance with compatible model types.
• InternLM Models: Added support for the InternLM family of models.

Fixes Worth Knowing

• Message sending is now blocked while the LLM is responding.
• Improved chat title generation quality.
• Corrected model loading progress display.
• Resolved several memory leaks for improved stability.

Before You Upgrade

If you have an NVIDIA GPU, ensure you have the latest drivers installed to take advantage of the CUDA backend. You can select it in the Settings menu.

TL;DR

GPT4All now supports NVIDIA GPUs (CUDA) via llama.cpp, significantly improving performance for compatible models and quantization types.

Breaking

• Removed support for Mamba, Persimmon, and PLaMo models (less common model architectures).

New

• CUDA Support: Utilize NVIDIA GPUs for faster prompt processing and generation.
• InternLM Models: Added support for the InternLM family of models (open-source language models).

Fixes Worth Knowing

• Message sending is now blocked while the LLM is responding.
• Improved chat title generation quality.
• Corrected model loading progress display.

Before You Upgrade

• If you have an NVIDIA GPU, ensure you have the latest drivers installed to take advantage of the CUDA backend. You can select it in Settings.