v3.8.0
📦 gpt4allView on GitHub →
✨ 5 features🐛 4 fixes
Summary
This release introduces robust native support for the DeepSeek-R1 model family and completely overhauls the chat template parser for improved model compatibility. Several critical bugs related to the code interpreter and local server functionality have also been resolved.
✨ New Features
- Native support added for the DeepSeek-R1 family of distillations, including several new model variants.
- Reasoning (wrapped in "think" tags) is now displayed similarly to the Reasoner model when using DeepSeek-R1.
- Support for the DeepSeek-R1 Qwen pretokenizer has been added, resolving previous loading failures.
- DeepSeek-R1 models are now configured with a GPT4All-compatible prompt template by default.
- The template parser for chat templating has been completely replaced for much better compatibility with common models.
🐛 Bug Fixes
- Fixed an issue preventing the code interpreter from logging a single string in v3.7.0.
- Resolved UI freezing issue while the code interpreter is running a computation.
- Fixed a server issue preventing LocalDocs from being used after the first request since v3.5.0.
- System messages are now correctly hidden from the message history in the local server.