mirror of
https://github.com/nocobase/nocobase.git
synced 2026-09-19 02:23:00 +08:00
ReLaMi96
9f0409aa2d
* feat: add Ollama LLM provider support - Add @langchain/ollama dependency to plugin-ai package - Implement OllamaProvider with chat and embedding support - Add dynamic model listing from Ollama API - Add client-side UI components for Ollama configuration: - ProviderSettings: Base URL configuration - ModelSettings: Model selection with proper form schema integration - Parameter controls: temperature, topP, topK, numPredict (maxTokens) - Register Ollama provider in both server and client plugins - Support for vision models (llava, bakllava) with image attachments - Default base URL: http://localhost:11434 - Use empty static model list to force dynamic fetching from Ollama instance * fix: improve Ollama tool calling behavior - Change default model from granite4:tiny-h to mistral-nemo:12b (better trained for tool/function calling) - Add format: undefined to prevent JSON formatting interference - Add keepAlive: 5m to keep model loaded for better performance - Addresses issue where smaller models repeatedly call the same tools Note: For reliable tool calling with Ollama, recommend using: - llama3.1:8b or larger - mistral-nemo - qwen2.5 Avoid models under 7B parameters for tool calling tasks.