v1.1.9

Added

• Added automatic GPU to CPU fallback for ONNX models when VRAM is insufficient.
• Added automatic max_length reduction when generator memory allocation fails.
• Added clear validation and error messages for incompatible ONNX models (e.g., transformers.js models missing genai_config.json).
• Added graceful handling of user cancellation during model loading in Module llama.cpp.
• Added macOS platform detection (is_macos(), is_apple_silicon()) to Module llama.cpp for optimal GPU configuration.
• Added cancel_loading() method to Module llama.cpp ModelHelper to support cancellation during model loading.
• Added GPU Layers setting to Module llama.cpp UI for configuring Metal GPU acceleration on macOS.
• Added pre-commit hooks configuration for linting, testing, and code validation.

Changed

• Refactored init_generator() to use _create_generator() helper for cleaner retry logic.
• Installation commands now use quoted syntax (pip install "notolog[llama]") for cross-platform shell compatibility.
• Module llama.cpp now automatically configures Metal GPU acceleration on Apple Silicon Macs (M1/M2/M3/M4).

Updated

• Updated FAQ with ONNX model compatibility and memory troubleshooting guidance.
• Improved the README.md with clearer installation and usage instructions.

Fixed

• Fixed ONNX module crash when GPU provider fails - now automatically falls back to CPU.
• Fixed ONNX generator memory allocation failures - now automatically retries with smaller max_length.
• Fixed Module llama.cpp logging errors when user closes dialog during model loading.
• Fixed inability to close app when Module llama.cpp model loading hangs - cancellation now properly signals the loading thread.