- Introduced new documentation for the llm-llm2 framework, detailing setup, inference, linking, unlinking, pausing, exiting, and task information retrieval.
- Added model configuration files for Qwen3 series models, including Qwen3-0.6B, Qwen3-1.7B, Qwen3.5-0.8B-Int4, Qwen3.5-2B-Int4, and Qwen3.5-4B-Int4.
- Updated main.cpp to support dynamic sampling parameter overrides during inference.
- Enhanced LLMPostprocess class to handle new sampling configurations and reset to defaults.
- Updated llm_pack.py to include new model entries for packaging.
- Updated VLMType to include PaddleOCRVL (ID 6).
- Implemented PaddleOCRVL image processing in image_processor.cpp and header.
- Enhanced disk cache loading to guard against corrupted files and added limits on key length and block count.
- Introduced port availability check in net_utils for better error handling.
- Updated vision module to handle PaddleOCR-VL input format and normalization.
- Improved error handling during cache save operations to prevent corruption.
- Adjusted mRoPE configuration to accommodate PaddleOCR-VL.
LittleMouse
dianjixz