MeloTTS adds Japanese support and optimizes audio quality. English MeloTTS models for en-default and en-us are added, along with the Japanese MeloTTS model for ja-jp.
Whisper adds the small model. The tiny base model is updated to optimize the recognition accuracy for English and Japanese.
VLM adds smolvlm-256M and smolvlm-500M models.
LLM adds the qwen3-0.6b model.
ModuleLLM-OpenAI-Plugin optimizes STT and TTS APIs.
A benchmark test is added.
Pzmq adds pzmq_data message type.
StackFlow changes the message passing type.
LLM-YOLO, LLM-LLM, and LLM-VLM switch to asynchronous inference to optimize inference latency.
Added experimental support for Axera VIN in llm-camera.
Added experimental support for BSON in llm-sys.
Added new models to llm-llm.
Optimized LLM post-processing algorithms.
Fixed the issue of incorrect token truncation in LLM output.
Updated the internVL2.5-MPO model.
Added the Deepseek-r1-distill-qwen2.5-0.5b model.
Added a new prefill 256 LLM model. Warning: This upgrade package requires installing lib-llm_1.6-m5stack1_arm64.deb first before installing other model packages.