Optimized open source AI models, fp16 and int8 (quantized) for vision, audio, and speech applications. Run models directly on real Qualcomm devices today.