A recent tutorial details how to deploy Qwen3-ASR-1.7B and Qwen3.5-9B on a single Haiguang DCU K100 GPU to create a voice-controlled file search system. The setup uses speech recognition to convert voice commands into text queries, which are then processed by the LLM to retrieve files intelligently. This integration demonstrates the growing capability of domestic Chinese hardware to run modern AI models, offering an alternative to NVIDIA-based solutions. For developers and engineers working with voice applications or exploring Chinese GPU ecosystems, this approach provides a practical reference for building similar systems. The tutorial covers model loading, inference optimization, and integration with file indexing, though the full step-by-step guide is best followed with caution due to potential copyright issues. The key takeaway is the feasibility of combining ASR and LLMs on a single GPU for real-time voice-driven tasks, which could inspire further innovation in voice-controlled productivity tools.
Deploying Qwen3-ASR and Qwen3.5-9B on a Haiguang DCU K100 for voice-driven file retrieval showcases the potential of domestic Chinese GPUs for AI workloads.