DarkUnitLab

Research and engineering for reliable, efficient AI systems. We focus on the path from model experimentation to practical deployment: training, evaluation, quantization, inference, and the systems that connect them.

Focus Areas

• Model Training and Adaptation Parameter-efficient fine-tuning, data preparation, reproducible experiments, and language and speech model adaptation.
• Inference and AI Infrastructure Quantization, model serving, runtime compatibility, and deployment across GPUs and Ascend NPUs.
• Multimodal Intelligence Vision-language models, video understanding, evidence-grounded reasoning, and retrieval-augmented generation.
• Language and Speech Cantonese and multilingual language modeling, automatic speech recognition, and evaluations that reflect real-world usage.