AI Research Engineer (Model Compression & Quantization)

Role overview Research and implement ways to make multimodal AI models smaller and more efficient while preserving their capabilities. The work covers language and vision-language models that process text, images, and audio, with the goal of enabling low-memory, low-latency inference on constrained devices such as smartphones.

Responsibilities - Apply low-bit and mixed-precision quantization to reduce model size and inference latency while maintaining output quality. - Use knowledge distillation to transfer capabilities from larger models to smaller models for multimodal tasks. - Prune redundant model parameters or attention heads to reduce computation while protecting task performance. - Build compression pipelines and define measures for accuracy, fidelity, memory use, throughput, and latency. - Investigate performance trade-offs and production inference bottlenecks, then refine techniques based on experimental results. - Document work for reproducibility and collaboration, and contribute research findings through technical publications.

Requirements - Degree in computer science or a related field; doctoral study in NLP, machine learning, or a related area is preferred. - Track record in AI research and development; strong publications at leading conferences are preferred. - Hands-on experience with PyTorch or an equivalent deep learning framework. - Practical knowledge of both quantization-aware training and post-training quantization. - Research or implementation experience with knowledge distillation and model pruning. - Strong understanding of neural networks, transformers, training, optimization, and fine-tuning.

Nice to have - C++ familiarity, especially for low-level quantization kernels or inference optimization.

Benefits and work setup - Remote work with an internationally distributed team. - Collaboration with colleagues working on AI and other technology areas.