AI Algorithm Engineer (AI Training/Inference Acceleration)

About the Role

We are collaborating with a leading global entity in information and communications technology (ICT) infrastructure and smart devices. The company delivers comprehensive AI solutions across various sectors including telecommunications, business enterprises, government institutions, and individual consumers worldwide. The position requires a highly skilled engineer to spearhead advancements in AI training and inference acceleration.

Key Responsibilities

• Lead the innovation and development of AI acceleration algorithms tailored to Agentic AI and multimodal applications, optimized for cutting-edge AI computing architectures to enhance computational efficiency and utilization.
• Manage the full lifecycle implementation of acceleration algorithms within proprietary AI frameworks and libraries, ensuring smooth integration and ongoing refinement calibrated by real-world performance data.
• Provide strategic technical insights and anticipate emerging trends in training and inference such as long-sequence modeling and sparsity, enabling proactive algorithm development that secures the platform's competitive edge.

Qualifications and Requirements

• Possess a Master's or PhD degree in Computer Science, Artificial Intelligence, Electronic Engineering, or an allied technical field.
• Demonstrate strong expertise in prevalent Large Language Model (LLM) structures and Mixture of Experts (MoE) models, with experience in large-scale training, fine-tuning, or deploying inference.
• Hands-on experience with AI acceleration technologies including zero-redundancy optimizers, distributed parallelism strategies (such as TP, PP, SP, VP, DP), communication compression, and memory efficiency techniques.
• Proficiency in developing high-performance attention kernels, KV-cache compression, and acceleration methods that consider model quantization and sparsity.
• Advanced Python skills paired with deep knowledge of state-of-the-art deep learning frameworks and acceleration libraries. Experience in kernel-level hardware optimization or operator tuning is highly desirable.
• Comprehensive understanding of hardware-software co-design, awareness of hardware attributes like memory bandwidth, compute cycles, and interconnects, with a successful record of enhancing the performance of models exceeding 100 billion parameters.
• Proven capability to optimize complex AI models in distributed or heterogeneous computing settings, effectively leveraging hardware-specific features to maximize algorithmic efficiency.

Additional Information

Interested candidates should apply promptly. Only shortlisted individuals will be contacted for further steps.

EA License No.: 01C4394

Data Privacy Notice: By submitting your CV and personal information, you consent to the collection, use, and disclosure of your data by PERSOL Singapore Pte Ltd and its affiliates according to their Privacy Policy.