全职 Artificial Intelligence Engineer Jobs, in Evolution Singapore - Ricebowl

About the Role

This role builds the foundation for production-grade distributed AI training at scale. You will design reusable training recipes, benchmarking frameworks, and evaluation standards that enable large customers to train and compare models efficiently across multi-node GPU clusters.

You’ll work closely with platform, orchestration, and application engineers to turn distributed training best practices into repeatable, customer-facing templates.

Job Details

• Build and maintain production-ready distributed training recipes using frameworks such as TorchTitan and Megatron-LM
• Define model scaling baselines and tuning guidance across GPU counts, parallelism strategies, and checkpointing patterns
• Design and run multi-node communication and performance benchmarks (throughput, MFU, cost, energy efficiency)
• Create standardized evaluation harnesses and offline benchmarking suites for model comparison
• Publish training efficiency playbooks and benchmark results to guide internal teams and customers

Job Requirements

• 5–7 years of hands-on experience with distributed ML training (PyTorch/JAX, FSDP, DeepSpeed, multi-node GPU systems)
• Deep expertise in GPU performance optimization, memory behavior, and NCCL communication patterns
• Proven ability to debug convergence issues and optimize large-scale training throughput
• Strong benchmarking discipline with experience designing controlled, repeatable experiments
• Practical knowledge of model parallelism trade-offs (FSDP, tensor, pipeline parallelism)