Member of Technical Staff - Efficient ML

Tech Stack

Responsibilities

• Optimize training efficiency using techniques like dataloaders, fusion, activation rematerialization, and gradient checkpointing.
• Enhance GPU and kernel performance through Nsight profiling, Triton/CUDA kernels, fused operations, and Flash-attention style speedups.
• Implement inference optimizations including low-latency serving, continuous batching, speculative decoding, and quantization.
• Ensure infrastructure and reliability by managing SLURM/Kubernetes multi-node jobs, checkpoint hygiene, and GPU failure handling.
• Contribute to building AI for creating world simulations.

Culture

Requirements