🛠 Tech Stack

💼 About This Role

You'll drive down wall-clock time to convergence by profiling and eliminating bottlenecks across the foundation model training stack. Your work will span from data pipelines to GPU kernels, optimizing for scalability and efficiency.

🎯 What You'll Do

• Profile and eliminate bottlenecks in model training stack
• Design and optimize distributed training systems for GPU clusters
• Implement efficient low-level code (CUDA, Triton, custom kernels)
• Develop monitoring and debugging tools for large-scale runs

📋 Requirements

• 8+ years experience in distributed systems or HPC
• Production-grade Python expertise
• Low-level performance mastery with CUDA/cuDNN/Triton
• Experience with PyTorch and model parallelism