Research Scientist / Engineer – Performance Optimization
Luma · Remote · Remote, EU ·
COMPENSATION
€•• – ••k
Listed on the employer's job board (Ashby)
- Fully remote
- Expert CUDA/Triton
- GPU optimization
- Transformer focus
- PyTorch kernels
Join Luma to enhance the speed and efficiency of their multimodal AI models. You will profile and optimize code across GPUs, CPUs, and accelerators to ensure efficient training and scalable deployment without compromising quality.
The work involves writing high-performance kernels and operations that maximize hardware utilization. It is designed for engineers with deep expertise in transformer internals and low-level optimization techniques.
What you'll do
- profile GPU, CPU, and accelerator code for minimal latency
- write high-performance PyTorch, Triton, and CUDA operations
- develop fused kernels leveraging tensor cores
- optimize model architectures for distributed multi-node deployment
- build performance monitoring tools and automation
- research and implement optimization techniques for transformers
What you'll need
- expert-level Triton and CUDA programming skills
- strong PyTorch proficiency including kernel development
- experience with profiling tools like NVIDIA Nsight
- deep understanding of transformer architectures
- knowledge of attention mechanisms
Nice to have
- experience with compilers like torch.compile or TensorRT
- background in optimizing inference workloads
- familiarity with Triton compiler and kernel fusion
- knowledge of warp-level intrinsics
Stack
- CUDA
- Triton
- PyTorch
- NVIDIA Nsight
- TensorRT
- ONNX
- XLA