Black Forest Labs
Member of Technical Staff - Research Engineer
$130K–$290K
San Francisco (United States) +1 · On-site · Staff · Full-time · Posted Jun 29, 2026
- Location
- San Francisco (United States)
Freiburg (Germany)
About the job
Improves the performance, reliability, and numerical stability of production training for large multimodal generative models.
What you'll do
- Profile and optimize full training steps, GPU kernels, data movement, memory use, and distributed training.
- Develop and validate low-precision training and GPU optimizations, and build trusted profiling and benchmarking tools.
- Partner with researchers to implement architecture changes and debug training failures while tying optimizations to model quality.
What they ask for
- Deep experience with large-scale training systems and hands-on improvements to throughput, memory, or stability in real training runs.
- Strong PyTorch fluency, including low-level training code.
- Experience with distributed training concepts such as FSDP, parallelism, activation checkpointing, NCCL, and overlapping compute and communication.
- Experience profiling GPU workloads and understanding low-precision training and quantization tradeoffs.
Nice to have
- Supported or co-owned training for a frontier foundation model that shipped or reached a major release.
- GPU kernel, attention, Hopper or Blackwell GPU, low-precision training, diffusion, flow-matching, DiT, or multimodal generative-model experience.
Pay
€130,000 - €240,000 EUR base salary; $180,000 - $290,000 USD base salary
Tools
PyTorch, FSDP, NCCL, Nsight Systems, Nsight Compute, CUDA, Triton, CUTLASS
Model-written summary · the original posting is authoritative