Posted
Senior AI R&D Engineer - Performance - Pre-training (f/m/d)
by AlephAlpha
- Artificial Intelligence
- Software
I’m interested
Free account · stay hidden from your current employer
Your interest is shared with employers only as an anonymized demand signal.
Introduction
Aleph Alpha is one of the few companies in Europe doing serious foundation model pre-training like our recently released Kolibri-1 model. Our customers - in finance, manufacturing, public administration - need models that understand German, meet European regulatory requirements, and work reliably in high-stakes settings. We're building that in Heidelberg.
We are hiring a Senior Performance Engineer to grow our pre-training efficiency team. If you are excited about making models fast and scaling them to thousands of GPUs, this is the role for you!
Tasks
- End-to-End Optimization: Profile training loops using PyTorch Profiler, Nsight Systems and Nsight Compute to identify system- and kernel-level bottlenecks in order to maximize model throughput.
- Distributed Strategy and Topology: Configure and tune composite parallelism strategies (e.g. TP, DP, HSDP/FSDP, EP), optimizing load balance, minimizing critical-path bottlenecks, and managing communication-to-computation trade-offs for large-scale LLM training.
- Hardware-Aware Modeling: Partner with AI Researchers to define model architectures for hardware efficiency without compromising convergence.
Requirements
- Are proficient in Python and the PyTorch library.
- Have a strong engineering background in parallel and/or distributed systems with proven track record of excellence.
- Have hands-on experience with modern machine learning techniques (especially large language models and their life cycle).
- Deeply understand the CUDA programming model.
- Have experience in distributed programming with APIs like NCCL or MPI.
- Have experience analysing profiling traces with tools such as PyTorch Profiler and Nvidia Nsight.
- Please note this role requires regular on-site collaboration in Heidelberg as a member of the Training Efficiency Team.
- Contributions to modern distributed training frameworks (e.g., TorchTitan, Megatron-LM, DeepSpeed).
- Familiarity with low-precision training formats (MXFP4, MXFP8) and their impact on numerical stability and throughput.
- A deep understanding of NCCL communication primitives, NVSHMEM or CUDA IPC and their performance.
- A proven track record of implementing and optimising modern transformer-based model training.
- A proven track record working on the NVIDIA Blackwell architecture.
Benefits
- Become part of an AI revolution!
- 30 days of paid vacation
- Access to a variety of fitness & wellness offerings via Wellhub
- Mental health support through nilo.health
- Substantially subsidized company pension plan for your future security
- Subsidized Germany-wide transportation ticket
- Budget for additional technical equipment
- Flexible working hours for better work-life balance and hybrid working model
- JobRad® Bike Lease