Machine Learning Engineer Model Bring Up
Cerebras builds wafer-scale AI computing systems and a cloud inference platform for training, fine-tuning, and serving AI models.
About Cerebras Systems, Inc.
Cerebras Systems is an AI-infrastructure company founded in 2015. It sells rack-scale wafer-scale computing systems and provides cloud-based, API-accessible AI inference alongside on-premises deployments.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will bring models from reference implementations to efficient accelerator execution. You will develop MLIR lowering paths, validate numerical correctness, and optimize model, compiler, kernel, and runtime performance. You will profile bottlenecks, improve inference performance, and work with hardware, compiler, kernel, and runtime teams to deliver reliable model support.
Requirements
- C++ and Python programming skills
- Experience bringing up and debugging machine learning models in PyTorch or a comparable framework
- MLIR experience, including dialects, rewrite patterns, transformation passes, and lowering pipelines
- Compiler fundamentals, including intermediate representations, dataflow analysis, and code generation
- Understanding of transformer architectures, attention mechanisms, tensor operations, and numerical precision
- Experience profiling and optimizing workloads on GPUs or other AI accelerators
- Ability to debug correctness and performance issues across model code, compiler-generated code, kernels, and runtime execution
Responsibilities
- Bring up new models by understanding architectures, loading and converting weights, implementing execution paths, and establishing correctness
- Develop MLIR dialects, graph transformations, lowering passes, and hardware-specific mappings
- Enable model operations through compiler and kernel changes
- Optimize operator fusion, tensor layouts, tiling, memory allocation, data movement, and parallel execution
- Tune prefill, decode performance, KV cache management, batching, and quantization
- Investigate numerical differences and measure accuracy impacts from precision changes and compiler optimizations
- Use profiling, execution traces, and hardware counters to identify bottlenecks
- Collaborate with hardware, compiler, kernel, and runtime teams to deliver reliable model support and performance benchmarks