ML Runtime and Kernel Engineer - Core ML

About the role

Cerebras Systems builds the world's largest AI chip, the Wafer-Scale Engine, which delivers training and inference speeds over 10 times faster than GPU-based cloud services. The Core ML team develops machine learning algorithms that exploit the unique capabilities of this architecture, covering efficient LLM training and inference, parallel and diffusion-based generation, sparsity, scaling laws and training dynamics. The role bridges research ideas and efficient execution on Cerebras systems, working across ML frameworks, compilers, runtimes and low-level kernels.

Key responsibilities

• Design and implement runtime components and high-performance kernels required by novel Core ML algorithms.
• Translate research prototypes into efficient implementations for the Cerebras platform, including reference implementations and comparisons on GPUs where useful.
• Profile and debug performance across the ML framework, compiler, runtime, communication and kernel layers.
• Optimize computation, memory movement, communication and concurrency for large-scale training and low-latency inference.
• Develop benchmarks, instrumentation and automated tests that validate functionality, performance and numerical correctness.
• Collaborate with Core ML researchers and compiler, runtime, kernel and inference engineers to evaluate design alternatives and deliver end-to-end capabilities.
• Contribute to software architecture and roadmap decisions by identifying recurring limitations and high-leverage platform improvements.

Required profile

• Bachelor's, Master's, PhD or equivalent practical experience in Computer Science, Computer Engineering, Electrical Engineering or a related field.
• Experience developing high-performance systems software, ML systems, runtimes, compilers or computational kernels.
• Proven ability to debug and profile complex software across multiple layers of a system.
• Familiarity with modern machine learning architectures and frameworks such as PyTorch or JAX.
• Ability to work with researchers and turn evolving algorithmic requirements into reliable software.

Required skills

• Strong programming in C++ and Python.
• Solid understanding of parallel programming, memory management, concurrency, data structures and performance optimization.
• Preferred: CUDA, Triton, low-level assembly, accelerator programming or a C-like domain-specific language.
• Preferred: compiler internals, distributed runtimes, custom hardware interfaces, HPC systems, LLM training or inference, and contributions to open-source systems.