ML Engineer: AI Accelerator & MLIR Optimization

Cerebras; United States; Remote; USD 140,000 - 210,000; Full time

Job summary

Cerebras Systems is seeking an ML Engineer to move models from reference implementations to efficient execution on Cerebras AI accelerators. You will own model bring-up, MLIR-based lowering paths, and optimization across model, compiler, kernels, and runtime.

Excellent C++/Python skills and experience with MLIR are required. Responsibilities include implementing hardware-specific mappings, improving operator fusion and memory layouts, and validating numerical accuracy against references.

Qualifications

• Proven experience bringing up and debugging ML models in PyTorch or similar.
• Experience with MLIR lowering paths and dialects.
• Strong C++ and Python programming skills.

Responsibilities

• Bring up new models: load and convert weights, implement execution paths, verify against references.
• Lower models to hardware via MLIR: develop dialects and graph transformations.
• Support model operations: enable attention, matmul, normalization, and embeddings through compiler/kernel changes.
• Optimize execution: fusion, layouts, tiling, memory, data movement, parallelism.
• Optimize inference: tune prefill/decode, KV cache, batching, and quantization.

Skills

C++

Python

MLIR

PyTorch

Compiler/Systems

Tools

MLIR tooling

Job description

Cerebras Systems is seeking an ML Engineer to move models from reference implementations to efficient execution on Cerebras AI accelerators. You will own model bring-up, MLIR-based lowering paths, and optimization across model, compiler, kernels, and runtime.

Excellent C++/Python skills and experience with MLIR are required. Responsibilities include implementing hardware-specific mappings, improving operator fusion and memory layouts, and validating numerical accuracy against references.