Senior Inference ML Engineer: Sparse Attention, Pruning

Cerebras; United States; Ibrido; USD 150.000 - 190.000; Tempo pieno

Vantaggi offerti da questo lavoro

0

Descrizione del lavoro

Cerebras Systems seeks a Senior Research Engineer to adapt language and vision models for our AI accelerator. You will work with ML researchers to prototype, validate, and optimize models on Cerebras hardware for end-to-end inference at scale.

You will explore speculative decoding, pruning/compression, sparse attention, and sparsity-driven techniques to deliver low latency and high throughput. Collaborate across teams to push the frontier of inference.

Competenze

• Combination of advanced degrees and deep ML software experience.
• Experience in implementing and evaluating transformer-based models.
• Demonstrated ability to translate research into production‑quality code.
• Strong background in performance optimization on specialized hardware.

Mansioni

• Design, implement, and optimize transformer architectures on Cerebras hardware.
• Research and prototype inference algorithms and novel model architectures.
• Train models to convergence, run hyperparameter sweeps, analyze results.
• Bring up models on Cerebras system, validate correctness, debug issues.
• Profile and optimize code to maximize throughput and minimize latency.
• Develop tooling to surface bottlenecks and guide inference optimization.
• Collaborate across software, hardware, and product teams from inception to delivery.

Conoscenze

Python/C++ programming

ML software development

Transformer models

Performance optimization

Research impact

Formazione

Bachelor's + 7+ years ML software

Master's + 4+ years software

PhD + 2+ years research

Equivalent practical experience

Strumenti

PyTorch

Transformers

vLLM

SGLang

Triton/CUDA

Descrizione del lavoro

Cerebras Systems seeks a Senior Research Engineer to adapt language and vision models for our AI accelerator. You will work with ML researchers to prototype, validate, and optimize models on Cerebras hardware for end-to-end inference at scale.

You will explore speculative decoding, pruning/compression, sparse attention, and sparsity-driven techniques to deliver low latency and high throughput. Collaborate across teams to push the frontier of inference.