New Grad AI Performance Engineer: Inference & Optimization

SambaNova, Inc.; United States; On-site; USD 131,000 - 197,000; Full time

Benefits offered by this job

Health insurance premium coverage

HSA with employer contribution

Wellness benefits package

Job summary

SambaNova Systems is seeking a new grad AI Performance Engineer to bring up open source foundation models on their stack and optimize execution for faster inference. You will profile model performance across compiler, runtime, and hardware layers, identify bottlenecks, and apply quantization and scheduling techniques to improve throughput and latency.

Collaborate across ML, compiler, runtime, and hardware teams to ship performance improvements in SambaNova products.

Qualifications

• B.S. in CS, EE, or related field.
• Strong programming in Python or C++.
• Understanding of algorithms and numerical computing.
• Experience with PyTorch, TensorFlow, or JAX.
• Experience with inference frameworks such as vLLM, TensorRT, or SGLang.
• GPU programming experience with CUDA or Triton.

Responsibilities

• Bring up new foundation models on SambaNova platform.
• Profile model performance across the compiler, runtime, and hardware layers.
• Identify bottlenecks and propose dataflow or scheduling optimizations for single-node and distributed systems.
• Apply quantization, scheduling, and memory optimization techniques.
• Benchmark throughput and latency across releases.
• Collaborate with ML, compiler, runtime, and hardware teams to ship improvements.

Skills

Python/C++

Algorithms & DS

ML frameworks

CUDA/Triton

PyTorch/TensorFlow/JAX

Education

B.S. in CS/EE or related

Tools

CUDA

Triton

PyTorch

TensorFlow

JAX

vLLM

TensorRT

SGLang

Job description

SambaNova Systems is seeking a new grad AI Performance Engineer to bring up open source foundation models on their stack and optimize execution for faster inference. You will profile model performance across compiler, runtime, and hardware layers, identify bottlenecks, and apply quantization and scheduling techniques to improve throughput and latency.

Collaborate across ML, compiler, runtime, and hardware teams to ship performance improvements in SambaNova products.