Boston, MA · Job # 8685BK
Our client is searching for a deep learning compiler engineer to build the compiler and software tool chain for deploying machine learning models on to a variety of photonics accelerators.
You will collaborate closely with engineers from architecture and research team etc. to scope out system/software/hardware requirements from chip up to algorithm level while engaging with hardware teams to understand the target hardware platform and its constraints like photonic circuits.
- Design, implement and test compiler features and capabilities related to IR infrastructure and compiler passes
- Develop graph compiler optimizations like operator fusion, layout optimization, etc. that are customized to the different accelerators
- Develop efficient custom operators for opto-electric pipeline and optimize data motion
- Integrate open-sourced compiler technology into internal compiler infrastructure
- Build performance tooling to evaluate, understand and improve ML performance of different models on different accelerators
- Collaborate with cross functional agile teams of presale and hardware engineers to guide the direction of machine learning
- Follow industry and academic developments in the ML compiler and algorithm domain for photonic quantization and gradient descents and provide performance guidelines and best practices for other partner teams
- GPU programming (CUDA) and familiarity with deep learning stack (e.g., cuDNN, cuBLAS)
- Experience with NVDLA and other accelerators etc.
- Experience with open-source deep learning stacks (TVM, XLA, etc.)
- 5+ years of experience in the field of compiler design and 2+ years of experience with deep learning
- Experience with deep learning frameworks (e.g., Tensorflow, Pytorch etc.) and software stack (e.g., TensorRT, TVM, etc.)
- Experience with ML accelerators and hardware architecture
- Strong expertise in writing production quality C++ code
- Comfortable and experienced in software development lifecycle - coding, debugging, optimization, testing, integration
- Familiarity with parallelization techniques for ML acceleration
- MS, or higher degree, in CS/CE/EE, or equivalent, in industry experience
Applicants must be authorized to work in the United States legally.
Apply For this Position