AI Software Solutions Engineer (AI Frameworks, Workloads)
Intel Corporation
Bengaluru, India
Job posting number: #7240009 (Ref:JR0263211)
Posted: April 26, 2024
Job Description
Job Description
We are looking for a dynamic software engineer to design, develop and optimize AI frameworks for training and inference on Intel Habana (https://habana.ai/) deep learning accelerators. In this role, you will work with a cross-geo team on enabling and optimizing state of the art deep learning models with a specific focus on the PyTorch framework. The roles and responsibilities that you would need to carry out may include the following:Design and develop SW techniques for AI frameworks - both HW-agnostic and HW-aware Contribute to enhancing and extending the Training and Inference capabilities in the Software stack. Profile deep learning inference and training workloads and identify optimization opportunities in the software stack.
Qualifications
BTech, MS or PhD in CS or related fields with an overall experience of 10 to 15 yearsProgramming skills in Advanced C++, Python and parallel programming skills
Previous exposure to Machine Learning (ML) frameworks such as PyTorch and Tensorflow.
Detailed understanding of machine learning systems optimization and deployment techniques such as quantization
understanding of optimization strategies for deployment of Large Language Models (LLMs)
knowledge of transformers, KV cache , prefill buffer etc optimzation technique for inference.
Working knowledge of operators in Pytorch or Tensorflow and Understanding of low level kernels.
Ability to debug complex issues in multi layered SW systems. Understanding of SW integration across open source framework and internal bridge layers.
Understanding of computer architecture and HW-SW optimization techniques
Practical knowledge of DL topologies for different use cases
Knowledge of compiler algorithms for heterogeneous systems
Experience working on frameworks/platforms that have gone to production
Effective communication skills and experience with working in a cross-geo setup
Preferred knowledge of open source compiler infrastructure like LLVM or gcc