Microsoft's AI Frameworks team is developing the AI software for training and deploying advanced AI models, collaborating with hardware teams on next-generation supercomputers and Maia AI accelerators, and optimizing model training and inference with ML researchers and developers, including work with OpenAI on Azure OpenAI service.
Requirements
- coding in languages including, but not limited to, C, C++, C, Java, JavaScript, or Python
- practical experience related to machine learning, including modeling, application, or system/kernel optimization.
- coding in languages including, but not limited to, C, C++, C, Java, JavaScript, or Python
- software development experience
- Experience in training or serving Deep Neural Network models.
- Experience with Language Models and ML system optimization are a plus.
- A strong technical background and solid foundation in algorithms, software engineering principles, and computer systems.
Responsibilities
- developing and evaluating core algorithmic and hardware technologies to enable large scale inferencing and training of the most advanced AI models on novel AI hardware.
- hands-on software design and development skills.
- solving hard technical problems and is motivated to learn new areas and tackle the hardest problems in building efficient AI systems.
- Collaborate broadly with ML researchers, system engineers, and production engineers.
- Engage with key partners to understand and evaluate performance and quality for state-of-the-art LLMs at different scales.
- Build software tools to support validation and exploration of LLM optimization technologies.
- Perform software development in model scripting and/or kernel languages, such as Python, C/C++, CUDA.
Other
- An entrepreneurial approach and ability to take initiative and move fast are essential.
- Great cross-team collaboration skills and the desire to collaborate in a team of researchers and developers.
- Strong communications skills (both written and oral)
- Identify requirements, scope solutions, estimate work, schedule deliverables.
