AI Research Scientist, Autodesk

• You are a research scientist who specializes in machine learning at scale. You have experience individually and as part of a team training large machine learning models with distributed deep learning. Your skills span across the design of large machine learning models, running large scale experiments informed by scaling laws, and optimizing data and compute GPU throughput to perform efficient model training
• You are passionate about solving problems and building things. You are excited to collaborate with AI researchers to build state-of-the-art generative AI features in Autodesk products. You are a good communicator and comfortable working at the intersection of research & product
• BSc or MSc in Computer Science, or equivalent industry experience
• Proven experience with large scale distributed ML model training
• Experience with 3D Parallelism including Tensor Parallelism, Sequence Parallelism and Pipeline Parallelism
• Knowledge of frameworks used for model sharding, e.g. DeepSpeed, FairScale
• Experience with large scale distributed data processing
• Experience with software version control, unit tests, and deployment pipelines
• Experience with cloud services & architectures (AWS, Azure, etc.)
• Excellent written communication skills to document code, architectures, and experiments
• Experience with hardware level systems for high performance computing
• A self-starter with initiative to search for solutions and execute on problems with minimal supervision
• A curious, creative problem-solver who is excited to learn & develop new technologies
• A fast learner & excellent communicator who can effectively collaborate across global locations
• Comfortable working in newly forming ambiguous areas where learning and adaptability are key skills

Desirable

• Lead the development and scaling of machine learning models at the intersection of research and product
• Leverage distributed deep learning frameworks, such as DeepSpeed, FSDP etc…, to perform scaling experiments with the support of engineering and data teams
• Collaborate with the engineering teams to find and resolve distributed deep learning training bottlenecks such as checkpointing, model/optimizer compression etc…
• Optimize data flows through the model and overall architecture
• Leverage frameworks such as Open Neural Network Exchange (ONNX) and TensorRT to create high performance model endpoints
• Writing robust, testable code that is well documented and easy to understand
• Analyze errors and provide solutions to problems that arise
• Present the results of your work to collaborators and leadership