- Raise experimental velocity by improving the speed of launching experiments and interpreting results.
- Maintain and evolve the JAX training framework for scalable, efficient distributed training on TPUs.
- Improve training data quality by investigating issues, building diagnostic tooling, and driving accuracy gains.
- Analyze production model accuracy and develop evaluation harnesses to guide development priorities.
- Refactor and modernize research prototypes into production-ready systems.
- Optimize production inference architectures for speech language models using techniques like quantization and speculative decoding.
- Investigate and resolve performance bottlenecks from low-level kernels to high-level system design.
- Collaborate across research, infrastructure, and production engineering to trace and resolve complex problems.
PythonMachine Learning