Build and improve the model-training pipeline and infrastructure, from cluster scheduling and reinforcement-learning code execution to evaluation throughput, while strengthening software quality and enabling faster model experimentation.
Build and improve the software, tooling, and infrastructure behind large-scale model training, from cluster scheduling and reinforcement-learning code execution to evaluation throughput and maintainable research code.
Optimize foundation-model pre-training on large GPU clusters by profiling workloads, tuning distributed parallelism, and improving hardware utilization and throughput from Python code to GPU kernels.
Research and scale LLM post-training methods to reduce hallucinations, improve evidence grounding and reliability, and deliver distributed training pipelines and rigorous evaluations.
Page 1 · 4 results
