Together AI is looking to improve the efficiency of training foundation models, both in terms of speed and resource utilization, by analyzing limitations of current techniques and proposing new algorithmic and systems optimizations.
Requirements
- Skilled at writing high-quality and efficient code in Python and PyTorch
- Algorithmic modifications of large neural network training (e.g., novel optimization algorithms or model adaptation techniques)
- Distributed optimization (including federated learning, communication-efficient optimization, and decentralized training)
- ML systems optimizations for distributed training, memory efficiency, or compute efficiency
- Writing optimized NVIDIA GPU kernels or communication collectives using NVIDIA’s networking stack (e.g., NCCL or NVSHMEM)
- Running large-scale experiments on GPU clusters
- Follow the latest advances in relevant subfields of AI
Responsibilities
- Define and drive the research agenda around efficiency and performance of foundation model training
- Study recent results from the broader AI research community, analyzing their relevance to the team’s research directions and ongoing projects
- Conduct experiments to empirically validate your hypotheses and compare the outcomes with relevant related work
- Share your findings both internally and externally (e.g., at top-tier conferences on ML and ML Systems)
- Partner with Machine Learning Engineers to integrate advanced methods into Together’s Model Shaping platform
Other
- Can autonomously design, implement, and validate your research ideas
- Have first-author publications at leading conferences on ML or ML Systems (ICLR, ICML, NeurIPS, MLSys)
- Are a strong communicator, ready to both discuss your research plans with other scientists and explain them to broader audience
- Passionate about seeing your research create real-world impact through Together AI's services and willing to work hands-on with production systems to achieve it
- flexibility in terms of remote work
