Research Scientist, Post-Training

What you'll do

• Research and implement methods to generate, refine, and curate instruction, preference, capability-specific, and domain-specific post-training data.
• Explore end-to-end and joint pre-training/post-training data curation to improve the post-trainability and performance of models.
• Source, evaluate, implement, and improve research ideas, translating them into practical product improvements in collaboration with engineers and customers.

What they ask for

• 3+ years of deep learning research experience.
• Experience post-training large vision, language, and multimodal models, including relevant algorithms or methods such as preference-based tuning, self-training, chain-of-thought distillation, or SFT.
• Experience developing post-training tooling and engineering.
• Strong deep learning fundamentals, software engineering skills, and deep learning framework skills to conduct large-scale experiments and build production prototypes; PyTorch or willingness to learn PyTorch. Demonstrated deep learning research success through papers, tools, or other artifacts.

Nice to have

• Experience with data management and distributed data processing solutions such as Spark or Snowflake.
• Experience building and shipping ML products.

Pay

$180,000 to $300,000 USD base salary

Tools

PyTorch, DPO, RLVR, RRHF, Spark, Snowflake

AI focus

Research and develop post-training data curation and synthetic data methods for foundation models, including instruction and preference data, and joint optimization of pre-training and post-training data.