Remote ML Research Scientist: Post-Training & RL Methods

Crossing Hurdles; United States; Remote; USD 600,000 - 1,800,000; Full time

Job summary

Crossing Hurdles is seeking a Machine Learning Research Scientist (Remote) to advance post-training methods for large language and multimodal models. You will lead experiments, optimize training dynamics, and craft training recipes that improve generalization and capability acquisition.

The role requires deep expertise in reinforcement learning or model distillation, strong Python and PyTorch skills, and the ability to iterate quickly based on empirical evidence.

Qualifications

• Strong foundations in machine learning, deep learning, and optimization.
• Hands-on experience with training or post-training large-scale language or multimodal models.
• Deep technical expertise in reinforcement learning, model distillation, or learning dynamics.
• Proficiency in Python and PyTorch for implementing complex training pipelines.
• Experience designing controlled experiments and translating results into research directions.
• Ability to iterate quickly based on experimental evidence.

Responsibilities

• Research, implement, and scale advanced post-training methods across large language and multimodal models.
• Investigate learning objectives and optimization dynamics to improve training stability and capability acquisition.
• Develop and refine training recipes by optimizing data mixtures and evaluation methodologies.
• Design methods for learning from expert demonstrations and human preferences.
• Research the co-evolution of generators and reward models to improve model performance.
• Design rigorous experiments and evaluation frameworks to measure generalization and performance.

Skills

Machine learning foundations

Deep learning

Optimization

Python

PyTorch

Experimentation

Job description

Crossing Hurdles is seeking a Machine Learning Research Scientist (Remote) to advance post-training methods for large language and multimodal models. You will lead experiments, optimize training dynamics, and craft training recipes that improve generalization and capability acquisition.

The role requires deep expertise in reinforcement learning or model distillation, strong Python and PyTorch skills, and the ability to iterate quickly based on empirical evidence.