- San Jose, California, United States
- On-site
- Full-time
- $180K–$450K per year
Building multimodal foundation models, focusing on mid-training recipes, synthetic data pipelines, and model distillation.
Compensation
The US base salary range for this full-time position is between $180,000 - $450,000 annually.
Responsibilities
- Develop mid-training recipes using high-quality and synthetic data
- Build synthetic data pipelines with code execution tests and filters
- Distill capabilities from larger models into smaller models
- Develop long-context data recipes
- Evaluate mid-training checkpoints
Requirements
- Nice to have: distillation
- Nice to have: rejection sampling
- Nice to have: model-based data generation
- Nice to have: long-context training
- Nice to have: rotary position embedding (RoPE) scaling)