Mid-Training ML Engineer: Synthetic Data & Evaluation

Maven Ventures; Palo Alto (CA); On-site; USD 180,000 - 440,000; Full time

Benefits offered by this job

Equity

Comprehensive medical coverage

Vision coverage

Dental coverage

401(k) retirement plan

Short & long-term disability insurance

Life insurance

Discounts & perks

Job summary

SpaceXSI is building Super Intelligence to help humanity understand the universe. Our teams are building the full Super Intelligence stack, including Grok models and related applications.

We are seeking researchers who can scale data to trillions of tokens and distill model intelligence into efficient mid-training processes. You will design ML experiments, curate data for text, image, audio, and video, and leverage Spark and Ray for large-scale processing.

Qualifications

• Expertise in ML and large model scaling.
• Strong ability to design ML experiments.
• Familiarity with state-of-the-art techniques for curating AI training data for multiple modalities.
• Strong engineering abilities in Spark, Ray, and other frameworks for large-scale data processing.

Responsibilities

• Scale synthetic coding data to trillions of tokens with large-scale docker verification.
• Distill the intelligence of flagship models into flash models through synthetic data generation.
• Optimize mid-training data mixtures to boost the ceiling for RL.
• Engineer long-context data recipes.
• Develop robust and diverse evaluation for mid-training checkpoints.

Skills

ML scaling

Experiment design

Data curation

Spark

Ray

Tools

Spark

Ray

Job description

SpaceXSI is building Super Intelligence to help humanity understand the universe. Our teams are building the full Super Intelligence stack, including Grok models and related applications.

We are seeking researchers who can scale data to trillions of tokens and distill model intelligence into efficient mid-training processes. You will design ML experiments, curate data for text, image, audio, and video, and leverage Spark and Ray for large-scale processing.