Staff AI Training Platform Architect

Unconventional, Inc.; Mountain View (CA); On-site; USD 180,000 - 260,000; Full time

Benefits offered by this job

Competitive salary

Significant equity

Health benefits

401k matching

Unlimited PTO

Complimentary meals (Palo Alto office)

Job summary

Unconventional, Inc. in Mountain View seeks a Member of Technical Staff for the AI Training Platform to build and scale the end-to-end tooling required to train, evaluate, and benchmark models.

You will help researchers map neural networks to novel hardware and advance physics-based compute. You will design distributed training systems, optimize kernels with CUDA and Triton, and expand our internal training framework while collaborating with theorists and hardware teams to convert model

Qualifications

• BS degree in a related field.
• 5+ years of AI/ML engineering experience.
• Experience mapping state-of-the-art AI models to system performance implications.
• Production-grade training frameworks such as Megatron-LM, DeepSpeed, Ray.

Responsibilities

• Train, evaluate, and benchmark models at scale.
• Architect, scale, and maintain the core AI/ML training platform and infrastructure.
• Design and scale multi-node distributed training with elastic sharding and robust data streaming pipelines.
• Develop and optimize kernels using CUDA and Triton.
• Expand and maintain the proprietary training framework.
• Collaborate with theorists and hardware teams to translate model requirements into infrastructure specifications.

Skills

AI/ML engineering

Transformer models

Mixture of Experts

Diffusion models

Cluster communication

Parallel programming

Python

C++

Education

BS in Computer Science, Physics, Electrical Engineering, or Applied Math

Tools

Megatron-LM

DeepSpeed

Ray

PyTorch Lightning

CUDA

Triton

Job description

Unconventional, Inc. in Mountain View seeks a Member of Technical Staff for the AI Training Platform to build and scale the end-to-end tooling required to train, evaluate, and benchmark models.

You will help researchers map neural networks to novel hardware and advance physics-based compute. You will design distributed training systems, optimize kernels with CUDA and Triton, and expand our internal training framework while collaborating with theorists and hardware teams to convert model