Staff ML Engineer: LLM Systems & Hardware Co-Design

MatX; Mountain View (CA); Hybrid; USD 160,000 - 800,000; Full time

Benefits offered by this job

Equity

Health insurance

PTO 4 weeks

Remote work

Professional development

401K

Commuter benefits

Family leave

Job summary

MatX in Mountain View, CA, is advancing systems-focused ML research to run the world's best AI models efficiently on specialized hardware.

The team trains and optimizes LLMs for our hardware, builds distributed training and inference infrastructure, and collaborates on hardware architecture from an ML perspective. A PhD or equivalent in ML or related fields is preferred, with strong software engineering skills and a track record of open research.

Qualifications

• PhD or equivalent experience with publications or shipped/open-sourced research in LLMs.
• Deep expertise in reinforcement learning, quantization or attention/architecture; inference/training optimization is a plus.
• Strong software engineering skills for building and scaling distributed training/inference codebases.

Responsibilities

• Train and optimize LLMs for our hardware.
• Run quality evaluations.
• Build and set up distributed infrastructure for training and inference.
• Advise on the hardware architecture from an ML perspective.
• Co-design hardware through ML experimentation and RTL prototypes.
• Work with other teams to validate numerics.

Skills

LLM research

Software engineering

Distributed training

ML hardware co-design

Education

PhD or equivalent

Tools

Python

C++

Distributed training frameworks

Job description

MatX in Mountain View, CA, is advancing systems-focused ML research to run the world's best AI models efficiently on specialized hardware.

The team trains and optimizes LLMs for our hardware, builds distributed training and inference infrastructure, and collaborates on hardware architecture from an ML perspective. A PhD or equivalent in ML or related fields is preferred, with strong software engineering skills and a track record of open research.