Sr. Staff Machine Learning Researcher (AI)

TL;DR

Sr. Staff Machine Learning Researcher (AI): Leading research and development for LLM training and inference optimization on custom AI accelerators with an accent on distributed compute and performance tuning. Focus on implementing techniques like speculative decoding, quantization, and kernel fusion to bridge the gap between breakthrough research and production-ready systems.

Location: Hybrid, based out of Toronto, Ontario, Canada

Compensation: $100k - $500k

Company

hirify.global is building next-generation AI systems and custom AI accelerators to revolutionize computing performance, ease of use, and cost efficiency.

What you will do

• Lead R&D efforts focused on LLM training and inference optimization.
• Train, evaluate, and optimize state-of-the-art AI models on custom hirify.global hardware.
• Improve performance using techniques such as speculative decoding, quantization, kernel fusion, and flash attention.
• Investigate system bottlenecks and collaborate cross-functionally to drive performance improvements.
• Translate cutting-edge ML research into scalable, production-ready solutions.

Requirements

• 4+ years of industry and/or academic experience in ML research and LLM development.
• Strong proficiency in Python and PyTorch for developing and training deep learning models.
• Deep understanding of ML architectures, LLM training, and inference optimization.
• Hands-on experience training large-scale machine learning models.
• Must be eligible to access U.S. export-controlled technology (compliance with EAR).

Nice to have

• PhD in a relevant field.
• Published research in machine learning.
• Experience with speculative decoding.

Culture & Benefits

• Highly competitive compensation package including base and variable targets.
• Opportunity to work at the intersection of AI research and high-performance hardware.
• Collaborative environment valuing curiosity and a commitment to solving hard problems.
• Direct impact on the development of future AI computing paradigms.