Engineering Manager - Models Team

Position Overview

• Location Type: Remote
• Job Type: FullTime
• Salary: $235K - $300K

Replicate is a company that makes it easy for software engineers to run and customize machine learning models in the cloud. They offer a library of thousands of open-source models, allowing users to get started with one line of code or fine-tune and deploy custom models. The company handles the infrastructure, enabling users to focus on building. The team is composed of individuals from companies like Docker, GitHub, and NVIDIA, and they are passionate about making AI intuitive. They build in public, ship fast, and prioritize details.

Team Focus

The models team at Replicate focuses on:

• Turning research into APIs: Packaging models with cog and making them accessible.
• Making models faster: Utilizing techniques like CUDA, quantization, and parallelism.
• Building new model features: Developing creative features and novel ways to utilize models.

Requirements

• Excitement about models, model performance, and AI infrastructure.
• Experience leading engineering teams.
• Comfort guiding a group of strong engineers and solving hard problems alongside them.
• Care about open source and collaboration with the AI community.

Responsibilities

• Leading and growing a team of 6-8 engineers focused on packaging, optimizing, and improving generative models.
• Building tools and workflows to help model creators ship their work on Replicate.
• Pushing model performance further through optimizations like CUDA, quantization, and parallelism.
• Experimenting with creative ideas to make models more useful and powerful.
• Setting the direction for short-term projects and long-term bets.
• Encouraging open-source contributions and contributing oneself.

About You

• You’re obsessed with optimizations, performance, and measurements.
• You’ve worked with open-source model ecosystems and want to make them better.
• You’ve led teams before but still enjoy doing technical work.
• You’re familiar with model inference performance and how to... (Description incomplete)