Job Description

About Templar AI

Templar is redefining how large language models are trained. The team enables permissionless pretraining — allowing collaborators across diverse computational environments to jointly train LLMs without centralized coordination. No single entity controls the process. Anyone can contribute compute, and the system is designed to handle it.

Templar's latest research, "Incentivizing Permissionless Distributed Learning of LLMs," introduces Gauntlet — an on-chain incentive system that powered a fully decentralized 1.2B parameter LLM training run. The work represents a meaningful step toward community-driven AI development at scale, and the team is now pushing further into pre-training and the infrastructure required to make distributed learning work at the frontier.

The Opportunity

Templar is looking for a Research Scientist to perform novel research in decentralized LLM pre-training. You will develop and evaluate ideas for scaling large-scale pre-training in bandwidth-constrained, heterogeneous, and error-prone environments — the exact conditions that define real-world decentralized infrastructure.

This is a research role with direct production impact. Your work will be ported to Templar's pre-training platform and published at major venues. If you want your research to advance the field and ship to a real system, this is that role.

What You'll Do

• •Perform novel research in decentralized training of LLMs with a focus on pre-training
• •Develop, implement, and evaluate ideas for scaling large-scale pre-training in bandwidth-constrained, heterogeneous, and error-prone environments
• •Contribute to porting methodology to Templar's pre-training platform
• •Publish and present work at major conferences

You Should Have

• •PhD in Machine Learning or equivalent research experience
• •Publications at top-tier venues such as NeurIPS, ICML, ICLR, CVPR, COLM, or EMNLP
• •Strong programming skills in PyTorch or JAX, with experience training models across multiple devices

Nice to Have

• •Experience with distributed training or federated learning
• •Familiarity with efficient LLM techniques including quantization, optimization, inference, or architecture design
• •Background in optimization theory