Design and implement the end‑to‑end reinforcement‑learning post‑training stack for decentralized large language models. Invent algorithms that tolerate high latency, partial trust, and asynchronous rollouts across consumer devices. Require strong Python/PyTorch engineering, hands‑on RL experience, and research‑level problem solving.
Key Highlights
Build a full RL post‑training pipeline that runs on consumer‑grade, geo‑distributed hardware.
Adapt and invent RL algorithms for asynchronous, high‑latency, partially‑trusted environments.
Ship the first decentralized post‑trained models and evaluation framework.
Key Responsibilities
Build the end‑to‑end RL post‑training stack, including rollout ingestion, reward computation, policy updates, and weight distribution across the network.
Invent and adapt RL algorithms to handle asynchronous, high‑latency, partially‑trusted generation and communication‑efficient updates.
Develop evaluation pipelines and ship the first decentralized post‑trained model releases.
Set technical direction for the post‑training system and drive execution across the team.
Technical Skills Required
Reinforcement Learning
Python
PyTorch
Discover our full range of relocation jobs with comprehensive support packages to help you relocate and settle in your new location.
Benefits & Perks
Significant equity ownership
Fully remote work environment
Visa sponsorship and relocation assistance
Interested in relocating to Australia? Check out our comprehensive Relocation Jobs in Australia page with detailed relocation packages and benefits.
Nice to Have
Experience training over slow networks or with decentralized/federated setups.
Familiarity with serving‑engine internals such as vLLM or SGLang.
Experience with reward modeling or building verifiable‑reward datasets.
Experience with P2P networking and NAT traversal.
Background at proprietary, open‑weight, or open‑source AI labs.
Job Description
Pluralis Research works on Protocol Learning: training and serving large models in a fully decentralized way on small consumer-grade devices connected via the internet. Despite being dismissed as infeasible, we have made significant advances on this problem, most recently Agora, a permissionless run that pretrained an 8B model from scratch on consumer GPUs spread over the internet, with no single participant ever holding the full weights (tech report). While many of the core research problems have been solved, Protocol Learning unlocks a series of new challenges. For the mission in full, read A Third Path: Protocol Learning.
Want the full job description?
Read the complete details on LinkedIn, the original posting.
Continue on LinkedIn
This is a short excerpt. All rights to the full description belong to its original publisher.
