Distributed RL Post-Training Engineer

Openkrill; United States; Remote; USD 180.000 - 260.000; Vollzeit

Benefits dieser Stelle

Equity-heavy package

Remote-first culture

Visa sponsorship and relocation

Zusammenfassung

Pluralis is advancing Protocol Learning by training and serving large models on decentralized consumer hardware. We are building a fully remote, globally distributed team to run RL post-training across geo-distributed inference pipelines, with weights never fully held by any single participant.

You will design end-to-end RL post-training stacks, invent asynchronous algorithms, and ship first post-trained models as public artifacts.

Qualifikationen

• RL post-training experience on large language models.
• Strong engineering in Python and PyTorch.
• Publications in RL post-training or related fields are a plus.
• Ability to work across time zones and in distributed teams.

Aufgaben

• Build the post-training RL stack end-to-end: rollout ingestion, reward computation, policy updates, and syncing weights.
• Invent algorithms for asynchronous, high-latency, partially trusted generation with staleness tolerance.
• Ship first post-trained models and release them as public artifacts.

Kenntnisse

RL post-training

Distributed systems

Async training loops

Team collaboration

Tools

Python

PyTorch

Distributed systems concepts

Jobbeschreibung

Pluralis is advancing Protocol Learning by training and serving large models on decentralized consumer hardware. We are building a fully remote, globally distributed team to run RL post-training across geo-distributed inference pipelines, with weights never fully held by any single participant.

You will design end-to-end RL post-training stacks, invent asynchronous algorithms, and ship first post-trained models as public artifacts.