Research Scientist / Engineer – Reinforcement Learning Infrastructure

Luma Ai

≈ 1,3 млн–2 млн ₽ · 14,6 тыс.–21,9 тыс. €

16 667 – 25 000 $

🌍 Удалённо

Senior

Полная занятость

🌐 Глобал

Продуктовая

Tech

Описание вакансии

Обязанности

•Build high throughput rollout generation

•Build reward model serving pipelines

•Collaborate with researchers to operationalize RL training ideas

•Create LLM as judge evaluation pipelines

•Design curriculum and task sampling

•Design distributed reinforcement learning post training systems

•Develop evaluation monitoring and debugging tooling

•Develop reward infrastructure and verifiable rewards

•Implement defenses against reward hacking

•Implement reinforcement learning environments for multi step tasks

•Implement sequence packing for long trajectories

•Improve training efficiency and stability

•Integrate inference engines into training loop

•Reuse KV cache across rollouts

•Schedule heterogeneous training and inference workloads

Технологии: Asynchronous training, Containerization, Curriculum learning, Distributed Systems, Distributed Training, FSDP, GPU Cluster, Human Feedback, KV cache, Kubernetes, LLM Evaluation, Language Models, Large Language Models, Learning from Human Feedback, MPI, Mixture of Experts, Model Synchronization, NCCL, Off Policy, Off Policy Learning, Pipeline parallelism, Policy Optimization, Policy learning, Proximal Policy Optimization, PyTorch, Ray, Reinforcement Learning, Reinforcement Learning from Human Feedback, Reward Hacking, Reward Modeling, SGLang, Sandboxing, Sequence Packing, Tensor Parallelism, Tool use, VLLM