Role Overview
Crusoe's Managed AI team builds services for customers to customize and run large language models without managing infrastructure. The platform supports post‑training fine‑tuning (SFT, RL) and managed inference serving of open‑weight and fine‑tuned models, including LoRA adapters, on Crusoe's GPU fleet. Research done by the intern is deployed and served, with RL training relying on high‑throughput inference.
Requirements & Skills
- Education: Currently enrolled in a PhD program in Computer Science, Machine Learning, Electrical Engineering or a related field, with an expected graduation date between Fall 2027 and Spring 2029.
- Research experience: 2+ years of research experience in machine learning or ML systems, including hands‑on work with large language models: training or fine‑tuning (SFT, RLHF, DPO, GRPO/PPO) or inference optimization.
- Programming: Strong Python and PyTorch; comfortable reading and modifying large ML codebases.
- Experimental rigor: You design controlled experiments, choose the right baselines and metrics, and can tell a real effect from noise.
- Systems awareness: Working knowledge of how LLM training and inference run on GPUs, such as batching, memory, mixed precision and parallelism.
- Communication: You explain complex results clearly in writing and in person, and you work well with engineers turning research into product.
- Publications: Experience publishing at venues such as NeurIPS, ICML, ICLR, MLSys, ACL, EMNLP or COLM on RL for LLMs, reward modeling, agents, efficient training or efficient inference.
- Inference & training frameworks: Hands‑on experience with inference engines (vLLM, SGLang, TensorRT‑LLM) or training frameworks (verl, OpenRLHF, TRL, Megatron‑LM, DeepSpeed).
- Advanced techniques: Experience with quantization, speculative decoding, GPU kernels (CUDA, Triton) or LoRA/PEFT methods.
- Agent development: Experience building agent environments, tool‑use sandboxes or evaluation harnesses.