About the job
As a Model Training engineer, you will design, build, and scale the post-training pipelines that turn a general LLM into a brand-fluent, production-ready assistant. Your innovations in fine-tuning and preference optimization (RLHF, DPO, GRPO, RLAIF) will directly improve reliability, alignment, and cost.
Responsibilities
Contribute to end-to-end post-training workflows—dataset curation, hyper-parameter search, evaluation, and rollout—using PyTorch, Torchtune, FSDP/DeepSpeed, and our internal orchestration stack.
Prototype and compare alignment techniques (e.g., curriculum RL, multi-objective reward modeling, tool-use fine-tuning) and push the best ideas into production.
Automate training at scale: build robust pipeline components, tools, scripts, and dashboards so experiments are reproducible and easy to trace.
Define the metrics that matter; run A/B tests and iterate quickly to meet aggressive quality targets.
Collaborate with inference, safety, and product teams to land improvements in customer-facing systems.
Qualifications
Minimum
Have hands-on experience training and fine-tuning large transformer models on multi-GPU / multi-node clusters.
Are fluent in PyTorch and its ecosystem tools (Torchtune, FSDP, DeepSpeed) and enjoy digging into distributed-training internals, mixed precision, and memory-efficiency tricks.
Have shipped or published work in RLHF, DPO, GRPO, or RLAIF and understand their practical trade-offs.
Care deeply about training tools, pipelines, and reproducibility—you automate the boring parts so you can iterate on the fun parts.
Balance research curiosity with product pragmatism—you know when to run an ablation and when to ship.
Communicate crisply with both technical and non-technical teammates.
Have a bachelor’s degree or equivalent in a related field to the offered position requirements.
Preferred
No preferred qualifications listed.