Responsibilities
- Build and own the end-to-end model fine-tuning pipeline: data preprocessing, training, evaluation, and model registry.
- Implement and optimize fine-tuning techniques (QLoRA, LoRA, PEFT, full fine-tune) for our training workloads.
- Design and maintain evaluation harnesses with task-specific benchmarks and automated regression testing.
- Drive the training iteration loop: analyze results, diagnose failure modes, improve data and configuration.
- Implement experiment tracking, hyperparameter optimization, and reproducible training workflows.
- Collaborate on training data strategy with data engineering, including synthetic data generation.
- Evaluate model quality across safety, accuracy, latency, and cost dimensions.
- Contribute to model serving architecture and inference optimization.
- Mentor ML engineers across the team.
Requirements
- 10+ years of software engineering experience, with 4+ years focused on ML/NLP systems.
- Hands‑on experience fine‑tuning large language models with parameter‑efficient methods.
- Strong knowledge of transformer architectures, tokenization, and training optimization.
- Experience building production ML training pipelines with experiment tracking.
- Proficiency in Python, PyTorch, and distributed training frameworks.
- Experience with GPU‑based training infrastructure in the cloud.
- Strong evaluation methodology: designing benchmarks, measuring quality, detecting regressions.
- Experience with RLHF, DPO, or other alignment techniques is a strong plus.
- BS/MS in Computer Science, Machine Learning, or equivalent experience.
#J-18808-Ljbffr