PyTorch developer job description

About the role
We are looking for a Mid-level PyTorch Engineer to build and maintain model training and inference code for applied machine learning products. You will work with researchers, data engineers, and backend engineers to turn prototypes into reliable PyTorch 2.x pipelines.

What you'll do
Implement training loops, datasets, metrics, and checkpointing for supervised and fine-tuning workflows. Debug tensor shape, dtype, device, and data loading issues. Use AMP, torch.profiler, and torch.compile where appropriate. Package models for batch or online inference with clear input and output contracts. Add tests, configuration, and documentation that make experiments repeatable.

What you'll bring
Hands-on PyTorch experience with nn.Module, autograd, DataLoader, optimizers, schedulers, and state_dict-based checkpoints. Comfort reading training logs, diagnosing convergence issues, and writing clean Python. Experience using CUDA-enabled environments, Git, containers, and common ML tools such as Hugging Face Transformers, Lightning, or Weights & Biases.

Nice to have
Experience with ONNX, TorchServe alternatives, Triton Inference Server, distributed training basics, or production monitoring for ML services.

About the role
We are hiring a Senior PyTorch Engineer to own model engineering for complex training and inference systems. You will make architecture decisions, guide implementation quality, and help teams move from experimental notebooks to maintained PyTorch codebases.

What you'll do
Design reusable training components for PyTorch 2.x while preserving flexibility for research. Lead debugging of distributed, memory, numerical, and performance issues. Choose between eager PyTorch, torch.compile, torch.export, ONNX, or runtime-specific deployment paths. Build robust checkpoint, evaluation, and model packaging workflows. Review code for reproducibility, operational safety, and long-term maintainability.

What you'll bring
Deep experience with PyTorch internals at the practical level: autograd behavior, CUDA memory, AMP or bfloat16, profiling, DDP or FSDP, and data pipeline performance. Strong Python engineering habits, including testing, typed interfaces where useful, configuration management, and CI. Ability to explain trade-offs clearly to researchers, platform engineers, and product stakeholders.

Nice to have
Experience fine-tuning transformer or vision-language models, using LoRA or FSDP, deploying GPU inference services, writing custom CUDA or Triton kernels, or migrating legacy TorchScript systems to newer PyTorch export paths.

About the role
We are seeking a Lead PyTorch Engineer to set technical direction for model training and inference across multiple projects. This role combines hands-on engineering, architecture ownership, and mentoring. You will define standards that let teams ship reliable PyTorch systems without blocking research velocity.

What you'll do
Lead design reviews for training platforms, fine-tuning pipelines, and GPU inference services. Establish patterns for PyTorch 2.x code structure, checkpoint compatibility, experiment reproducibility, distributed training, model export, and observability. Mentor engineers through difficult debugging work involving FSDP, torch.compile, memory pressure, numerical instability, and production regressions. Partner with engineering managers and research leads to sequence work, reduce technical risk, and make build-versus-buy recommendations.

What you'll bring
A track record owning PyTorch systems beyond a single model. Strong judgment across model quality, software design, infrastructure constraints, and operational risk. Experience leading code reviews, setting engineering standards, and guiding other engineers without taking over every implementation detail. Fluency with Python, CUDA-aware debugging, profiling, distributed training, and deployment trade-offs.

Nice to have
Experience with multi-tenant GPU platforms, inference runtimes such as Triton or ONNX Runtime, large-model fine-tuning, custom operators, privacy-aware ML logging, or governance for model release processes.