Model Optimization Engineer (PyTorch Infrastructure Development)

Key Skills

Algorithms, Infrastructure as code, JAX framework, Machine learning techniques, Optimization, PyTorch, TensorFlow

Job Description

We work on a python library that implements a variety of training time and post training quantization algorithms and provides them to developers as simple to use, turnkey APIs, and ensures that these optimizations work seamlessly with the Core ML inference stack and Apple hardware. Our algorithms are implemented using PyTorch. We optimize models across domains, including NLP, vision, text, generative models etc. In this role, the Model Optimization Engineer will be an expert in understanding the internal workings of PyTorch, graph capturing and graph editing mechanisms, methods to observe and modify intermediate activations and weights, tensor subclasses, custom ops, different types of parallelism for training models, and use this knowledge to implement and update the core infrastructure of the optimization library which enables an efficient and scalable implementation of various classes of compression algorithms. You'll also set up and debug training jobs, datasets, evaluation, performance benchmarking pipelines. Ability to ramp up quickly on new training code bases and run experiments. Run detailed experiments and ablation studies to profile algorithms on various models, tasks, across different model sizes.

Minimum Qualifications

• Bachelors in Computer Sciences, Engineering, or related discipline.
• 3+ years of industry and/or research experience
• Highly proficient in Python programming
• Proficiency in at least one ML authoring framework, such as PyTorch, TensorFlow, JAX, MLX
• Experience in the area of model compression and quantization techniques, specially in one of the optimization libraries for an ML framework (e.g. torch.ao).
• Design and develop the core infrastructure which powers the implementations of various compression algorithms (training time, post training, data free, calibration data based etc)
• Implement latest algorithms from research papers for model compression in the optimization library.
• Collaborate with software and hardware engineers, from the ML compiler inference stack, to co-develop new compression operations, model export flows for on device deployment.
• Design clean, intuitive, maintainable APIs

Preferred Qualifications

• Demonstrated ability to design user friendly and maintainable APIs
• Experience in training, fine tuning, and optimizing neural network models
• Primary contributor to a model optimization/compression library.
• Self prioritize and adjust to changing priorities and asks
• Improving model optimization documentation, writing tutorials and guides
• Good communication skills, including ability to communicate with cross-functional audiences

Pay & Benefits

At Apple, base pay is one part of our total compensation package and is determined within a range. This provides the opportunity to progress as you grow and develop within a role. The base pay range for this role is between $143,100 and $264,200, and your base pay will depend on your skills, qualifications, experience, and location.

Apple employees also have the opportunity to become an Apple shareholder through participation in Apple’s discretionary employee stock programs. Apple employees are eligible for discretionary restricted stock unit awards, and can purchase Apple stock at a discount if voluntarily participating in Apple’s Employee Stock Purchase Plan. You’ll also receive benefits including: Comprehensive medical and dental coverage, retirement benefits, a range of discounted products and free services, and for formal education related to advancing your career at Apple, reimbursement for certain educational expenses — including tuition. Additionally, this role might be eligible for discretionary bonuses or commission payments as well as relocation. Learn more about Apple Benefits.

Note: Apple benefit, compensation and employee stock programs are subject to eligibility requirements and other terms of the applicable plan or program.