PyTorch Conference 2024

As AI models become increasingly powerful and pervasive, trust and safety have become top priorities. Join us for a timely talk on Llama 3, our latest foundation model, and the cutting-edge trust and safety models and tools we've developed to ensure responsible AI development. In this talk, we'll dive into: •The advancements of Llama 3 and its applications •Our innovative trust and safety approaches, including toxicity detection and mitigation •The open-source tools and resources we're sharing to empower the community Discover how Meta is pushing the boundaries of trust and safety and learn how you can integrate these solutions into your own projects. Let's build a safer, more responsible AI future together!

The DeepChem library is a scientific computing library that implements deep learning infrastructure for drug discovery, materials discovery, and biology. The DeepChem community is one of the largest scientific open source projects built in PyTorch, with over 5K stars on Github and thousands of citations. The DeepChem community has learned a number of useful lessons for building and maintaining high quality scientific code built on top of PyTorch. In this talk, I will share our learnings with the PyTorch community and also highlight opportunities for improving scientific support in the ecosystem.

torch.compile is a graph compilation technique that improves GPU utilization. A key challenge in getting torch.compile to perform well is to minimize (or eliminate) graph breaks, however, this isn't trivial as even the Llama implementation provided by Meta has many graph breaks resulting in reduced training throughput. In this talk we discuss 1. how we addressed these challenges in order to train a model using torch.compile 2. how we combined torch.compile with FSDP and selective activation checkpointing to achieve the maximum throughput for training 3. model quality comparison between models trained with compile and no-compile, and lastly 4. the best setup we have for different model sizes in the Llama family that achieves the maximum throughput and MFU number (e.g. 68% MFU for the 7B model on A100 GPUs!)

This talk gives a run through of who builds PyTorch, new and upcoming improvements to the framework and how to get involved. All thanks to our awesome community of contributors, partners and ecosystem tools.

Journey across the PyTorch stack and see all the extension points that exist from nn.Module to the c++ Dispatcher through autograd and subclasses. This sessions will cover example use cases and when each one should be used while pointing to reference for in-depth details.

Large-scale model pretraining crucially relies on specialized and dedicated dataloaders that can, for example, partition and stream data asynchronously across multiple processes and physical nodes. In this talk we discuss one of the torch-native dataloaders we built and use at IBM Research for addressing these needs. Intended for use in large-scale model pretraining, particularly in research settings where rapid iteration between datasets may be required, our dataloader is distributed, stateful, checkpointable, composable and rescalable – while remaining a simple extension of the existing PyTorch dataloading framework. It automatically and invisibly handles data sharding, shuffling, subdataset weighting, checkpoint saving and loading, and custom user-defined preprocessing functions, with minimal overhead and high throughput. We discuss these properties and how we achieved them, such as reducing overhead by implementing a custom LCG random number generator, and demonstrate proof of concept on production-scale training of a 7B parameter Llama model over 4 trillion tokens.

This work explores performance optimization strategies for training 3D generative models using PyTorch. We focus on training Variational Autoencoders (VAEs) on the ShapeNet dataset, a popular benchmark for this task. Our objective is to achieve high-fidelity reconstructions while minimizing the computational footprint and training time. We focus on: 1) Large-scale 3D dataset loading strategies using PyTorch & Google Cloud Storage Buckets 2) Implementation details and insights for 3D VAEs using PyTorch 2.x 3) Training using Automatic Mixed-precision regimes 4) Optimized training using torch.compile and different quantization techniques (as supported) - Dynamic Quantization - Static Quantization - Static Quantization-aware Training 5) Comparative Benchmark over several experiments performed with a focus on execution time and memory footprint Through this comprehensive study, we present a comparative analysis of the performance gains achieved by our optimized models. Our findings present empirical insights into the trade-offs between model accuracy, computational complexity, and hardware resource utilization.

A hallmark feature of PyTorch is the natural expression of computation. This enables practitioners to implement AI models with ease. However, it prompts the question how to optimize the workload for a given hardware setup because those optimizations clutter our code and are tricky to combine. Lightning Thunder provides a Python-to-Python compiler to scale and optimize PyTorch programs that focuses on usability, understandability, and extensibility. A key tool in delivering on these goals is the composability of transformations: without changing the user code, we can stack quantization, distributing the computation across multiple GPUs, dispatching to optimized kernels, offloading, and other pluggable optimizations. Lightning Thunder flourishes in the PyTorch ecosystem: with PyTorch eager and with executors like torch.compile and nvFuser. It also dispatches to libraries like cuDNN, TransformerEngine, Apex, OpenAI Triton. The ability to apply multiple optimizations just-in-time leads to significant compounded speed-ups over unoptimized code out of the box. Luca will discuss the design of Thunder and demonstrate applications on training and inference for large language and multimodal models.

In the last 2 years we've optimized performance of PyTorch on Arm processors. The optimizations have included changes to ATen, C10, MKLDNN operators, GEMM backend, and Torch inductor. In many cases instead of writing our own kernel we integrated the Arm compute library, used fastmath kernels with format types like bf16, implemented operator caching, selected optimal backend based on the input context etc. Through these optimizations we improved performance by over 2x. In this presentation first we will talk about how we went across this process, what those optimizations are, performance numbers for AWS Graviton3 processors for around 75 models, and CI/CD workflow details. Next, we will walk through a sample PyTorch application showing basic usage, how to tune runtime and the resulting speed up. At the end of the presentation attendees will learn about PyTorch performance optimizations on Arm processors, how to use them, and the areas where they can collaborate to further improve PyTorch for aarch64 CPUs.

Artificial Intelligence (AI) has the potential to revolutionize various sectors, including environmental conservation and water management. However, the deployment of AI technologies raises ethical questions about the environmental impact, particularly water resources. This presentation will discuss the ethical implications of AI concerning water while also exploring how AI can both positively and negatively affect water resources along with the broader ecosystem. My goal is to facilitate a critical conversation around how to balance technological advancements with environmental stewardship. Objectives: Understanding Ethical Implications: Provide an in depth overview of how AI impacts water resources. Focus on ethical concerns related to AI's water footprint, including, but not limited to energy consumption and water usage in data centers. Explore Positive Applications: Talk about the possible successful implementations of AI in water conservation, pollution monitoring, and efficient resource management. Discuss potential future applications where AI could contribute to sustainable water management and connect stakeholders to address ethical concerns and solutions.