Get Started
Home
Topics
Search
Library
Research questionHow can PyTorch 2 programs avoid graph breaks without manual refactoring?Graph breaks interrupt FX graph compilation and send execution back to Python eager mode, causing CPU-GPU synchronization and reducing optimization opportunities. Developers therefore face performance costs or must refactor otherwise valid model code.
AI
Code Generation & Program Synthesis
Inference Optimization
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.GraphMend: Code Transformations for Fixing Graph Breaks in PyTorch 2The problem concerns TorchDynamo and FX graph capture in PyTorch 2 programs. The supplied evidence examines 195 Hugging Face models, including 27 with graph breaks, and evaluates semantics-preserving source transformations on NVIDIA GPUs; it does not establish coverage for all graph-break patterns or PyTorch workloads.research paper · Sep 3, 2026
Related questions
How can full-graph visibility improve fusion and hardware utilization during Transformer training?How can we test whether language models genuinely execute multi-step graph logic when static benchmarks become contaminated?How can compiler intermediate representations preserve high-level abstractions in dynamically typed languages while keeping them optimizable?How can symbolic regression optimize constants efficiently across heterogeneous expression-tree populations on GPUs?