PyTorch Profiling and Monitoring

See the Quick Start guide on how to install and configure Graphsignal.

Graphsignal automatically instruments and profiles PyTorch.

What’s captured

• Profiling: common PyTorch operator and module hot paths (for example torch.nn.Linear.forward, attention layers, distributed collectives, CUDA sync points).
• Metrics: CUDA memory metrics from torch.cuda.memory_stats (allocated/reserved/peaks, OOMs, utilization, fragmentation) when CUDA is available.

Integration into your Python application that uses PyTorch

Call graphsignal.configure(...) in your app and import/use PyTorch normally:

import graphsignal

graphsignal.configure(api_key="my-api-key")
# or pass the API key in GRAPHSIGNAL_API_KEY environment variable