Real-world uses

Created for this library

1. 1.

   An ML platform team profiles per-node compute time in a TensorFlow graph to find bottlenecks during inference.

2. 2.

   A research engineer applies operator fusion across nodes in a TensorFlow graph to reduce serving latency for production models.

3. 3.

   An ML platform team uses graph-level optimizations across nodes in TensorFlow to lower compute cost for production models.