TensorFlow's benchmark graveyard and its one active project
A split repository where half the code is retired and the other half, PerfZero, still runs performance tests for TensorFlow models.

What it does
This repo houses two things: PerfZero, a living benchmark framework for TensorFlow, and tf_cnn_benchmarks, a collection of TensorFlow 1 CNN benchmarks that is explicitly no longer maintained. The README nudges you toward TensorFlow Official Models if you actually want to run and measure things.
The interesting bit PerfZero appears to be the sole survivor—what the README calls “a benchmark framework” without elaboration, which is either confidence or brevity taken too far. The dead project lingers in the tree like a cautionary exhibit.
Key highlights
- PerfZero: active benchmark framework for TensorFlow (details live in its subdirectory)
scripts/tf_cnn_benchmarks: TensorFlow 1 CNN benchmarks, explicitly unmaintained- README suggests TensorFlow Official Models as the preferred alternative for performance measurement
- 1,147 stars suggest historical relevance more than current momentum
Caveats
- The top-level README is thin; you must dig into the PerfZero subdirectory to understand what the framework actually does
- One of the two listed projects is deprecated, which may confuse newcomers
Verdict Worth a look if you need to benchmark TensorFlow and want to evaluate whether PerfZero fits your workflow, but most developers should probably start with the TensorFlow Official Models instead.