stormlog.tensorflow.context_profiler

TensorFlow Context Profiling

Functions

`clear_global_profiler`()	Clear global profiler state.
`clear_profiles`()	Reset profiling data without discarding the global profiler.
`get_global_profiler`()	Get or create global profiler instance.
`get_profile_summaries`([limit])	Return aggregated profiling summaries for recent functions/contexts.
`profile_context`([name, profiler])	Context manager for profiling code blocks.
`profile_function`([func, profiler, name])	Decorator to profile function memory usage.
`profile_keras_training`(model, x_train, y_train)	Profile Keras model training.
`profile_model`(model[, profiler])	Profile all layers in a TensorFlow model.
`set_global_profiler`(profiler)	Set global profiler instance.

Classes

`ProfiledLayer`(layer[, profiler, name])	Wrapper for TensorFlow layers with automatic profiling.
`TensorFlowProfiler`([device])	High-level TensorFlow profiling interface.

stormlog.tensorflow.context_profiler.get_global_profiler()[source]

Get or create global profiler instance.

Return type:: TFMemoryProfiler

stormlog.tensorflow.context_profiler.set_global_profiler(profiler)[source]

Set global profiler instance.

Parameters:: profiler (TFMemoryProfiler)
Return type:: None

stormlog.tensorflow.context_profiler.profile_function(func=None, *, profiler=None, name=None)[source]

Decorator to profile function memory usage.

Parameters:

func (F | None) – Function to profile
profiler (TFMemoryProfiler | None) – Profiler instance (uses global if None)
name (str | None) – Custom name for profiling

Return type:

Callable[[F], F] | F

stormlog.tensorflow.context_profiler.profile_context(name='context', profiler=None)[source]

Context manager for profiling code blocks.

Parameters:

name (str) – Name for the profiling context
profiler (TFMemoryProfiler | None) – Profiler instance (uses global if None)

Return type:

Iterator[None]

class stormlog.tensorflow.context_profiler.ProfiledLayer(layer, profiler=None, name=None)[source]

Bases: object

Wrapper for TensorFlow layers with automatic profiling.

Parameters:

layer (Any)
profiler (TFMemoryProfiler | None)
name (str | None)

stormlog.tensorflow.context_profiler.profile_model(model, profiler=None)[source]

Profile all layers in a TensorFlow model.

Parameters:

model (Any) – TensorFlow model
profiler (TFMemoryProfiler | None) – Profiler instance

Returns:

Model with profiled layers

Return type:

Any

class stormlog.tensorflow.context_profiler.TensorFlowProfiler(device=None)[source]

Bases: object

High-level TensorFlow profiling interface.

Parameters:: device (str | None)

profile_training(model, dataset, epochs=1, steps_per_epoch=None)[source]

Profile model training.

Parameters:

model (Any) – TensorFlow model
dataset (Any) – Training dataset
epochs (int) – Number of epochs
steps_per_epoch (int | None) – Steps per epoch

Return type:

None

profile_inference(model, data, batch_size=32)[source]

Profile model inference.

Parameters:

model (Any) – TensorFlow model
data (Any) – Input data
batch_size (int) – Batch size for inference

Return type:

None

get_results()[source]

Get profiling results.

Return type:: Any

reset()[source]

Reset profiler state.

Return type:: None

stormlog.tensorflow.context_profiler.profile_keras_training(model, x_train, y_train, epochs=1, batch_size=32, validation_data=None, profiler=None)[source]

Profile Keras model training.

Parameters:

model (Any) – Keras model
x_train (Any) – Training data
y_train (Any) – Training labels
epochs (int) – Number of epochs
batch_size (int) – Batch size
validation_data (Any | None) – Validation data tuple (x_val, y_val)
profiler (TFMemoryProfiler | None) – Profiler instance

Return type:

None

stormlog.tensorflow.context_profiler.clear_global_profiler()[source]

Clear global profiler state.

Return type:: None

stormlog.tensorflow.context_profiler.clear_profiles()[source]

Reset profiling data without discarding the global profiler.

Return type:: None

stormlog.tensorflow.context_profiler.get_profile_summaries(limit=None)[source]

Return aggregated profiling summaries for recent functions/contexts.

Parameters:: limit (int | None)
Return type:: List[Dict[str, Any]]