stormlog.tensorflow.profiler

Core TensorFlow Stormlog

Main profiling engine for capturing and analyzing GPU memory usage during TensorFlow model training and inference.

Classes

MemorySnapshot(timestamp, name, ...[, ...])	Represents a point-in-time memory snapshot.
ProfileResult(start_time, end_time, ...[, ...])	Comprehensive profiling results.
TFMemoryProfiler([device, ...])	Main TensorFlow Stormlog class.
TensorTracker()	Tracks TensorFlow tensor lifecycle and memory usage.

class stormlog.tensorflow.profiler.MemorySnapshot(timestamp, name, gpu_memory_mb, cpu_memory_mb, gpu_memory_reserved_mb, gpu_utilization, num_tensors, tensor_sizes=<factory>, operation_name=None, graph_node=None)

Bases: object

Represents a point-in-time memory snapshot.

Parameters:

• timestamp (float)

• name (str)

• gpu_memory_mb (float)

• cpu_memory_mb (float)

• gpu_memory_reserved_mb (float)

• gpu_utilization (float)

• num_tensors (int)

• tensor_sizes (Dict[str, int])

• operation_name (str | None)

• graph_node (str | None)

timestamp: float

name: str

gpu_memory_mb: float

cpu_memory_mb: float

gpu_memory_reserved_mb: float

gpu_utilization: float

num_tensors: int

tensor_sizes: Dict[str, int]

operation_name: str | None = None

graph_node: str | None = None

class stormlog.tensorflow.profiler.ProfileResult(start_time, end_time, peak_memory_mb, average_memory_mb, min_memory_mb, total_allocations, total_deallocations, snapshots=<factory>, function_profiles=<factory>, tensor_lifecycle=<factory>, memory_fragmentation=0.0, efficiency_score=0.0)

Bases: object

Comprehensive profiling results.

Parameters:

• start_time (float)

• end_time (float)

• peak_memory_mb (float)

• average_memory_mb (float)

• min_memory_mb (float)

• total_allocations (int)

• total_deallocations (int)

• snapshots (List[MemorySnapshot])

• function_profiles (Dict[str, Dict[str, Any]])

• tensor_lifecycle (Dict[str, Any])

• memory_fragmentation (float)

• efficiency_score (float)

start_time: float

end_time: float

peak_memory_mb: float

average_memory_mb: float

min_memory_mb: float

total_allocations: int

total_deallocations: int

snapshots: List[MemorySnapshot]

function_profiles: Dict[str, Dict[str, Any]]

tensor_lifecycle: Dict[str, Any]

memory_fragmentation: float = 0.0

efficiency_score: float = 0.0

property duration: float

Total profiling duration in seconds.

property memory_growth_rate: float

Memory growth rate in MB/second.

class stormlog.tensorflow.profiler.TensorTracker

Bases: object

Tracks TensorFlow tensor lifecycle and memory usage.

track_tensor(tensor, operation_name='unknown')

Track a new tensor.

Parameters:

• tensor (tensorflow.Tensor)

• operation_name (str)

Return type:

None

get_active_tensors()

Get information about currently active tensors.

Return type:

Dict[str, Any]

get_tensor_lifecycle()

Get complete tensor lifecycle history.

Return type:

List[Dict[str, Any]]

class stormlog.tensorflow.profiler.TFMemoryProfiler(device=None, enable_tensor_tracking=True)

Bases: object

Main TensorFlow Stormlog class.

Parameters:

• device (str | None)

• enable_tensor_tracking (bool)

capture_snapshot(name='snapshot')

Capture current memory state.

Parameters:

name (str)

Return type:

MemorySnapshot

profile_function(func)

Decorator to profile function memory usage.

Parameters:

func (Callable[[~P], R])

Return type:

Callable[[~P], R]

profile_context(name='context')

Context manager for profiling code blocks.

Parameters:

name (str)

Return type:

Iterator[None]

start_continuous_profiling(interval=1.0)

Start continuous memory profiling.

Parameters:

interval (float)

Return type:

None

stop_continuous_profiling()

Stop continuous memory profiling.

Return type:

None

get_results()

Get comprehensive profiling results.

Return type:

ProfileResult

reset()

Reset profiler state.

Return type:

None