Module memory 

Memory timeline analysis — TensorFlow Memory Profile model.

Structs

DeviceMemoryStats

Per-device memory stats.

LiveAllocation

Active allocation at the global peak (TensorFlow breakdown table).

MemoryProfile

Full memory profile reconstructed from trace evidence.

MemorySummary

Aggregated memory statistics for one probe run.

MemoryTimelinePoint

One point on the memory-vs-time curve for a device.

NodeMemoryMetrics

Per-span / per-op TensorFlow memory metrics.

Enums

MemoryAction

Allocation or deallocation recorded during a probe run.

MemoryCategory

TensorFlow-profiler-style memory category for timeline breakdown.

Functions

analyze_memory

Build a TensorFlow-style memory profile from a parsed trace document.

category_for_step

Map a training step (+ tensor flags) to a PyTorch-style memory category.

dtype_size_bytes

Derive element size in bytes from a Candle-style dtype label (f32, F32, …).

elem_count

Product of shape dimensions; returns 0 for empty shape.

node_memory_metrics

Attribute memory metrics to graph node ids (span ids and {span}/op/{n} op ids).

resolve_storage_bytes

Resolve explicit bytes or derive from shape/dtype.

storage_bytes

Storage bytes for a dense tensor (elem_count × dtype_bytes).