cuda_record_memory_history: Enable Recording of Memory Allocation Stack Traces

In torch: Tensors and Neural Networks with 'GPU' Acceleration

Enable Recording of Memory Allocation Stack Traces

Description

Enables recording of stack traces associated with memory allocations, allowing users to identify the source of memory allocation in CUDA snapshots.

Usage

cuda_record_memory_history(
  enabled,
  context = "all",
  stacks = "all",
  max_entries = 1
)

Arguments

enabled	Character or NULL. Controls memory history recording. Options: NULL Disable recording of memory history. "state" Record currently allocated memory information. "all" Record the history of all allocation and free events (default).
context	Character or NULL. Controls traceback recording. Options: NULL Do not record any tracebacks. "state" Record tracebacks for currently allocated memory. "alloc" Record tracebacks for allocation events. "all" Record tracebacks for both allocation and free events (default).
stacks	Character. Defines the stack trace frames to include. Options: "all" Include all frames (default).
max_entries	Integer. The maximum number of allocation/free events to retain.

Details

Alongside tracking stack traces for each current allocation and free event, this function can also keep a historical log of all allocation and free events.

Use cuda_memory_snapshot() to retrieve recorded information. Visualization can be performed using pytorch.org/memory_viz.

Value

None; function invoked for side effects.

Examples

if (torch_is_installed()) {
## Not run: 
cuda_record_memory_history(enabled = 'all', context = 'all', stacks = 'all', max_entries = 10000)

## End(Not run)
}

torch documentation built on Aug. 21, 2025, 5:50 p.m.