Skip to main content

Module distributed_profiling

Module distributed_profiling 

Source
Expand description

Enhanced Distributed Training Profiling

This module provides comprehensive profiling support for distributed training scenarios, including multi-node coordination, gradient synchronization analysis, and communication pattern optimization.

Structs§

Bottleneck
Detected performance bottleneck
CommunicationEvent
Communication event between nodes
CommunicationSummary
Summary of communication patterns
DistributedProfiler
Distributed training profiler
DistributedProfilerConfig
Configuration for distributed profiling
DistributedProfilingReport
Distributed profiling report
LoadBalanceAnalysis
Load balance analysis across nodes
NodeInfo
Information about a node in the distributed cluster
NodePerformanceSnapshot
Performance snapshot for a single node
RealtimeStats
Real-time statistics for dashboards
SynchronizationEvent
Gradient synchronization event
SynchronizationSummary
Summary of synchronization operations

Enums§

BottleneckType
Type of performance bottleneck
CommunicationType
Type of communication between nodes
NodeRole
Node role in distributed training
NodeStatus
Node status
SyncType
Type of gradient synchronization