Add memory profiling instrumentation to distributed MoE training workflows.
npx skills add https://github.com/mlc-ai/pith-train --skill add-memory-prints
Please help me install this Agent Skill. Skill: add-memory-prints Source: https://github.com/mlc-ai/pith-train/tree/main/.agents/skills/add-memory-prints Command: npx skills add https://github.com/mlc-ai/pith-train --skill add-memory-prints
Add detailed memory profiling instrumentation throughout the pithtrain training framework. This enables visibility into distributed setup, model creation, checkpoint loading, pipeline scheduling, per-layer activations, saved tensor profiling, expert MLP internals, and memory snapshot dumps to diagnose memory-related bottlenecks.
Run the memory instrumentation on your training script to enable detailed memory profiling during a run.
High-intent search queries and answers about installing and using this skill.
Profile GPU memory in distributed MoE training by integrating in-framework memory hooks and profiling utilities to surface allocation patterns across initialization, pipeline scheduling, and per-layer execution.
Debug MoE expert MLP memory spikes by applying targeted per-layer memory instrumentation to capture precise footprints of activations, weights, and caches to guide optimization.
Capture memory snapshots for activations and weights in PyTorch CUDA training by enabling optional saved tensor instrumentation to ground memory budgets and identify leaks.
Yes, this memory profiling works with pipeline scheduling and distributed setups by applying comprehensive instrumentation across setup, pipeline steps, and per-layer execution.
Diagnose GPU memory shortages during MoE initialization and checkpoint loading by integrating memory profiling to surface allocation patterns and breakdowns throughout the distributed setup workflow.
Rawgrowth-ConsultingTrain PyTorch models with distributed strategies and callbacks via PyTorch Lightning.
NVIDIADiagnose and optimize MoE training bottlenecks across memory, compute, and communication.
KappTech88Automate PyTorch Lightning training loops with distributed and mixed-precision support across CPU/GPU/TPU.
Mang30Profiles specified execution targets to diagnose bottlenecks across CPU, memory, GPU, and interconnect communication.
