Analyze Relax training scripts for performance and memory misconfigurations.
npx skills add https://github.com/EnkeZ/tgsopd --skill perf-doctor
Please help me install this Agent Skill. Skill: perf-doctor Source: https://github.com/EnkeZ/tgsopd/tree/main/skills/perf-doctor Command: npx skills add https://github.com/EnkeZ/tgsopd --skill perf-doctor
This Skill analyzes Relax training launch scripts to identify and address misconfigurations that may impact performance (time/MFU) or GPU memory usage.
Run the perf-doctor skill on your training script to analyze its performance and memory usage.
High-intent search queries and answers about installing and using this skill.
To diagnose GPU memory usage in a Relax training script, you can analyze the launch script to identify misconfigured flags and memory inefficiencies that cause unnecessary memory consumption or out-of-memory conditions.
Poor performance and low MFU during training are frequently caused by misconfigured flags in the training launch script, which can be identified through a performance analysis of the script.
You can check a training script for misconfigurations by running a diagnostic analysis to detect slow or inefficient flags, ensuring the script is optimized for performance and memory before execution.
Yes, analyzing training launch scripts beforehand checks for memory inefficiencies and potential out-of-memory conditions, ensuring memory usage is optimized and preventing training failures before execution.
perf-doctor specifically analyzes and optimizes training scripts built for the Relax framework, checking for performance and memory inefficiencies tailored to that environment.
You should run a training script analysis right before launching the script, as this identifies and addresses performance bottlenecks and memory issues, potentially avoiding long training times or failures.
DeepExperienceMigrates RedAccel reinforcement learning algorithms to Relax-compatible modules and interfaces.
cxnaiveFine-tune large language models with LoRA and QLoRA on limited GPUs.
xotongBenchmark and optimize Claude-flow v3 memory, CPU, and swarm coordination.
NVIDIA-NeMoReduce GPU peak memory usage during large-scale model training by adjusting parallelism and activation recompute strategies.
