القائمة
GPU memory and utilization headroom
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
NCCL proxy wait decomposition — culprit (send_gpu_wait) vs victim (recv_wait)
One-shot health check: CPU, GPU, tables, torch step progress, TorchProbe overhead, cluster nodes, NCCL profiler health
SRE first-response triage for distributed training incidents. Automates the manual checks from PyTorch/NCCL debugging runbooks.
Diagnose PyTorch NCCL watchdog timeouts with Flight Recorder collective sequence alignment.
Detect monotonic GPU memory growth across training steps
Find slowest PyTorch modules in recent steps
| name | gpu_pressure |
| description | GPU memory and utilization headroom |
| category | memory |
| tables | ["gpu.utilization","gpu.devices","python.torch_trace"] |
| tags | ["gpu","VRAM","utilization","显存","利用率"] |
| keywords | {"en":["GPU memory","VRAM","GPU utilization","GPU idle"],"zh":["显存不够","GPU 利用率","VRAM","显存占用","GPU 空闲"]} |
| parameters | {"sample_limit":{"type":"integer","default":20}} |
查看 gpu.utilization 采样与 python.torch_trace 中的 allocated 是否一致， 判断是「真 OOM 风险」还是「利用率低 / 内存碎片」。
sample_limit (integer, default 20):