Domains
Forms
Mechanism
Breaks down how the skill works and what it produces, so you can quickly judge whether it fits your scenario and value.
When planning memory headroom for a training run on NVIDIA DGX Spark GB10, when a job OOMs on unified memory, or when monitoring temperature/power during multi-hour training.
free -g and the UMA worksheet, not nvidia-smi.assets/thermal-sample.sh; recognize ~100W as sustained cap.Stable long-running ML jobs on DGX Spark: correct memory sizing, ordered OOM recovery, and thermal-aware slowdown diagnosis without misconfigured tuning.
Use Cases
A team launches a 70B QLoRA fine-tune on DGX Spark. Before starting, they use the skill's UMA worksheet and free -g to size headroom, compare against the 40GB anchor, and choose packing length to avoid mid-run OOM. The skill prevents wasted launches and gives confidence the run fits the 128GB pool.
A training job OOMs mid-step. The engineer follows the OOM ladder: flushes page cache with drop_caches, then reduces packing length, avoiding premature method downgrade. Separately, thermal sampling shows ~100W sustained — a normal cap, not a bug. The skill turns panic into a known procedure.
Skill Relationships
Dependency relationships read as "the upper tier points to the lower tier." The current Skill sits in the middle tier — above are Skills that depend on it, below are Skills it depends on.
Tier 1 · These Skills Use Me
Tier 2 · Current Skill
Tier 3 · I Use These Skills
Browse this skill's relationships within its skillset. Click a node to switch the side panel; use the search box to jump to any skill.
Skill File