Agent Skills for NVIDIA products — install into Claude Code, Codex, and other coding agents to run Physical AI, robotics, simulation, CUDA, and RAG workflows end to end.
Indexed The index holds package records for this repository. The package count is historical registry records; it does not mean those packages are available now.
Repository
Canonical name
NVIDIA/skills
GitHub repository ID
1167034425
Package records
394 (historical registry records, not current availability)
Discovery and queue
First recorded source
marketplace:anthropics/claude-plugins-official
Discovered
Queue state
ok
Recorded errors
0
Last settled processing
Next-check eligibility
Eligible from
The first recorded source is what the queue stored first, not the full discovery history. The error count increases when processing ends in the error state and resets after a successful settlement; it is not a count of all attempts. “Last settled processing” is when it finished, not when it started.
Eligibility is calculated as of . It is not a schedule: each collector run handles only a limited number of sources, and being eligible does not promise when a check happens.
Last saved scan
Repository metadata read
Scan status
complete
Scan saved
Extractor version
2
Continuation record
None
The scan record is separate from the queue state: the queue state shows processing progress, the scan record shows what was last saved. A continuation record only means unfinished state is stored; it has no remaining count.
Source-file evidence coverage
Raw counts of the pinned source files that stored package versions reference in this repository. They show what has been recorded, not how complete the repository is.
Scope and limits
Counts distinct pinned file links (repository, commit, path) across all stored versions, from any package, that name exactly this repository's canonical name. Files of packages owned by this repository are not counted unless some version links to them here. Alias names are not merged and no history is guessed from a current name.
“Known” means a regular-file identity is recorded. It does not mean a whole directory or package is covered. Links that are not valid pinned file URLs are not counted.
395 pinned source files referenced
395 known
0 absent at commit
0 file listing truncated
0 unsupported
0 not recorded
0 read failed
Deduplicated by repository, commit and path; the same path at two commits counts as two file identities. These are not component-group counts.
Failed source-commit reads 0
No source commit has a stored failed read for files that are still unrecorded.
Linked packages 394
Linked by verified repository identity. The count is historical registry records, not current availability.
Convert single-node scripts to multi-node Slurm sbatch jobs and debug common multi-node failures. Covers srun-native vs uv run torch.distributed approaches, container setup, NCCL timeouts, OOM sizing for MoE models, and interactive allocat…
Validate and use selective and full activation recompute in Megatron Bridge to reduce GPU memory usage at the cost of extra compute. Use for activation memory OOMs or regressions involving recompute_granularity, recompute_num_layers, recom…
Validate and use CPU offloading in Megatron Bridge, including layer-level activation offloading and fractional optimizer state offloading with HybridDeviceOptimizer.
Validate and use CUDA graph capture in Megatron Bridge, including local full-iteration graphs and Transformer Engine scoped graphs for attention, MLP, and MoE modules.
Validate and use MoE expert-parallel communication overlap in Megatron-Bridge, including overlap_moe_expert_parallel_comm, delay_wgrad_compute, and flex dispatcher backends such as DeepEP and HybridEP.
Choose the right MoE token dispatcher (`alltoall`, DeepEP, or HybridEP) for the hardware, EP degree, and optimization stage. Summarizes patterns from DSV3, Qwen3, Qwen3-Next, and VLM bring-up work.
Representative, point-in-time MoE training playbooks by hardware and model family. Use them as candidate seeds, then revalidate the exact runtime, semantics, topology, and steady-state throughput.
Long-context MoE training guidance for Megatron Bridge. Covers CP sizing, selective recompute, dispatcher choices, and practical patterns from DSV3, Qwen3, and Qwen3-Next long-context experiments.
Evidence-gated workflow for MoE performance optimization in Megatron Bridge. Covers measurement contracts, the Three Walls framework, parallel folding, profiling, matched A/B tuning, and final validation.
Practical guidance for training MoE VLMs in Megatron Bridge. Compares FSDP and 3D-parallel approaches, using rounded lessons from Qwen3-VL, Qwen3-Next, and other multimodal experiments.
Operational guide for choosing and combining parallelism strategies in Megatron Bridge, including sizing rules, hardware topology mapping, and combined parallelism configuration.
Validate and use packed sequences and long-context training in Megatron-Bridge, including offline LLM packing, collate-time VLM packing, Energon online packing, and CP constraints.
Operational guide for enabling TP, DP, and PP communication overlap in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
Recommend and customize Megatron Bridge library and benchmark recipes for a user's model, GPU count, hardware, sequence length, and pretrain/SFT/PEFT goal. Use when selecting a starting recipe, comparing library and benchmark configs, resi…
Use this skill when NeMo Relay is installed or imported but application-side runtime behavior is missing or incorrect, including load failures, inactive scopes, missing events, and plugin or adaptive wiring problems.
When last recorded, these names pointed to this repository. This is a recorded resolution, not a live GitHub check; packages and metadata are never merged because of it.
No aliases pointing to this repository are recorded.
Unsupported marketplace entries 0
Entries in this repository's marketplace that this registry does not support yet. They are preserved source data and are never fetched or executed.