diff --git a/docs/skills-index.md b/docs/skills-index.md index c6d76dcff9..f8fd1d0d2d 100644 --- a/docs/skills-index.md +++ b/docs/skills-index.md @@ -34,6 +34,7 @@ skills/nemo-rl-e2e-testing/SKILL :caption: Performance Techniques :maxdepth: 1 +skills/nemo-mbridge-perf-cpu-offloading/SKILL skills/nemo-mbridge-perf-moe-long-context/SKILL skills/nemo-mbridge-perf-moe-optimization-workflow/SKILL skills/nemo-mbridge-perf-moe-vlm-training/SKILL @@ -63,5 +64,5 @@ skills/nemo-mbridge-multi-node-slurm/references/templates :caption: Resiliency :maxdepth: 1 -skills/resiliency/SKILL +skills/nemo-mbridge-resiliency/SKILL ``` diff --git a/docs/training/cpu-offloading.md b/docs/training/cpu-offloading.md index 81ee640685..56a8a239ff 100644 --- a/docs/training/cpu-offloading.md +++ b/docs/training/cpu-offloading.md @@ -5,7 +5,7 @@ during training, trading throughput for the ability to train models or configurations that would otherwise not fit in GPU memory. For operational setup, code anchors, and verification commands, see -[skills/perf-cpu-offloading/SKILL.md](../skills/perf-cpu-offloading/SKILL.md). +[skills/nemo-mbridge-perf-cpu-offloading/SKILL.md](../skills/nemo-mbridge-perf-cpu-offloading/SKILL.md). ## What It Is @@ -96,7 +96,7 @@ CPU offloading is configured through two independent config namespaces: `model.cpu_offloading_num_layers`, and related `model.cpu_offloading_*` fields For config examples, parameter tables, and runnable commands, see -[skills/perf-cpu-offloading/SKILL.md](../skills/perf-cpu-offloading/SKILL.md). +[skills/nemo-mbridge-perf-cpu-offloading/SKILL.md](../skills/nemo-mbridge-perf-cpu-offloading/SKILL.md). ## Common Failure Modes @@ -115,4 +115,4 @@ For config examples, parameter tables, and runnable commands, see - [docs/training/megatron-fsdp.md](megatron-fsdp.md) - [docs/training/optimizer-scheduler.md](optimizer-scheduler.md) - [docs/training/cuda-graphs.md](cuda-graphs.md) -- [skills/perf-cpu-offloading/SKILL.md](../skills/perf-cpu-offloading/SKILL.md) +- [skills/nemo-mbridge-perf-cpu-offloading/SKILL.md](../skills/nemo-mbridge-perf-cpu-offloading/SKILL.md) diff --git a/docs/training/resiliency.md b/docs/training/resiliency.md index 0b67c10965..002cf00960 100644 --- a/docs/training/resiliency.md +++ b/docs/training/resiliency.md @@ -7,7 +7,7 @@ due to failures and interruptions during training. This page is the stable overview for what each resiliency feature is, when to use it, and which constraints are durable. For operational setup, config knobs, -parameter tables, code anchors, and verification commands, see [skills/resiliency/SKILL.md](../skills/resiliency/SKILL.md). +parameter tables, code anchors, and verification commands, see [skills/nemo-mbridge-resiliency/SKILL.md](../skills/nemo-mbridge-resiliency/SKILL.md). ## What It Is @@ -124,7 +124,7 @@ Local checkpointing is useful when: - Node-local storage must have sufficient capacity for at least one checkpoint - Replication degree must be configured to survive the expected failure rate -- Requires compatible checkpoint format (see [skills/resiliency/SKILL.md](../skills/resiliency/SKILL.md)) +- Requires compatible checkpoint format (see [skills/nemo-mbridge-resiliency/SKILL.md](../skills/nemo-mbridge-resiliency/SKILL.md)) ## Re-run State Machine @@ -186,7 +186,7 @@ with job-level fault tolerance. - [docs/training/checkpointing.md](checkpointing.md) - [docs/performance-guide.md](../performance-guide.md) -- [skills/resiliency/SKILL.md](../skills/resiliency/SKILL.md) -- [skills/resiliency/card.yaml](../skills/resiliency/card.yaml) +- [skills/nemo-mbridge-resiliency/SKILL.md](../skills/nemo-mbridge-resiliency/SKILL.md) +- [skills/nemo-mbridge-resiliency/card.yaml](../skills/nemo-mbridge-resiliency/card.yaml) - [NVIDIA Resiliency Extension](https://github.com/NVIDIA/nvidia-resiliency-ext) - [In-Process Restart Guide](https://nvidia.github.io/nvidia-resiliency-ext/inprocess/index.html) diff --git a/skills/nemo-mbridge-perf-cpu-offloading/BENCHMARK.md b/skills/nemo-mbridge-perf-cpu-offloading/BENCHMARK.md new file mode 100644 index 0000000000..a2c677a23f --- /dev/null +++ b/skills/nemo-mbridge-perf-cpu-offloading/BENCHMARK.md @@ -0,0 +1,72 @@ +# Evaluation Report + +Evaluation of the `nemo-mbridge-perf-cpu-offloading` skill before publication through NVSkills-Eval. + +This benchmark summarizes 3-Tier Evaluation from NVSkills-Eval results for the skill. The goal is to document whether the skill is safe, discoverable, effective, and useful for agents before it is published for broader workflow use. + +## Evaluation Summary + +- Skill: `nemo-mbridge-perf-cpu-offloading` +- Evaluation date: 2026-05-28 +- NVSkills-Eval profile: `external` +- Overall verdict: FAIL +- Tier 3 live agent evaluation: not available in this report + +## Agents Used + +- Tier 3 agent details were not available in this report. + +## Metrics Used + +Reported benchmark dimensions: + +- Security: checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access. +- Correctness: checks whether the agent follows the expected workflow and produces the correct final output. +- Discoverability: checks whether the agent loads the skill when relevant and avoids using it when irrelevant. +- Effectiveness: checks whether the agent performs measurably better with the skill than without it. +- Efficiency: checks whether the agent uses fewer tokens and avoids redundant work. + +Underlying evaluation signals used in this run: + +- No Tier 3 evaluation signal details were available in this report. + +## Test Tasks + +Tier 3 evaluation task details were not available in this report. + +## Results + +Tier 3 dimension rollup was not available in this report. + +## Tier 1: Static Validation Summary + +Tier 1 validation reported findings. NVSkills-Eval ran 9 checks and found 14 total findings. + +Top findings: + +- MEDIUM PII/gps_coordinates: GPS coordinates (location information) (`SKILL.md:101`) +- MEDIUM QUALITY/quality_correctness: SKILL_SPEC recommended field missing: 'metadata.author' (`skills/nemo-mbridge-perf-cpu-offloading/SKILL.md`) +- MEDIUM QUALITY/quality_correctness: SKILL_SPEC recommended field missing: 'metadata.tags' (`skills/nemo-mbridge-perf-cpu-offloading/SKILL.md`) +- MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Instructions' (`skills/nemo-mbridge-perf-cpu-offloading/SKILL.md`) +- MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Examples' (`skills/nemo-mbridge-perf-cpu-offloading/SKILL.md`) + +## Tier 2: Deduplication Summary + +Tier 2 validation reported findings. NVSkills-Eval ran 2 checks and found 1 total findings. + +Top findings: + +- HIGH DUPLICATE/duplicate: Duplicate content found within SKILL.md: + "### Activation CPU offloading (small/medium models only)" in SKILL.md (lines 46-58) + vs "### Activation offloading" in SKILL.md (lines 70-79) + vs "### Activation offloading" in SKILL.md (lines 82-89) + vs "### Activation offload (small model, PP=1)" in SKILL.md (lines 119-129) + vs "### Weight offload only (small model, PP=1)" in SKILL.md (lines 130-140) + vs "### Both activations and weights (small model, PP=1)" in SKILL.md (lines 141-156) + vs "### MCore activation offload constraints" in SKILL.md (lines 188-206) + vs "### MCore fine-grained offloading mutual exclusion" in SKILL.md (lines 214-222) + vs "### MCore model_parallel_config fields" in SKILL.md (lines 257-267) (`SKILL.md:46`) + +## Publication Recommendation + +The skill should be reviewed before NVSkills-Eval publication. Skill owners should address the findings above and rerun NVSkills-Eval to refresh this benchmark. diff --git a/skills/perf-cpu-offloading/SKILL.md b/skills/nemo-mbridge-perf-cpu-offloading/SKILL.md similarity index 98% rename from skills/perf-cpu-offloading/SKILL.md rename to skills/nemo-mbridge-perf-cpu-offloading/SKILL.md index 16cccca913..1ea0d1ce9a 100644 --- a/skills/perf-cpu-offloading/SKILL.md +++ b/skills/nemo-mbridge-perf-cpu-offloading/SKILL.md @@ -1,6 +1,7 @@ --- -name: perf-cpu-offloading +name: nemo-mbridge-perf-cpu-offloading description: Validate and use CPU offloading in Megatron Bridge, including layer-level activation offloading and fractional optimizer state offloading with HybridDeviceOptimizer. +license: Apache-2.0 when_to_use: Enabling CPU offload to reduce GPU memory, or investigating a commit that changed CPU offloading config and caused OOM or a crash; 'cpu_offloading', 'optimizer_cpu_offload', 'optimizer_offload_fraction', 'HybridDeviceOptimizer', 'move optimizer to CPU'. --- @@ -9,7 +10,7 @@ when_to_use: Enabling CPU offload to reduce GPU memory, or investigating a commi ## References - Stable docs: @docs/training/cpu-offloading.md -- Structured metadata: @skills/perf-cpu-offloading/card.yaml +- Structured metadata: @skills/nemo-mbridge-perf-cpu-offloading/card.yaml ## What It Is diff --git a/skills/perf-cpu-offloading/card.yaml b/skills/nemo-mbridge-perf-cpu-offloading/card.yaml similarity index 100% rename from skills/perf-cpu-offloading/card.yaml rename to skills/nemo-mbridge-perf-cpu-offloading/card.yaml diff --git a/skills/nemo-mbridge-perf-cpu-offloading/evals/evals.json b/skills/nemo-mbridge-perf-cpu-offloading/evals/evals.json new file mode 100644 index 0000000000..fe51488c70 --- /dev/null +++ b/skills/nemo-mbridge-perf-cpu-offloading/evals/evals.json @@ -0,0 +1 @@ +[] diff --git a/skills/nemo-mbridge-perf-cpu-offloading/skill-card.md b/skills/nemo-mbridge-perf-cpu-offloading/skill-card.md new file mode 100644 index 0000000000..29f33c550e --- /dev/null +++ b/skills/nemo-mbridge-perf-cpu-offloading/skill-card.md @@ -0,0 +1,53 @@ +## Description:
+Validate and use CPU offloading in Megatron Bridge, including layer-level activation offloading and fractional optimizer state offloading with HybridDeviceOptimizer.
+ +This skill is ready for commercial/non-commercial use.
+ +## Owner +NVIDIA
+ +### License/Terms of Use:
+Apache 2.0
+## Use Case:
+Developers and engineers enabling CPU offload to reduce GPU memory pressure during LLM training, or investigating commits that changed CPU offloading configuration and caused OOM or crashes.
+ +### Deployment Geography for Use:
+Global
+ +## Known Risks and Mitigations:
+Risk: Review before execution as proposals could introduce incorrect or misleading guidance into skills.
+Mitigation: Review and scan skill before deployment.
+ +## Reference(s):
+- [CPU Offloading Documentation](docs/training/cpu-offloading.md)
+- [Performance Tuning Guide](docs/performance-guide.md)
+- [Skill Structured Metadata](skills/nemo-mbridge-perf-cpu-offloading/card.yaml)
+ + +## Skill Output:
+**Output Type(s):** [Configuration instructions, Shell commands, Analysis]
+**Output Format:** [Markdown with inline Python and bash code blocks]
+**Output Parameters:** [1D]
+**Other Properties Related to Output:** [None]
+ +## Evaluation Tasks:
+Evaluated via NVSkills-Eval 3-Tier Evaluation (external profile) on 2026-05-28. Tier 1 static validation ran 9 checks with 14 findings. Tier 2 deduplication ran 2 checks with 1 finding. Tier 3 live agent evaluation was not available.
+ +## Evaluation Metrics Used:
+Reported benchmark dimensions:
+- Security: Checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access.
+- Correctness: Checks whether the agent follows the expected workflow and produces the correct final output.
+- Discoverability: Checks whether the agent loads the skill when relevant and avoids using it when irrelevant.
+- Effectiveness: Checks whether the agent performs measurably better with the skill than without it.
+- Efficiency: Checks whether the agent uses fewer tokens and avoids redundant work.
+ + + +## Skill Version(s):
+b058d061 (source: git SHA, committed 2026-05-28)
+ +## Ethical Considerations:
+NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal team to ensure this skill meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
+ +(For Release on NVIDIA Platforms Only)
+Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns [here](https://app.intigriti.com/programs/nvidia/nvidiavdp/detail).
diff --git a/skills/nemo-mbridge-perf-cpu-offloading/skill.oms.sig b/skills/nemo-mbridge-perf-cpu-offloading/skill.oms.sig new file mode 100644 index 0000000000..92947f761d --- /dev/null +++ b/skills/nemo-mbridge-perf-cpu-offloading/skill.oms.sig @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json","verificationMaterial":{"x509CertificateChain":{"certificates":[{"rawBytes":"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"},{"rawBytes":"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"},{"rawBytes":"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"}]},"tlogEntries":[]},"dsseEnvelope":{"payload":"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","payloadType":"application/vnd.in-toto+json","signatures":[{"sig":"MGQCMFt1geIwKurmInB7R6Byvc2dyzKNpcZAVAwC6YlEBQCSQPv6gp1SFOJT5WcpOUwJ+AIwIef9cewjrZpG+AphYzG1VUYF+KBsegzL/Wn5sBBBWf+i/n9UnAZYYxoQpM5jDVJs","keyid":""}]}} \ No newline at end of file diff --git a/skills/nemo-mbridge-resiliency/BENCHMARK.md b/skills/nemo-mbridge-resiliency/BENCHMARK.md new file mode 100644 index 0000000000..ce93c99e27 --- /dev/null +++ b/skills/nemo-mbridge-resiliency/BENCHMARK.md @@ -0,0 +1,64 @@ +# Evaluation Report + +Evaluation of the `nemo-mbridge-resiliency` skill before publication through NVSkills-Eval. + +This benchmark summarizes 3-Tier Evaluation from NVSkills-Eval results for the skill. The goal is to document whether the skill is safe, discoverable, effective, and useful for agents before it is published for broader workflow use. + +## Evaluation Summary + +- Skill: `nemo-mbridge-resiliency` +- Evaluation date: 2026-05-28 +- NVSkills-Eval profile: `external` +- Overall verdict: FAIL +- Tier 3 live agent evaluation: not available in this report + +## Agents Used + +- Tier 3 agent details were not available in this report. + +## Metrics Used + +Reported benchmark dimensions: + +- Security: checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access. +- Correctness: checks whether the agent follows the expected workflow and produces the correct final output. +- Discoverability: checks whether the agent loads the skill when relevant and avoids using it when irrelevant. +- Effectiveness: checks whether the agent performs measurably better with the skill than without it. +- Efficiency: checks whether the agent uses fewer tokens and avoids redundant work. + +Underlying evaluation signals used in this run: + +- No Tier 3 evaluation signal details were available in this report. + +## Test Tasks + +Tier 3 evaluation task details were not available in this report. + +## Results + +Tier 3 dimension rollup was not available in this report. + +## Tier 1: Static Validation Summary + +Tier 1 validation reported findings. NVSkills-Eval ran 9 checks and found 15 total findings. + +Top findings: + +- MEDIUM QUALITY/quality_correctness: SKILL_SPEC recommended field missing: 'metadata.author' (`skills/nemo-mbridge-resiliency/SKILL.md`) +- MEDIUM QUALITY/quality_correctness: SKILL_SPEC recommended field missing: 'metadata.tags' (`skills/nemo-mbridge-resiliency/SKILL.md`) +- MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Instructions' (`skills/nemo-mbridge-resiliency/SKILL.md`) +- MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Examples' (`skills/nemo-mbridge-resiliency/SKILL.md`) +- MEDIUM SCHEMA/author_missing: Author not specified in metadata (`skills/nemo-mbridge-resiliency/SKILL.md`) + +## Tier 2: Deduplication Summary + +Tier 2 validation passed. NVSkills-Eval ran 2 checks and found 0 total findings. + +Notable observations: + +- Context Deduplication: Collected 1 file(s) +- Inter-Skill Deduplication: Parsed skill 'nemo-mbridge-resiliency': 144 char description + +## Publication Recommendation + +The skill should be reviewed before NVSkills-Eval publication. Skill owners should address the findings above and rerun NVSkills-Eval to refresh this benchmark. diff --git a/skills/resiliency/SKILL.md b/skills/nemo-mbridge-resiliency/SKILL.md similarity index 94% rename from skills/resiliency/SKILL.md rename to skills/nemo-mbridge-resiliency/SKILL.md index a8124f0669..643ade7933 100644 --- a/skills/resiliency/SKILL.md +++ b/skills/nemo-mbridge-resiliency/SKILL.md @@ -1,13 +1,14 @@ --- -name: resiliency +name: nemo-mbridge-resiliency description: Resiliency features in Megatron Bridge including fault tolerance, straggler detection, in-process restart, preemption, and re-run state machine. +license: Apache-2.0 when_to_use: Enabling resiliency features, or investigating a commit that caused training hangs, straggler detection failures, or broken restarts; 'fault tolerance', 'straggler detection', 'hang detection', 'automatic restart', 'in-process restart', 'preemption', 'nvidia-resiliency-ext'. --- # Resiliency Stable docs: @docs/training/resiliency.md, @docs/training/checkpointing.md -Card: @skills/resiliency/card.yaml +Card: @skills/nemo-mbridge-resiliency/card.yaml ## Enablement @@ -221,16 +222,16 @@ cfg.checkpoint.non_persistent_local_ckpt_algo = "fully_parallel" - Config: `src/megatron/bridge/training/config.py` — `FaultToleranceConfig` - Runtime: `src/megatron/bridge/training/fault_tolerance.py` - Plugin: `src/megatron/bridge/recipes/run_plugins.py` — `FaultTolerancePlugin` -- Perf plugin: `scripts/performance/resiliency_plugins.py` +- Perf plugin: `scripts/performance/nemo-mbridge-resiliency_plugins.py` - Tests: `tests/unit_tests/training/test_fault_tolerance.py` -- Example: `examples/training_features/resiliency/fault_tolerance/` +- Example: `examples/training_features/nemo-mbridge-resiliency/fault_tolerance/` ### Straggler detection - Config: `src/megatron/bridge/training/config.py` — `NVRxStragglerDetectionConfig` - Runtime: `src/megatron/bridge/training/nvrx_straggler.py` - Train loop: `src/megatron/bridge/training/train.py` — `check_nvrx_straggler_detection` - Tests: `tests/unit_tests/training/test_nvrx_straggler.py`, `tests/functional_tests/training/test_nvrx_straggler.py` -- Example: `examples/training_features/resiliency/straggler_detection/` +- Example: `examples/training_features/nemo-mbridge-resiliency/straggler_detection/` ### In-process restart - Config: `src/megatron/bridge/training/config.py` — `InProcessRestartConfig` @@ -281,8 +282,8 @@ cfg.checkpoint.non_persistent_local_ckpt_algo = "fully_parallel" ### Fault tolerance ```bash -./examples/training_features/resiliency/fault_tolerance/run_fault_tolerance.sh -./examples/training_features/resiliency/fault_tolerance/run_fault_tolerance.sh --simulate-fault +./examples/training_features/nemo-mbridge-resiliency/fault_tolerance/run_fault_tolerance.sh +./examples/training_features/nemo-mbridge-resiliency/fault_tolerance/run_fault_tolerance.sh --simulate-fault ``` Look for `[FaultTolerance]` / `[RankMonitorServer]` log lines with section timeouts. Simulated fault should trigger restart from checkpoint. @@ -290,7 +291,7 @@ timeouts. Simulated fault should trigger restart from checkpoint. ### Straggler detection ```bash uv run python -m torch.distributed.run --nproc_per_node=2 \ - examples/training_features/resiliency/straggler_detection/straggler_detection_example.py + examples/training_features/nemo-mbridge-resiliency/straggler_detection/straggler_detection_example.py ``` Look for `GPU relative performance` and `GPU individual performance` reports with per-rank scores. diff --git a/skills/resiliency/card.yaml b/skills/nemo-mbridge-resiliency/card.yaml similarity index 95% rename from skills/resiliency/card.yaml rename to skills/nemo-mbridge-resiliency/card.yaml index 651be11a81..ad6be8b779 100644 --- a/skills/resiliency/card.yaml +++ b/skills/nemo-mbridge-resiliency/card.yaml @@ -103,7 +103,7 @@ evidence: - src/megatron/bridge/training/config.py - src/megatron/bridge/training/utils/sig_utils.py - src/megatron/bridge/recipes/run_plugins.py - - scripts/performance/resiliency_plugins.py + - scripts/performance/nemo-mbridge-resiliency_plugins.py - tests/unit_tests/training/test_fault_tolerance.py - tests/unit_tests/training/test_nvrx_straggler.py - tests/unit_tests/training/test_inprocess_restart.py @@ -111,8 +111,8 @@ evidence: - tests/functional_tests/training/test_nvrx_straggler.py - tests/functional_tests/training/test_inprocess_restart.py - tests/functional_tests/training/test_local_checkpointing.py - - examples/training_features/resiliency/fault_tolerance/ - - examples/training_features/resiliency/straggler_detection/ + - examples/training_features/nemo-mbridge-resiliency/fault_tolerance/ + - examples/training_features/nemo-mbridge-resiliency/straggler_detection/ follow_up_validation: - Add a Slurm-based FT end-to-end test with actual hang recovery. - Add a checked-in recipe that enables FT + straggler detection by default. diff --git a/skills/nemo-mbridge-resiliency/evals/evals.json b/skills/nemo-mbridge-resiliency/evals/evals.json new file mode 100644 index 0000000000..fe51488c70 --- /dev/null +++ b/skills/nemo-mbridge-resiliency/evals/evals.json @@ -0,0 +1 @@ +[] diff --git a/skills/nemo-mbridge-resiliency/skill-card.md b/skills/nemo-mbridge-resiliency/skill-card.md new file mode 100644 index 0000000000..80a94fc5b4 --- /dev/null +++ b/skills/nemo-mbridge-resiliency/skill-card.md @@ -0,0 +1,53 @@ +## Description:
+Resiliency features in Megatron Bridge including fault tolerance, straggler detection, in-process restart, preemption, and re-run state machine.
+ +This skill is ready for commercial/non-commercial use.
+ +## Owner +NVIDIA
+ +### License/Terms of Use:
+Apache 2.0
+## Use Case:
+Developers and engineers enabling resiliency features for distributed GPU training, including fault tolerance, straggler detection, preemption handling, and automatic restart capabilities in Megatron Bridge workloads.
+ +### Deployment Geography for Use:
+Global
+ +## Known Risks and Mitigations:
+Risk: Review before execution as proposals could introduce incorrect or misleading guidance into skills.
+Mitigation: Review and scan skill before deployment.
+ +## Reference(s):
+- [Resiliency Documentation](docs/training/resiliency.md)
+- [Checkpointing Documentation](docs/training/checkpointing.md)
+- [NVIDIA NeMo Megatron Bridge Documentation](https://docs.nvidia.com/nemo/megatron-bridge/latest/)
+ + +## Skill Output:
+**Output Type(s):** [Configuration instructions, Shell commands, Code]
+**Output Format:** [Markdown with inline Python and bash code blocks]
+**Output Parameters:** [1D]
+**Other Properties Related to Output:** [None]
+ +## Evaluation Tasks:
+Evaluated through NVSkills-Eval 3-Tier evaluation framework with external profile on 2026-05-28.
+ +## Evaluation Metrics Used:
+Reported benchmark dimensions:
+- Security: Checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access.
+- Correctness: Checks whether the agent follows the expected workflow and produces the correct final output.
+- Discoverability: Checks whether the agent loads the skill when relevant and avoids using it when irrelevant.
+- Effectiveness: Checks whether the agent performs measurably better with the skill than without it.
+- Efficiency: Checks whether the agent uses fewer tokens and avoids redundant work.
+ + + +## Skill Version(s):
+v0.2.0rc6-1467-gb058d061 (source: git tag)
+ +## Ethical Considerations:
+NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal team to ensure this skill meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
+ +(For Release on NVIDIA Platforms Only)
+Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns [here](https://app.intigriti.com/programs/nvidia/nvidiavdp/detail).
diff --git a/skills/nemo-mbridge-resiliency/skill.oms.sig b/skills/nemo-mbridge-resiliency/skill.oms.sig new file mode 100644 index 0000000000..7635fe193c --- /dev/null +++ b/skills/nemo-mbridge-resiliency/skill.oms.sig @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json","verificationMaterial":{"x509CertificateChain":{"certificates":[{"rawBytes":"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"},{"rawBytes":"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"},{"rawBytes":"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"}]},"tlogEntries":[]},"dsseEnvelope":{"payload":"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","payloadType":"application/vnd.in-toto+json","signatures":[{"sig":"MGUCMCu+JH7t0ghpQud1jLocPWK69WwKongxzNbxDKABOH9MqqZE5MrsfUrncgoUxNAAtgIxAMJ5qq14V7U3Y7Bypc2mdU91q3xdtPSlkhgYppZ+XSiJeWvgNEFhj1ChBS9E0adX0w==","keyid":""}]}} \ No newline at end of file