[skills] chore: rename memory and moe config skills with nemo mbridge prefix - #4032
Conversation
Signed-off-by: Chen Cui <chcui@nvidia.com>
|
/nvskills-ci |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| templates already include it. | ||
| 2. **Add selective activation recompute** (`recompute_modules=[core_attn]`) if | ||
| not already enabled. See @skills/perf-activation-recompute/SKILL.md. | ||
| not already enabled. See @skills/mbridge-perf-activation-recompute/SKILL.md. |
There was a problem hiding this comment.
Bug: skills/mbridge-perf-activation-recompute/ does not exist. The actual directory is skills/perf-activation-recompute/. This reference (and the other three like it in this file) was incorrectly renamed — mbridge- is not a valid prefix for any existing skill.
If these skills are being renamed in a future PR, leave the old references for now and update them in that PR. Otherwise revert to the original paths:
| not already enabled. See @skills/mbridge-perf-activation-recompute/SKILL.md. | |
| not already enabled. See @skills/perf-activation-recompute/SKILL.md. |
| | Increase TP | Fewer params per GPU | Severe (-28% on 70B) | Last resort | | ||
| | Distributed optimizer | Shards optimizer state across DP ranks | ~1-2% | Recommended for large models | | ||
| | FSDP | Shards params + grads + optimizer | Varies | See @skills/perf-megatron-fsdp/SKILL.md | | ||
| | FSDP | Shards params + grads + optimizer | Varies | See @skills/mbridge-perf-megatron-fsdp/SKILL.md | |
There was a problem hiding this comment.
Same issue: skills/mbridge-perf-megatron-fsdp/ does not exist — the actual path is skills/perf-megatron-fsdp/.
| | FSDP | Shards params + grads + optimizer | Varies | See @skills/mbridge-perf-megatron-fsdp/SKILL.md | | |
| | FSDP | Shards params + grads + optimizer | Varies | See @skills/perf-megatron-fsdp/SKILL.md | |
| ### Activation recompute | ||
|
|
||
| See @skills/perf-activation-recompute/SKILL.md for full details. | ||
| See @skills/mbridge-perf-activation-recompute/SKILL.md for full details. |
There was a problem hiding this comment.
| See @skills/mbridge-perf-activation-recompute/SKILL.md for full details. | |
| See @skills/perf-activation-recompute/SKILL.md for full details. |
| Selective activation recompute with `mlp` saved ~3 GB peak memory but cost | ||
| ~16% GPU utilization on this workload. See | ||
| @skills/perf-activation-recompute/SKILL.md for full results. | ||
| @skills/mbridge-perf-activation-recompute/SKILL.md for full results. |
There was a problem hiding this comment.
| @skills/mbridge-perf-activation-recompute/SKILL.md for full results. | |
| @skills/perf-activation-recompute/SKILL.md for full results. |
Light ReviewBroken cross-references in nemo-mbridge-perf-memory-tuning/SKILL.md (bug) Four internal references were renamed to a mbridge- prefix that does not match any existing directory:
These should either stay as-is (pointing to the current names) or be updated in the same PR that renames those skills. See inline comments with suggested fixes. Stale references in files not touched by this PR Several files still reference the old (pre-rename) skill paths that no longer exist after this PR:
These should be updated in this PR to point to the new nemo-mbridge- prefixed paths. Suggested test cases No perf tests impacted. |
yaoyu-33
added
area:misc
Signed-off-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com>
Signed-off-by: Chen Cui <chcui@nvidia.com>
|
/nvskills-ci |
Signed-off-by: Chen Cui <chcui@nvidia.com>
|
Scoped this signing PR back down to the renamed skill directories only. The cross-skill reference updates that touched old skill directories were moved to draft cleanup PR #4044 so NVSkills should not validate unrelated old skill dirs in this batch. |
|
/nvskills-ci |
Signed-off-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com>
|
/ok to test b431a82 |
Signed-off-by: Chen Cui <chcui@nvidia.com>
… prefix (NVIDIA-NeMo#4032) Signed-off-by: Chen Cui <chcui@nvidia.com> Signed-off-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com> Co-authored-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com> Signed-off-by: Vasudevan Rengasamy <vrengasamy@nvidia.com>
3 participants
Summary
nemo-mbridge-*prefix:nemo-mbridge-perf-memory-tuningnemo-mbridge-perf-moe-dispatcher-selectionnemo-mbridge-perf-moe-hardware-configslicense: Apache-2.0frontmatter for these public skills.evals/evals.jsonfiles where needed for NVSkills signing.docs/skills-index.mdreferences for the renamed skills.Validation
SKILL.mdfrontmatter names.license: Apache-2.0.evals/evals.jsonfiles.SKILL.mdfiles are <= 500 lines.git diff --checkpassed.uv run --no-sync pre-commit run --all-filespassed.