Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Fully sharded data-parallel training for large models.
.claude/skills/nousresearch-pytorch-fsdp/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 7 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-17 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 32% | 0% |
| case-10 | ✓→✓ | ＝ Same ✓ | -28% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | -24% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 37% | 0% |
Assistance with pytorch-fsdp development, generated from official documentation.
This skill should be triggered when:
The full common-patterns catalog (~157k chars of runnable FSDP snippets) lives in references/common-patterns.md — load it with read_file when you need wrapping, sharding-strategy, checkpoint, or mixed-precision examples. Start there rather than reconstructing FSDP incantations from memory.
This skill includes comprehensive documentation in references/:
Use view to read specific reference files when detailed information is needed.
Start with the getting_started or tutorials reference files for foundational concepts.
Use the appropriate category reference file (api, guides, etc.) for detailed information.
The quick reference section above contains common patterns extracted from the official docs.
Organized documentation extracted from official sources. These files contain:
Add helper scripts here for common automation tasks.
Add templates, boilerplate, or example projects here.
To refresh this skill with updated documentation:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 5,585 | 4,868 | -13% | 1 | 1 | 0% | 1,044 | 1,381 | +32% | 0 | 0 | — |
case-10 | pass→pass | 14,333 | 10,867 | -24% | 1 | 1 | 0% | 2,734 | 1,973 | -28% | 0 | 0 | — |
case-02 | pass→pass | 10,893 | 6,024 | -45% | 1 | 1 | 0% | 2,073 | 1,583 | -24% | 0 | 0 | — |
case-03 | pass→pass | 4,341 | 3,734 | -14% | 1 | 1 | 0% | 861 | 1,177 | +37% | 0 | 0 | — |
case-04 | pass→pass | 6,126 | 3,416 | -44% | 1 | 1 | 0% | 1,123 | 1,096 | -2% | 0 | 0 | — |
case-05 | pass→pass | 8,354 | 3,842 | -54% | 1 | 1 | 0% | 1,627 | 1,122 | -31% | 0 | 0 | — |
case-06 | pass→pass | 6,967 | 4,817 | -31% | 1 | 1 | 0% | 1,284 | 1,369 | +7% | 0 | 0 | — |
case-07 | pass→pass | 7,024 | 7,179 | +2% | 1 | 1 | 0% | 1,398 | 1,934 | +38% | 0 | 0 | — |
case-08 | pass→pass | 3,844 | 3,704 | -4% | 1 | 1 | 0% | 720 | 1,123 | +56% | 0 | 0 | — |
case-09 | pass→pass | 5,728 | 5,227 | -9% | 1 | 1 | 0% | 1,047 | 1,413 | +35% | 0 | 0 | — |
case-11 | pass→pass | 7,089 | 7,431 | +5% | 1 | 1 | 0% | 1,423 | 2,299 | +62% | 0 | 0 | — |
case-12 | pass→pass | 7,165 | 4,357 | -39% | 1 | 1 | 0% | 1,500 | 1,305 | -13% | 0 | 0 | — |
case-13 | pass→pass | 4,482 | 4,432 | -1% | 1 | 1 | 0% | 1,044 | 1,401 | +34% | 0 | 0 | — |
case-14 | pass→pass | 5,277 | 4,292 | -19% | 1 | 1 | 0% | 1,302 | 1,316 | +1% | 0 | 0 | — |
case-15 | pass→pass | 4,606 | 3,218 | -30% | 1 | 1 | 0% | 890 | 1,124 | +26% | 0 | 0 | — |
case-16 | pass→pass | 3,937 | 3,305 | -16% | 1 | 1 | 0% | 738 | 1,105 | +50% | 0 | 0 | — |
case-17 | fail→pass | 5,817 | 5,366 | -8% | 1 | 1 | 0% | 1,146 | 1,477 | +29% | 0 | 0 | — |
case-18 | pass→pass | 11,776 | 6,653 | -44% | 1 | 1 | 0% | 2,382 | 1,788 | -25% | 0 | 0 | — |
case-19 | pass→pass | 6,607 | 6,756 | +2% | 1 | 1 | 0% | 1,462 | 1,705 | +17% | 0 | 0 | — |
case-20 | pass→pass | 8,558 | 10,757 | +26% | 1 | 1 | 0% | 1,796 | 2,247 | +25% | 0 | 0 | — |
case-21 | pass→pass | 13,892 | 13,768 | -1% | 1 | 1 | 0% | 2,990 | 3,004 | +0% | 0 | 0 | — |
case-22 | pass→pass | 18,595 | 16,580 | -11% | 1 | 1 | 0% | 3,627 | 3,705 | +2% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/8/2026 | — |
Other measured skills in the registry, with their headline benchmark lift.