Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Configure RuVLLM local inference with model selection, MicroLoRA fine-tuning, and SONA adaptation
.claude/skills/ruvnet-llm-config/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -51% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -16% | 0% |
Configure RuVLLM for local inference and fine-tuning.
When you need to configure local LLM inference, create MicroLoRA adapters for task-specific fine-tuning, or set up SONA for real-time adaptation.
mcp__plugin_ruflo-core_ruflo__ruvllm_status to see current model and adapter statemcp__plugin_ruflo-core_ruflo__ruvllm_generate_config with model parametersmcp__plugin_ruflo-core_ruflo__ruvllm_microlora_create for task-specific adaptersmcp__plugin_ruflo-core_ruflo__ruvllm_microlora_adapt with training datamcp__plugin_ruflo-core_ruflo__ruvllm_sona_create for real-time neural adaptationmcp__plugin_ruflo-core_ruflo__ruvllm_sona_adapt with feedback signals| Feature | MicroLoRA | SONA | |---------|-----------|------| | Speed | Minutes to train | <0.05ms adaptation | | Scope | Task-specific fine-tuning | Real-time micro-adjustments | | Persistence | Saved as adapter weights | Session-scoped | | Use case | Specialized domain tasks | Continuous feedback loops |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 14,736 | 2,637 | -82% | 1 | 1 | 0% | 3,032 | 810 | -73% | 0 | 0 | — |
case-02 | fail→fail | 20,702 | 2,065 | -90% | 1 | 1 | 0% | 4,140 | 781 | -81% | 0 | 0 | — |
case-03 | fail→fail | 23,044 | 1,559 | -93% | 1 | 1 | 0% | 5,277 | 688 | -87% | 0 | 0 | — |
case-04 | pass→pass | 9,741 | 8,043 | -17% | 1 | 1 | 0% | 1,523 | 2,079 | +37% | 0 | 0 | — |
case-05 | pass→pass | 18,644 | 15,455 | -17% | 1 | 1 | 0% | 3,672 | 3,461 | -6% | 0 | 0 | — |
case-06 | pass→pass | 17,794 | 14,640 | -18% | 1 | 1 | 0% | 3,693 | 3,477 | -6% | 0 | 0 | — |
case-07 | fail→pass | 9,953 | 2,558 | -74% | 1 | 1 | 0% | 1,955 | 887 | -55% | 0 | 0 | — |
case-08 | fail→fail | 9,794 | 5,645 | -42% | 1 | 1 | 0% | 1,858 | 675 | -64% | 0 | 0 | — |
case-09 | fail→pass | 13,545 | 4,591 | -66% | 1 | 1 | 0% | 2,434 | 1,408 | -42% | 0 | 0 | — |
case-10 | fail→pass | 10,442 | 3,282 | -69% | 1 | 1 | 0% | 1,861 | 1,007 | -46% | 0 | 0 | — |
case-11 | fail→pass | 14,484 | 5,145 | -64% | 1 | 1 | 0% | 2,676 | 1,302 | -51% | 0 | 0 | — |
case-12 | fail→pass | 6,502 | 2,783 | -57% | 1 | 1 | 0% | 1,090 | 921 | -16% | 0 | 0 | — |
case-13 | pass→pass | 7,280 | 1,802 | -75% | 1 | 1 | 0% | 1,233 | 773 | -37% | 0 | 0 | — |
case-14 | pass→pass | 10,380 | 2,982 | -71% | 1 | 1 | 0% | 1,756 | 987 | -44% | 0 | 0 | — |
case-15 | fail→pass | 5,559 | 2,266 | -59% | 1 | 1 | 0% | 924 | 762 | -18% | 0 | 0 | — |
case-16 | pass→pass | 11,427 | 3,924 | -66% | 1 | 1 | 0% | 2,006 | 1,182 | -41% | 0 | 0 | — |
case-17 | fail→pass | 14,058 | 5,943 | -58% | 1 | 1 | 0% | 2,639 | 1,569 | -41% | 0 | 0 | — |
case-18 | fail→pass | 28,095 | 3,706 | -87% | 1 | 1 | 0% | 1,075 | 1,243 | +16% | 0 | 0 | — |
case-19 | fail→pass | 7,549 | 3,356 | -56% | 1 | 1 | 0% | 1,404 | 614 | -56% | 0 | 0 | — |
case-20 | fail→pass | 10,057 | 2,137 | -79% | 1 | 1 | 0% | 1,902 | 805 | -58% | 0 | 0 | — |
case-21 | fail→pass | 7,787 | 1,548 | -80% | 1 | 1 | 0% | 1,400 | 652 | -53% | 0 | 0 | — |
case-22 | fail→pass | 8,454 | 1,453 | -83% | 1 | 1 | 0% | 1,506 | 643 | -57% | 0 | 0 | — |
case-23 | pass→pass | 21,322 | 2,663 | -88% | 1 | 1 | 0% | 1,732 | 853 | -51% | 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. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +52 percentage points is the difference between those two pass rates over the 21 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.
Other measured skills in the registry, with their headline benchmark lift.