Validation Suite: 10 Industrial Tasks
Select an automation task to compare stock base model drift vs. fine-tuned adapter outputs:
❌ Stock Ornith-1.5-9B (Base)
SYNTAX DRIFT
// Loading base failure trace...
✅ adarrshDev/ornith-1.5-9b-sft-r16
COMPILABLE IEC 61131-3 ✓
// Loading adapter Structured Text...
⚡ Run Live Neural Model via Edge Proxy
Send any custom industrial automation prompt. Requests proxy through Cloudflare Edge (api.paulcreates.online) running Cloudflare Workers AI (@cf/meta/llama-3.2-3b-instruct) paired with deterministic AST auto-healing. To execute genuine 9B SFT adapter weights (adarrshDev/ornith-1.5-9b-sft-r16), use the copyable Python PEFT script below.
Status: Standby • Ready to dispatch
Latency: 0 ms
// Output from live model will stream here...
Empirical Benchmark Summary (23-Task Harness)
Measured across 23 pinned industrial verification tasks (eval/plc/) against deterministic IronPLC & MATIEC compiler gates:
| Evaluation Category | Task Count | Target Verification Gate | Empirical Pass Rate | Status |
|---|---|---|---|---|
| IEC 61131-3 ST Compilation | 10 Tasks | ironplc-plc2x Stage 4 AST & Type Resolution |
9 / 10 (90.0%) | 10/10 with AST Auto-Heal |
| Specification to Logic | 5 Tasks | Spec constraints & safety invariant completeness | 5 / 5 (100.0%) | Verified |
| Digital Twin Assertions | 3 Tasks | Dynamic tank differential physics & conveyor assertion loops | 3 / 3 (100.0%) | Verified |
| IIoT Protocol Extraction | 5 Tasks | OPC UA (asyncua), Sparkplug B & SenML canonical mapping | 5 / 5 (100.0%) | Verified |
| Total Multi-Gate Benchmark | 23 Tasks | Closed-Loop Multi-Gate Verification | 22 / 23 (95.7%) | Production Certified |