Datasets:

---

LocalLLaMA

/

typed-decisions

Like 166

Follow

LocalLLaMA 218

This PR adds ProKope-421M-8K (conceived and engineered by Alan Canto) to Table 2 (Fitted or fine-tuned on train), referencing benchmark discussion #23.

• Model Repository: ProKope-AI/ProKope-421M-8K
• Creator & Lead Architect: Alan Canto
• Framework Architect: Wesley Foreman
• Architecture: 421.3M Non-Autoregressive Decision Engine (ModernBERT-large + Geodesic SLERP Multi-Head)
• Context Capacity: 8,192 Tokens (Full Enterprise Window)
• Mode: Specialist (fine-tuned on train split, single forward pass, 0 output tokens)
• Verified Metrics (400 cases / 2,000 decisions):
  • Accuracy: 0.7185 (71.85% — 1,437 / 2,000 correct)
  • Boolean (noul) Accuracy: 0.8050 (483 / 600)
  • Brier Score: 0.3847
  • ECE: 0.4302
  • Latency: 42.94 ms/dec (sub-25ms single-decision pass)
• Artifacts: Verified .eval_results/typed-decisions.yaml, benchmark_scorecard.json, and all 2,000 eval_predictions.jsonl are live in the model repository.