Datasets:

---

LocalLLaMA

/

typed-decisions

Like 159

Follow

LocalLLaMA 216

This PR adds Phocinae-Largha-150M-v1 (our model) to Table 2 (Fitted or fine-tuned on train), referencing benchmark discussion #18.

• Model repository: Phocinae/Phocinae-Largha-150M-v1
• Architecture: mmBERT-small encoder, 144.3M params — single forward pass, 0 output tokens
• Mode: specialist — fitted on this dataset's train split (with flip-augmented option reorderings)
• Metrics (400 cases / 2,000 decisions, self-reported): Accuracy 0.906 · KL from gold 0.0548 · Brier 0.0248 · ECE 0.2519 (our definition) · p50 latency 21.0 ms (RTX 5090, fp16)
• Context: the score is above the 0.735 teacher self-agreement reference — per the dataset card's own note, we flag that as a label-specific-overfitting caution (disclosed on our card and in BENCHMARKS.md); zh (machine-translated cases) 0.848.
• Artifacts: full numbers, methodology and evaluation code: BENCHMARKS.md.