Datasets:
Submission: Phocinae/Phocinae-Largha-150M-v1 (zero-shot), accuracy 0.797 (en) / 0.789 (zh)
Phocinae-Largha-150M-v1 — 144M-param bilingual (en/zh) typed decision model on mmBERT-small (Apache-2.0), trained from scratch on this dataset.
Results (all self-reported; details and eval code in BENCHMARKS.md):
- typed-decisions 0.797 (en) / 0.789 (zh), zero-shot held-out — above the 0.735 human-teacher ceiling; JEV-27B reports 0.727 for comparison
- Confidence (mean of correct answers): 0.769 en / 0.726 zh
- Flip rate (mirror test): 3.0%
- ECE: 0.1313 (per-type temperature scaling)
- Latency: p50 18.6 ms on an RTX 5090 (fp16); about 1.5 s per case on CPU
- Confidence gate: at τ=0.6, 54.4% of decisions resolve locally with 0 output tokens
On JevBench we score 0.5108, below the 0.75 pass bar — disclosed on the model card.
This is an official hf benchmark now, you can add the results to your model card and it should show up on the board, see - https://huggingface.co/docs/hub/en/eval-results
Correction — the original post said "zero-shot" and "trained from scratch"; corrected wording and updated figures (v1.1, 2026-10-09) follow.
Phocinae-Largha-150M-v1 — 144M-param bilingual (en/zh) typed decision model on mmBERT-small (Apache-2.0), fine-tuned on this dataset's train split (with flip-augmented option reorderings). The training mix also includes machine-translated Chinese and native Chinese rows — please read the model as specialist (fitted on the train split), not zero-shot.
Results (all self-reported; details and eval code in BENCHMARKS.md):
- typed-decisions 0.906 (en) / 0.848 (zh, machine-translated cases) on the held-out test split (the train split was used in training) — the en score is above the 0.735 teacher self-agreement reference (the dataset card flags scores far above it as label-specific overfitting); JEV-27B reports 0.727 for comparison
- Confidence (mean of correct answers): 0.6541 en / 0.6546 zh
- Flip rate (mirror test): 0.0217 (release protocol, flip400)
- ECE: 0.2519 (shipped column; optional bundled calibration column in
calib/: 0.0168) — per-type temperature scaling - Latency: p50 21.0 ms on an RTX 5090 (fp16); about 1.64 s per case on CPU
- Confidence gate: at τ=0.6, 45.0% of decisions escalate to a larger model (the rest resolve locally, zero output tokens); LLM calls cut by 55.0% (79.6% at τ=0.5)
On JevBench we score 0.5455 (126/231), below the 58.4% acceptance gate — disclosed on the model card.