teacher-labels
neyshekar-fa-wimp-teacher-labels
Persian word-importance teacher labels (neyshekar v4)
Per-token word-importance soft labels for a DHH-oriented semantic-WER (ACE-style) metric.
Teacher: google/gemma-4-31b-it (OpenRouter, ModelRun fp4) + 5 Persian few-shot examples.
Source text: shekar-ai/neyshekar-v4-persian-asr-fa (text only, deduplicated).
Utterances: 26490 ok (3 format-fails dropped).
Scale: 0.0–1.0 per whitespace token (0 = filler/function word, 1 = essential content).
Provenance: distilled with no human… See the full description on the dataset page: https://huggingface.co/datasets/Reza2kn/neyshekar-fa-wimp-teacher-labels.gpt51-teacher-labels
GPT-5.1 Teacher Labels for SLM Distillation
Anonymous NeurIPS 2026 E&D companion dataset. This repository contains
derived teacher-label metadata only: row identifiers, human labels inherited
from the source datasets, GPT-5.1 teacher labels, and verdict-token top-logprob
vectors. It does not redistribute source query/passage text, resumes, or
job descriptions.
Files
data/llmjudge_gpt51_teacher/dev.jsonl
data/llmjudge_gpt51_teacher/test.jsonl… See the full description on the dataset page: https://huggingface.co/datasets/anonymous2026whl/gpt51-teacher-labels.sangeetkar-teacher-labels
Ostrich: Surgical Teacher-Generated Mood Beliefs
High-precision continuous mood beliefs generated by a CLAP teacher model
for knowledge distillation into the lightweight Ostrich student.
Dataset Stats
Samples: 6862
Window: First 15 seconds
Values: Continuous probabilities [0, 1]
Emotions
Energetic, Calm, Happy, Sad, Angry, Romantic, Mysterious, Nostalgic
Usage
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/beastLucifer/sangeetkar-teacher-labels.