train_500k / README.md
athaze's picture
Update README.md
84f4204 verified
|
Raw
History Blame
1.04 kB
metadata
dataset_info:
  - config_name: hi-ml
    features:
      - name: audio_filepath
        dtype: audio
      - name: text
        dtype: string
      - name: duration
        dtype: float64
      - name: target_lang
        dtype: string
      - name: source_lang
        dtype: string
      - name: pnc
        dtype: string
      - name: hi-ml
        dtype: string
    splits:
      - name: train
        num_bytes: 114807371510
        num_examples: 475187
    download_size: 114968073480
    dataset_size: 114807371510
configs:
  - config_name: hi-ml
    data_files:
      - split: train
        path: data/train-*
license: cc-by-nc-4.0

hi-ml Dataset Summary

Metric Value
Audio Duration 996.72 hours

Translation Quality Report (COMET Kiwi‑DA)

Metric Value
Dataset hi-ml
Sentence pairs 475,181
COMET mean score 0.8087
Score (0‑100) 80.87

💡 Note: COMET Kiwi-DA scores range from 0–1, where higher values indicate better translation quality. A score of 80.87/100 suggests strong translation performance.