bad49wolf's picture
add tokenized shard data/video=-RywY2MP3SU/tokens.parquet
80051d3 verified
---
language:
- ar
task_categories:
- text-to-speech
pretty_name: Mimir Auto Tokenized SNAC
---
# Mimir Auto Tokenized SNAC
Tokenized TTS training dataset derived from the paired audio/text dataset.
Rows have exactly three columns:
- `inputs`: single token stream containing byte text, special markers, and flattened SNAC tokens
- `labels`: same length as `inputs`, with `-100` for text/prompt positions
- `attentionmask`: sequence mask, `1` for valid positions