feat: tts-norm LoRA — dataset generator + training script

Browse Source

gen_tts_dataset.py: 4960 synthetic examples, 22 categories (numbers,
currencies, dates, times, temperatures, acronyms, NATO phonetic, URLs,
markdown, etc). Bilingual EN/PL with explicit [lang] tag prefix.

train_tts_norm.py: Unsloth LoRA training for Qwen2.5-7B-Instruct.
Rank 16, 3 epochs, packing, max_seq 768. Trained on H100 in 20m38s,
final loss 0.091. Adapter: 154MB.

...

This commit is contained in:

marauder-actual

2026-05-26 00:14:51 +02:00

parent 8137d278db

commit 122e73860b

2 changed files with 1525 additions and 0 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

gen_tts_dataset.py

+1352

File diff suppressed because it is too large Load Diff