H1K0 d09a08d553 feat: add src/evaluate.py and scripts/evaluate.py ...

Implements perplexity computation, chord distribution extraction (qualities,
extensions, inversions, root-motion intervals), 4-panel comparison plot, and
paired qualitative example generation for pretrained vs finetuned model.

Results on user val set: pretrained PPL 3.58 → finetuned PPL 2.15 (−40 %).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-06-04 14:57:49 +03:00

..

evaluate.py

feat: add src/evaluate.py and scripts/evaluate.py

2026-06-04 14:57:49 +03:00

generate.py

feat: add --bars arg to control output length

2026-05-21 20:29:44 +03:00

make_colab_zip.py

refactor: reorganize data/processed/ into mcgill/ and user/ subdirs

2026-05-21 19:47:32 +03:00

prepare_data.py

feat: add dataset, prepare_data pipeline and fix McGill converter

2026-05-19 18:09:46 +03:00

pretrain.py

feat: remove BAR token; bump spec to v2.3; fix max_seq_len

2026-05-20 13:56:34 +03:00

train.py

scripts: update fine-tune defaults to lr=3e-5, epochs=30

2026-05-21 21:18:37 +03:00