hamori/notebooks/colab_pretrain.ipynb

hamori — pre-training on McGill Billboard corpus¶

This notebook runs scripts/pretrain.py on Google Colab (GPU T4 recommended).

Steps:

1. Check GPU
2. Upload hamori_colab.zip (built locally via python scripts/make_colab_zip.py)
3. Extract and install dependencies
4. Run pre-training
5. Download checkpoint and logs

In [ ]:

# ── 1. GPU check ────────────────────────────────────────────────────────────
import torch
if torch.cuda.is_available():
    print(f"GPU: {torch.cuda.get_device_name(0)}")
    print(f"VRAM: {torch.cuda.get_device_properties(0).total_memory / 1e9:.1f} GB")
else:
    print("No GPU found — training will be slow on CPU.")
    print("Go to Runtime → Change runtime type → T4 GPU and re-run.")

In [ ]:

# ── 2. Upload hamori_colab.zip ───────────────────────────────────────────────
# Build it locally first:  python scripts/make_colab_zip.py
from google.colab import files
uploaded = files.upload()   # select hamori_colab.zip from your machine
print("Uploaded:", list(uploaded.keys()))

In [ ]:

# ── 3. Extract and install dependencies ─────────────────────────────────────
import zipfile, os

WORK_DIR = "/content/hamori"
os.makedirs(WORK_DIR, exist_ok=True)

with zipfile.ZipFile("hamori_colab.zip") as zf:
    zf.extractall(WORK_DIR)
    print(f"Extracted {len(zf.namelist())} files to {WORK_DIR}")

os.chdir(WORK_DIR)
print("Working directory:", os.getcwd())

In [ ]:

# Colab ships torch; only install the extra deps
!pip install -q pretty_midi mido matplotlib

In [ ]:

# ── 4. Verify data ───────────────────────────────────────────────────────────
from pathlib import Path
train_pt = list(Path("data/processed/mcgill/train").glob("*.pt"))
val_pt   = list(Path("data/processed/mcgill/val").glob("*.pt"))
print(f"Train: {len(train_pt)} files")
print(f"Val:   {len(val_pt)} files")
if not train_pt:
    print("ERROR: no training data found — did you build the zip with data included?")

In [ ]:

# ── 5. Pre-train ─────────────────────────────────────────────────────────────
# Outputs:
#   checkpoints/pretrained.pt
#   checkpoints/pretrained.log.csv
#   checkpoints/pretrained_curves.png
#   checkpoints/pretrained.report.txt
!python scripts/pretrain.py

In [ ]:

# ── 6. Show report ───────────────────────────────────────────────────────────
report = Path("checkpoints/pretrained.report.txt")
if report.exists():
    print(report.read_text(encoding="utf-8"))
else:
    print("Report not found — training may have failed.")

In [ ]:

# Show loss curves inline
from IPython.display import Image
Image("checkpoints/pretrained_curves.png")

In [ ]:

# ── 7. Download results ──────────────────────────────────────────────────────
import shutil
from google.colab import files

# Bundle all checkpoint outputs into a single zip for download
shutil.make_archive("/content/pretrain_results", "zip", WORK_DIR, "checkpoints")
files.download("/content/pretrain_results.zip")