train: train_ounce100m -- download_checkpoint prefix/landing-dir fix (found by P1 v2)
Browse files- train/train_ounce100m.py +2 -1
train/train_ounce100m.py
CHANGED
|
@@ -204,7 +204,8 @@ def main():
|
|
| 204 |
say(f"downloaded {got['files']} files, {got['bytes']/1e6:.0f} MB in {got['seconds']}s")
|
| 205 |
if got["bytes"] < 1_000_000:
|
| 206 |
raise SystemExit("auto-resume downloaded a checkpoint that is not real weights")
|
| 207 |
-
args.resume =
|
|
|
|
| 208 |
cur = SD.load_cursor(args.resume) if args.resume else None
|
| 209 |
if args.resume:
|
| 210 |
say(f"resume from {args.resume} cursor={cur}")
|
|
|
|
| 204 |
say(f"downloaded {got['files']} files, {got['bytes']/1e6:.0f} MB in {got['seconds']}s")
|
| 205 |
if got["bytes"] < 1_000_000:
|
| 206 |
raise SystemExit("auto-resume downloaded a checkpoint that is not real weights")
|
| 207 |
+
args.resume = got["dir"]
|
| 208 |
+
say(f"resuming from {args.resume}")
|
| 209 |
cur = SD.load_cursor(args.resume) if args.resume else None
|
| 210 |
if args.resume:
|
| 211 |
say(f"resume from {args.resume} cursor={cur}")
|