| import sys |
| import glob |
| from tensorboard.backend.event_processing.event_accumulator import EventAccumulator |
|
|
| |
| versions = sorted(glob.glob("/workspace/boccaccioAI/logs/finetune/version_*")) |
| if not versions: |
| print("No versions found") |
| sys.exit(1) |
| path = versions[-1] |
| print(f"Reading: {path}") |
|
|
| ea = EventAccumulator(path) |
| ea.Reload() |
| tags = ea.Tags() |
| print(f"Tags: {tags['scalars']}") |
|
|
| if 'train/loss' in tags['scalars']: |
| events = ea.Scalars('train/loss') |
| print(f"\ntrain/loss: {len(events)} events (first 10)") |
| for e in events[:10]: |
| print(f" step={e.step}, value={e.value:.4f}") |
| print("...") |
| for e in events[-5:]: |
| print(f" step={e.step}, value={e.value:.4f}") |
|
|
| if 'val/loss' in tags['scalars']: |
| events = ea.Scalars('val/loss') |
| print(f"\nval/loss: {len(events)} events (last 5)") |
| for e in events[-5:]: |
| print(f" step={e.step}, value={e.value:.4f}") |
|
|
| if 'perf/tokens_per_sec' in tags['scalars']: |
| events = ea.Scalars('perf/tokens_per_sec') |
| print(f"\nperf/tokens_per_sec: {len(events)} events (last 3)") |
| for e in events[-3:]: |
| print(f" step={e.step}, value={e.value:.0f}") |
|
|
