pscamillo commited on
Commit
fdc9818
·
verified ·
1 Parent(s): 6fb716d

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +8 -10
README.md CHANGED
@@ -1,6 +1,5 @@
1
  ---
2
- license: other
3
- license_name: vesuvius-challenge-derived
4
  tags:
5
  - vesuvius-challenge
6
  - ink-detection
@@ -53,12 +52,11 @@ uniform grey even when they carry signal.
53
 
54
  ## Two things to know before comparing against anything else
55
 
56
- **Step 77,500, not 78,125.** `num_iterations: 78125` with `save_every: 2500`
57
- writes no checkpoint at the final step, because the trainer only saves when
58
- `(step+1) % save_every == 0` and the closing call does not force it. All five
59
- arms are at the same step, so the comparisons among them are clean. The same
60
- applies to the official release with `save_every: 5000` — its published
61
- `step-075000` is the last one, not the final one.
62
 
63
  **Labels.** All five arms trained on the `ink_9um` labels as they stood before
64
  the 2026-08-18 correction, when the seam regions from the alignment pipeline
@@ -72,8 +70,8 @@ GitHub repo: median change 0.0%, mean +2.5% pulled by a single outlier.
72
 
73
  One seed per arm. One judge for all 300 blind comparisons. Region-held-out
74
  rather than segment-held-out validation, so nothing here speaks to cross-scroll
75
- generalisation. The full list is in the repository README, and it is worth
76
- reading before drawing conclusions from these weights.
77
 
78
  ---
79
 
 
1
  ---
2
+ license: mit
 
3
  tags:
4
  - vesuvius-challenge
5
  - ink-detection
 
52
 
53
  ## Two things to know before comparing against anything else
54
 
55
+ **Step 77,500, not 78,125.** With `num_iterations: 78125` and
56
+ `save_every: 2500`, the trainer writes at multiples of 2,500, so the last file
57
+ is 77,500 — 99.2% of training. All five arms are at the same step, so the
58
+ comparisons among them are clean. Worth knowing when comparing against the
59
+ released checkpoints, which follow the same pattern at `save_every: 5000`.
 
60
 
61
  **Labels.** All five arms trained on the `ink_9um` labels as they stood before
62
  the 2026-08-18 correction, when the seam regions from the alignment pipeline
 
70
 
71
  One seed per arm. One judge for all 300 blind comparisons. Region-held-out
72
  rather than segment-held-out validation, so nothing here speaks to cross-scroll
73
+ generalisation. The full list is in the repository README, worth reading
74
+ alongside these weights.
75
 
76
  ---
77