Mulah commited on
Commit
eb92d34
·
verified ·
1 Parent(s): 33fccd7

README: define Truncated as true 128-token cap hit

Browse files
Files changed (1) hide show
  1. data/Supplementary_Data/README.md +1 -1
data/Supplementary_Data/README.md CHANGED
@@ -41,7 +41,7 @@ Methods, "Expert Curation".
41
  | `Curator_score` | Expert rating, 1 (unrelated or contradictory) to 5 (equivalent) |
42
  | `Specificity` | Prediction more / equally / less specific than the Swiss-Prot description |
43
  | `Information` | Prediction more / equally / less informative than the Swiss-Prot description |
44
- | `Truncated` | Whether the prediction does not end in terminal punctuation, used as a proxy for generation having reached the 128-token cap. On the unseen set, 93% of predictions flagged this way did reach the cap |
45
  | `SwissProt_description` | Reference text shown to the curator |
46
  | `ProtTale_prediction` | Generated text shown to the curator |
47
  | `Curator_notes` | Free-text justification, present for 175 of the 232 items |
 
41
  | `Curator_score` | Expert rating, 1 (unrelated or contradictory) to 5 (equivalent) |
42
  | `Specificity` | Prediction more / equally / less specific than the Swiss-Prot description |
43
  | `Information` | Prediction more / equally / less informative than the Swiss-Prot description |
44
+ | `Truncated` | Whether generation reached the 128-token cap without emitting an end-of-sequence token (82 of the 232 curated predictions; 233 of the 1,031 unseen predictions, 22.6%) |
45
  | `SwissProt_description` | Reference text shown to the curator |
46
  | `ProtTale_prediction` | Generated text shown to the curator |
47
  | `Curator_notes` | Free-text justification, present for 175 of the 232 items |