add comment

Hgherzog · Hgherzog · commit 4b4647d3be45 · 2025-10-23T14:42:30.000-07:00
diff --git a/olmoearth_pretrain/train/masking.py b/olmoearth_pretrain/train/masking.py
@@ -1172,7 +1172,7 @@ def apply_bandset_mask_rules(
             num_encoded = self._accumulate_token_counts(num_encoded, enc_count)
             num_decoded = self._accumulate_token_counts(num_decoded, dec_count)
 
-        # Handle edge cases where samples have no encoded or decoded tokens (only happens for some ablations)
+        # Handle edge cases where samples have no encoded or decoded tokens (only happens for some ablations like S2  only with small HW)
         self._handle_no_tokens_edge_cases(
             masked_batch_dict, num_encoded, num_decoded, patch_size
         )

Original file line number	Diff line number	Diff line change
`@@ -1172,7 +1172,7 @@ def apply_bandset_mask_rules(`
`1172`	`1172`	`num_encoded = self._accumulate_token_counts(num_encoded, enc_count)`
`1173`	`1173`	`num_decoded = self._accumulate_token_counts(num_decoded, dec_count)`
`1174`	`1174`
`1175`		`- # Handle edge cases where samples have no encoded or decoded tokens (only happens for some ablations)`
	`1175`	`+ # Handle edge cases where samples have no encoded or decoded tokens (only happens for some ablations like S2 only with small HW)`
`1176`	`1176`	`self._handle_no_tokens_edge_cases(`
`1177`	`1177`	`masked_batch_dict, num_encoded, num_decoded, patch_size`
`1178`	`1178`	`)`