summaryrefslogtreecommitdiff
path: root/tokenizer/bpe/validate.go
diff options
context:
space:
mode:
authorJonas Knobloch <jonas.knobloch@t-online.de>2026-04-11 01:09:16 +0200
committerJonas Knobloch <jonas.knobloch@t-online.de>2026-04-14 19:38:34 +0200
commit0c97f22069d77cf3cac92d68173e38f83b258a9a (patch)
tree127bf00ad1f69ce1368089247471fd52b8f9d305 /tokenizer/bpe/validate.go
parentaa1fdb5f3b5d6c54c3bc44269482c31f4dfafe67 (diff)
Add lesci module
Diffstat (limited to 'tokenizer/bpe/validate.go')
-rw-r--r--tokenizer/bpe/validate.go14
1 files changed, 14 insertions, 0 deletions
diff --git a/tokenizer/bpe/validate.go b/tokenizer/bpe/validate.go
index f9bd265..84910cf 100644
--- a/tokenizer/bpe/validate.go
+++ b/tokenizer/bpe/validate.go
@@ -106,3 +106,17 @@ func ByteCoverage(t *Tokenizer) bool {
return covered
}
+
+func ReachableMerges(t *Tokenizer, merges [][2]string) []bool {
+ mask := make([]bool, len(merges))
+
+ for i, merge := range merges {
+ a := t.Tokenize(merge[0])
+ b := t.Tokenize(merge[1])
+ c := t.Tokenize(merge[0] + merge[1])
+
+ mask[i] = len(a) == 1 && len(b) == 1 && len(c) == 1
+ }
+
+ return mask
+}