summaryrefslogtreecommitdiff
path: root/research/sander
diff options
context:
space:
mode:
authorJonas Knobloch <jonas.knobloch@t-online.de>2026-04-24 11:49:26 +0200
committerJonas Knobloch <jonas.knobloch@t-online.de>2026-04-24 11:50:02 +0200
commit12f15fdd8c048e5958e294f12a9f163f6f950b3e (patch)
tree16bb23c134f1fe476d977dd846e6dda6c7667c48 /research/sander
parent408536516bdfedc93d93226dc08280ed224361fc (diff)
Recover tokenization based on config
Diffstat (limited to 'research/sander')
-rw-r--r--research/sander/cmd/sander/main.go6
1 files changed, 5 insertions, 1 deletions
diff --git a/research/sander/cmd/sander/main.go b/research/sander/cmd/sander/main.go
index 6f68434..56f1488 100644
--- a/research/sander/cmd/sander/main.go
+++ b/research/sander/cmd/sander/main.go
@@ -28,7 +28,11 @@ func main() {
}
func run(src, dst string) error {
- t := must(bpe.NewTokenizerFromFiles(path.Join(src, "vocab.json"), path.Join(src, "merges.txt")))
+ cfg := bpe.Config{
+ Recover: true,
+ }
+
+ t := must(bpe.NewTokenizerFromFiles(path.Join(src, "vocab.json"), path.Join(src, "merges.txt"), cfg))
e := must(sander.NewExperiment(dst, path.Join(src, "model.onnx"), sander.UnusedTokensMBPE(t)))
return e.Run()