From 12f15fdd8c048e5958e294f12a9f163f6f950b3e Mon Sep 17 00:00:00 2001 From: Jonas Knobloch Date: Fri, 24 Apr 2026 11:49:26 +0200 Subject: Recover tokenization based on config --- research/sander/cmd/sander/main.go | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'research/sander/cmd') 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() -- cgit v1.3.1