summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJonas Knobloch <jonas.knobloch@t-online.de>2026-03-31 17:05:43 +0200
committerJonas Knobloch <jonas.knobloch@t-online.de>2026-03-31 17:31:47 +0200
commitfa4e25b4992c25cdec41bdccb8af909b93fc9c39 (patch)
treef69359d436ad65d08e6f92a2666629b798af60ba
parentce9c65cb1e13f7ae6eed4b419232a4da9dd63fdb (diff)
Count skipped texts
-rw-r--r--llm/run.go16
1 files changed, 14 insertions, 2 deletions
diff --git a/llm/run.go b/llm/run.go
index df48ff6..227a365 100644
--- a/llm/run.go
+++ b/llm/run.go
@@ -52,13 +52,25 @@ func (e *Evaluator[R]) Run(title string, data dataset.Reader, window, stride int
return int(e.completed.Load())
})
- defer pb.Close()
+ s := 0 // skipped
+ n := 0 // total
- n := 0
+ defer func() {
+ fmt.Printf("skipped %d of %d texts\n", s, n)
+ }()
+
+ defer pb.Close()
for d := range data.Texts("text") {
tokens := toInt64(e.tokenizer.Tokenize(d))
+ if len(tokens) == 0 {
+ n++
+ s++
+
+ continue
+ }
+
e.schedule(n, tokens, window, stride, e.batchSize)
pb.SetTotal(pb.Total() + e.estimateJobs(tokens, window, stride))