update

2025-10-20 15:25:22 +08:00 · 2025-01-09 07:16:13 +00:00
parent ed5f65a694
commit 99d0921049
1 changed files with 2 additions and 2 deletions
--- a/llm/server/server/engine/infer.py
+++ b/llm/server/server/engine/infer.py
@@ -500,8 +500,8 @@ class ModelRunner:
            if self.proposer is not None:
                self.proposer.run(
                    self.share_inputs,
-                    real_batch_size=self.args.max_batch_size,
-                    seq_lens_this_time=self.share_inputs["seq_lens_this_time"],
+                    real_batch_size=seq_lens_this_time.shape[0],
+                    seq_lens_this_time=seq_lens_this_time,
                )

            self.infer_engine.predictor.run()