fix execute_model_normal in empty run (#6968)

2026-04-23 08:21:53 +08:00 · 2026-03-23 14:07:46 +08:00
parent 33e79f922a
commit 7a78001be2
1 changed files with 1 additions and 1 deletions
@@ -2039,7 +2039,7 @@ class GPUModelRunner(ModelRunnerBase):
    ) -> None:
        model_inputs, p_done_idxs, _ = self._preprocess(model_forward_batch, num_running_requests)
        model_output = self._execute(model_inputs)
-        if model_output is None:
+        if model_output is None or self.share_inputs["seq_lens_this_time_cpu"].numpy().sum().item() <= 0:
            return
        model_output_data, sampler_output, post_process_event = self._postprocess(
            model_output, p_done_idxs, model_forward_batch, num_running_requests