Recapture cuda graph after model weight update to resolve IMA error (#11780)

2025-10-19 19:50:03 -07:00
parent 283c8ba031
commit c726d44cc7
1 changed files with 4 additions and 0 deletions
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -981,6 +981,10 @@ class ModelRunner:
        self.server_args.load_format = load_format
        self.load_config = load_config

+        # Recapture device graph after model weight update.
+        if not self.server_args.disable_cuda_graph and self.device == "cuda":
+            self.init_device_graphs()
+
        logger.info("Update weights end.")
        return True, "Succeeded to update model weights."