[Optimization][Perf] Disable the GC during CUDA graph capture to speed up by up to 3x

python/sglang/srt/model_executor/cuda_graph_runner.py

@@ -16,6 +16,7 @@
 from __future__ import annotations
 
 import bisect
+import gc
 import inspect
 import logging
 import os
@@ -75,6 +76,24 @@ def model_capture_mode():
     is_capture_mode = False
 
 
+@contextmanager
+def freeze_gc(enable_cudagraph_gc: bool):
+    """
+    Optimize garbage collection during CUDA graph capture.
+    Clean up, then freeze all remaining objects from being included
+    in future collections if GC is disabled during capture.
+    """
+    gc.collect()
+    should_freeze = not enable_cudagraph_gc
+    if should_freeze:
+        gc.freeze()
+    try:
+        yield
+    finally:
+        if should_freeze:
+            gc.unfreeze()
+
+
 def _to_torch(model: torch.nn.Module, reverse: bool, num_tokens: int):
     for sub in model._modules.values():
         if isinstance(sub, CustomOp):
@@ -423,7 +442,12 @@ def capture(self) -> None:
                 record_shapes=True,
             )
 
-        with graph_capture() as graph_capture_context:
+        # Trigger CUDA graph capture for specific shapes.
+        # Capture the large shapes first so that the smaller shapes
+        # can reuse the memory pool allocated for the large shapes.
+        with freeze_gc(
+            self.model_runner.server_args.enable_cudagraph_gc
+        ), graph_capture() as graph_capture_context:
             with profile_context as prof:
                 self.stream = graph_capture_context.stream
                 avail_mem = get_available_gpu_memory(

python/sglang/srt/server_args.py

@@ -215,6 +215,7 @@ class ServerArgs:
     disable_cuda_graph: bool = False
     disable_cuda_graph_padding: bool = False
     enable_profile_cuda_graph: bool = False
+    enable_cudagraph_gc: bool = False
     enable_nccl_nvls: bool = False
     enable_tokenizer_batch_encode: bool = False
     disable_outlines_disk_cache: bool = False
@@ -1551,6 +1552,11 @@ def add_cli_args(parser: argparse.ArgumentParser):
             action="store_true",
             help="Enable profiling of cuda graph capture.",
         )
+        parser.add_argument(
+            "--enable-cudagraph-gc",
+            action="store_true",
+            help="Enable garbage collection during CUDA graph capture. If disabled (default), GC is frozen during capture to speed up the process.",
+        )
         parser.add_argument(
             "--enable-nccl-nvls",
             action="store_true",