[CI] Add comment about the single cudagraph capture size that is used (#26252)

tdoublep · web-flow · commit d3c84297c38c · 2025-10-06T02:35:37.000Z
diff --git a/tests/conftest.py b/tests/conftest.py
@@ -740,6 +740,10 @@ def __init__(
         )
 
         if not kwargs.get("compilation_config", None):
+            # Note(@tdoublep): This is set to 4 because some tests (e.g., hybrid
+            # model tests) may set max_num_seqs=4. If min cudagraph_capture_size is
+            # set to larger than max_num_seqs, then it will lead to *no* graphs
+            # being captured which can trigger edge cases that we don't handle yet.
             kwargs["compilation_config"] = {"cudagraph_capture_sizes": [4]}
 
         with init_ctx:

Original file line number	Diff line number	Diff line change
`@@ -740,6 +740,10 @@ def __init__(`
`740`	`740`	`)`
`741`	`741`
`742`	`742`	`if not kwargs.get("compilation_config", None):`
	`743`	`+ # Note(@tdoublep): This is set to 4 because some tests (e.g., hybrid`
	`744`	`+ # model tests) may set max_num_seqs=4. If min cudagraph_capture_size is`
	`745`	`+ # set to larger than max_num_seqs, then it will lead to no graphs`
	`746`	`+ # being captured which can trigger edge cases that we don't handle yet.`
`743`	`747`	`kwargs["compilation_config"] = {"cudagraph_capture_sizes": [4]}`
`744`	`748`
`745`	`749`	`with init_ctx:`