[misc] Use --cuda-graph-max-bs-decode in tests, examples, and docs (#29591)

test/registered/input_embedding/test_input_embeddings.py

@@ -31,7 +31,7 @@ class TestInputEmbeds(CustomTestCase):
             cls.model,
             cls.base_url,
             timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=["--disable-radix", "--cuda-graph-max-bs", 4],
+            other_args=["--disable-radix", "--cuda-graph-max-bs-decode", 4],
         )
         cls.texts = [
             "The capital of France is",

test/registered/input_embedding/test_input_embeds_chunked.py

@@ -94,7 +94,7 @@ class TestInputEmbedsChunkedAndRetract(CustomTestCase):
                     "--disable-radix-cache",
                     "--chunked-prefill-size",
                     str(CHUNKED_PREFILL_SIZE),
-                    "--cuda-graph-max-bs",
+                    "--cuda-graph-max-bs-decode",
                     "4",
                 ],
             )