minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

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

Browse Source
This commit is contained in:
Liangsheng Yin
2026-06-28 18:38:28 -07:00
committed by GitHub
parent 3217410cf6
commit 909123ddb8
140 changed files with 304 additions and 290 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/manual/mla/test_flashmla_archived.py
+1 -1
View File
@@ -35,7 +35,7 @@ class TestFlashMLAAttnBackend(unittest.TestCase):
if torch.cuda.is_available() and torch.version.cuda:
other_args.extend(
[
"--cuda-graph-max-bs",
"--cuda-graph-max-bs-decode",
"2",
"--attention-backend",
"flashmla",
test/manual/mla/test_mla_deepseek_v3.py
+6 -4
View File
@@ -22,7 +22,9 @@ class TestMLADeepseekV3(CustomTestCase):
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = ["--trust-remote-code", "--chunked-prefill-size", "256"]
if is_cuda():
other_args.extend(["--enable-torch-compile", "--cuda-graph-max-bs", "2"])
other_args.extend(
["--enable-torch-compile", "--cuda-graph-max-bs-decode", "2"]
)
cls.process = popen_launch_server(
cls.model,
cls.base_url,
@@ -59,7 +61,7 @@ class TestMLADeepseekV3DisableFusedFunc(CustomTestCase):
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = ["--trust-remote-code", "--chunked-prefill-size", "256"]
if is_cuda():
other_args.extend(["--cuda-graph-max-bs", "2"])
other_args.extend(["--cuda-graph-max-bs-decode", "2"])
cls.process = popen_launch_server(
cls.model,
cls.base_url,
@@ -107,7 +109,7 @@ class TestMLADeepseekV3Fa3Fp8Kvcache(CustomTestCase):
"fa3",
"--mem-fraction-static",
"0.8",
"--cuda-graph-max-bs",
"--cuda-graph-max-bs-decode",
"2",
]
)
@@ -145,7 +147,7 @@ class TestDeepseekV3MTP(CustomTestCase):
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--trust-remote-code",
"--cuda-graph-max-bs",
"--cuda-graph-max-bs-decode",
"2",
"--disable-radix",
"--enable-torch-compile",
test/manual/mla/test_mla_flashinfer_archived.py
+1 -1
View File
@@ -31,7 +31,7 @@ class TestFlashinferMLA(CustomTestCase):
other_args.extend(
[
"--enable-torch-compile",
"--cuda-graph-max-bs",
"--cuda-graph-max-bs-decode",
"4",
"--attention-backend",
"flashinfer",
test/manual/mla/test_mla_int8_deepseek_v3_archived.py
+2 -2
View File
@@ -31,7 +31,7 @@ class TestMLADeepseekV3ChannelInt8(CustomTestCase):
if torch.cuda.is_available() and torch.version.cuda:
other_args.extend(
[
"--cuda-graph-max-bs",
"--cuda-graph-max-bs-decode",
"16",
"--enable-torch-compile",
"--torch-compile-max-bs",
@@ -75,7 +75,7 @@ class TestMLADeepseekV3BlockInt8(CustomTestCase):
if torch.cuda.is_available() and torch.version.cuda:
other_args.extend(
[
"--cuda-graph-max-bs",
"--cuda-graph-max-bs-decode",
"16",
"--enable-torch-compile",
"--torch-compile-max-bs",