[CI] Temporarily disable prefill cuda graph for qwen3.5 nightly test (#33772)
This commit is contained in:
@@ -30,7 +30,7 @@ class TestQwen35(unittest.TestCase):
|
||||
"--tool-call-parser=qwen3_coder",
|
||||
"--mem-fraction-static=0.8",
|
||||
]
|
||||
dp_args = ["--dp=8", "--enable-dp-attention"]
|
||||
dp_args = ["--dp=8", "--enable-dp-attention", "--disable-prefill-cuda-graph"]
|
||||
mtp_args = [
|
||||
"--speculative-algorithm=EAGLE",
|
||||
"--speculative-num-steps=3",
|
||||
|
||||
Reference in New Issue
Block a user