[PD]: Support incremental transfer for mooncake transfer engine (#24257)
Co-authored-by: Shangming Cai <csmthu@gmail.com>
This commit is contained in:
co-authored by
Shangming Cai
parent
62265ca7fc
commit
9a5450ad73
@@ -63,6 +63,28 @@ class TestLoadBalanceMethod(unittest.TestCase):
|
||||
str(context.exception),
|
||||
)
|
||||
|
||||
def test_pd_decode_radix_cache_allows_mooncake(self):
|
||||
server_args = ServerArgs(
|
||||
model_path="dummy",
|
||||
disaggregation_mode="decode",
|
||||
disaggregation_decode_enable_radix_cache=True,
|
||||
disaggregation_transfer_backend="mooncake",
|
||||
)
|
||||
|
||||
self.assertFalse(server_args.disable_radix_cache)
|
||||
|
||||
def test_pd_decode_radix_cache_rejects_unknown_backend(self):
|
||||
with self.assertRaises(ValueError) as context:
|
||||
ServerArgs(
|
||||
model_path="dummy",
|
||||
disaggregation_mode="decode",
|
||||
disaggregation_decode_enable_radix_cache=True,
|
||||
disaggregation_transfer_backend="fake",
|
||||
)
|
||||
|
||||
self.assertIn("('nixl', 'mooncake')", str(context.exception))
|
||||
self.assertIn("'fake'", str(context.exception))
|
||||
|
||||
|
||||
class TestPortArgs(unittest.TestCase):
|
||||
@patch("sglang.srt.server_args.get_free_port")
|
||||
|
||||
Reference in New Issue
Block a user