Migrate all callers from /get_server_info to /server_info (#21463)
This commit is contained in:
@@ -87,7 +87,7 @@ class TestEAGLE3EngineDPAttention(CustomTestCase):
|
||||
metrics = run_eval(args)
|
||||
print(f"{metrics=}")
|
||||
|
||||
server_info = requests.get(self.base_url + "/get_server_info")
|
||||
server_info = requests.get(self.base_url + "/server_info")
|
||||
server_data = server_info.json()
|
||||
|
||||
# Try to get avg_spec_accept_length
|
||||
|
||||
@@ -33,7 +33,7 @@ def test_gsm8k(base_url: str, model: str):
|
||||
num_threads=128,
|
||||
)
|
||||
metrics = run_eval(args)
|
||||
server_info = requests.get(base_url + "/get_server_info")
|
||||
server_info = requests.get(base_url + "/server_info")
|
||||
avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
"avg_spec_accept_length"
|
||||
]
|
||||
|
||||
@@ -113,7 +113,7 @@ class TestStandaloneSpeculativeDecodingBase(CustomTestCase):
|
||||
metric_key = "score"
|
||||
self.assertGreater(metrics[metric_key], self.accuracy_threshold)
|
||||
|
||||
server_info = requests.get(self.base_url + "/get_server_info")
|
||||
server_info = requests.get(self.base_url + "/server_info")
|
||||
avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
"avg_spec_accept_length"
|
||||
]
|
||||
@@ -176,7 +176,7 @@ class TestStandaloneV2SpeculativeDecodingBase(CustomTestCase):
|
||||
metric_key = "score"
|
||||
self.assertGreater(metrics[metric_key], self.accuracy_threshold)
|
||||
|
||||
server_info = requests.get(self.base_url + "/get_server_info")
|
||||
server_info = requests.get(self.base_url + "/server_info")
|
||||
avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
"avg_spec_accept_length"
|
||||
]
|
||||
|
||||
Reference in New Issue
Block a user