Revert "[EAGLE] Refactor code for page size > 1 & more simplifications" (#7210)
This commit is contained in:
@@ -441,71 +441,5 @@ class TestEAGLEServerTriton(TestEAGLEServer):
|
||||
)
|
||||
|
||||
|
||||
class TestEAGLEServerPageSize(TestEAGLEServer):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--speculative-algorithm",
|
||||
"EAGLE",
|
||||
"--speculative-draft-model-path",
|
||||
DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
|
||||
"--speculative-num-steps",
|
||||
5,
|
||||
"--speculative-eagle-topk",
|
||||
1,
|
||||
"--speculative-num-draft-tokens",
|
||||
6,
|
||||
"--mem-fraction-static",
|
||||
0.7,
|
||||
"--chunked-prefill-size",
|
||||
128,
|
||||
"--max-running-requests",
|
||||
8,
|
||||
"--page-size",
|
||||
4,
|
||||
"--attention-backend",
|
||||
"flashinfer",
|
||||
],
|
||||
)
|
||||
|
||||
|
||||
class TestEAGLEServerPageSizeTopk(TestEAGLEServer):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--speculative-algorithm",
|
||||
"EAGLE",
|
||||
"--speculative-draft-model-path",
|
||||
DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
|
||||
"--speculative-num-steps",
|
||||
5,
|
||||
"--speculative-eagle-topk",
|
||||
8,
|
||||
"--speculative-num-draft-tokens",
|
||||
64,
|
||||
"--mem-fraction-static",
|
||||
0.7,
|
||||
"--chunked-prefill-size",
|
||||
128,
|
||||
"--max-running-requests",
|
||||
8,
|
||||
"--page-size",
|
||||
4,
|
||||
"--attention-backend",
|
||||
"flashinfer",
|
||||
],
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
|
||||
Reference in New Issue
Block a user