Fix DSpark + DeepSeek V4 prefill CP compatibility (#33865)
This commit is contained in:
@@ -198,5 +198,49 @@ class TestDSV4FlashFP4B200Balanced_CP_NonDeepEP(
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
|
||||
# DSPARK draft is bundled with the -DSpark checkpoint.
|
||||
DSPARK_MODEL = "deepseek-ai/DeepSeek-V4-Flash-DSpark"
|
||||
|
||||
|
||||
class TestDSV4FlashFP4B200_CP_DSpark(
|
||||
BasicDecodeCorrectnessMixin,
|
||||
GSM8KMixin,
|
||||
CustomTestCase,
|
||||
):
|
||||
"""DSPARK speculation + prefill CP (interleave, CP_V2, attn_cp=tp)."""
|
||||
|
||||
gsm8k_accuracy_thres = 0.90
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = try_cached_model(DSPARK_MODEL)
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=SERVER_LAUNCH_TIMEOUT,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--attn-cp-size",
|
||||
"4",
|
||||
"--speculative-algorithm",
|
||||
"DSPARK",
|
||||
"--enable-prefill-cp",
|
||||
"--cp-strategy",
|
||||
"interleave",
|
||||
"--moe-runner-backend", # for fp4 checkpoint
|
||||
"flashinfer_mxfp4",
|
||||
],
|
||||
env={"SGLANG_ENABLE_CP_V2": "1"},
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
if hasattr(cls, "process") and cls.process:
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
|
||||
Reference in New Issue
Block a user