[CI] Add Kimi-K3 low-latency performance check (#34089)
This commit is contained in:
@@ -1,7 +1,8 @@
|
||||
"""B300 per-commit CI coverage for Kimi-K3 serving recipes.
|
||||
|
||||
Runs the Low Latency DSPARK recipe and the Balanced DCP/HiCache recipe on
|
||||
eight B300 GPUs. Each server must preserve basic model quality on GSM8K.
|
||||
eight B300 GPUs. Each server must preserve basic model quality on GSM8K, and
|
||||
the Low Latency recipe must also preserve single-request decode performance.
|
||||
"""
|
||||
|
||||
import unittest
|
||||
@@ -9,6 +10,7 @@ import unittest
|
||||
from sglang.srt.utils import kill_process_tree
|
||||
from sglang.test.ci.ci_register import register_cuda_ci
|
||||
from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
|
||||
from sglang.test.kits.spec_decoding_kit import SpecDecodingMixin
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
@@ -33,11 +35,13 @@ def _stop_server(process):
|
||||
_wait_for_gpu_idle_in_ci(timeout=GPU_IDLE_TIMEOUT)
|
||||
|
||||
|
||||
class TestKimiK3B300LowLatency(GSM8KMixin, CustomTestCase):
|
||||
class TestKimiK3B300LowLatency(GSM8KMixin, SpecDecodingMixin, CustomTestCase):
|
||||
"""TP8 Low Latency recipe with DSPARK linear ReplaySSM speculation."""
|
||||
|
||||
gsm8k_score_threshold = 0.95
|
||||
gsm8k_num_examples = 200
|
||||
accept_length_thres = 6.6
|
||||
bs_1_speed_thres = 440
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
|
||||
Reference in New Issue
Block a user