[CI] Add deepep tests to CI (#7872)
This commit is contained in:
Cheng Wan
@@ -170,21 +170,22 @@ suites = {
|
||||
TestFile("test_pp_single_node.py", 150),
|
||||
TestFile("test_multi_instance_release_memory_occupation.py", 64),
|
||||
],
|
||||
"per-commit-4-gpu-deepep": [
|
||||
TestFile("test_deepep_small.py", 531),
|
||||
],
|
||||
"per-commit-4-gpu-amd": [
|
||||
TestFile("test_pp_single_node.py", 150),
|
||||
],
|
||||
"per-commit-8-gpu": [
|
||||
# Disabled deepep tests temporarily because it takes too much time.
|
||||
# TODO: re-enable them after reducing the test time with compilation cache and smaller models.
|
||||
# TestFile("test_deepep_intranode.py", 50),
|
||||
# TestFile("test_deepep_low_latency.py", 50),
|
||||
# TestFile("test_moe_deepep_eval_accuracy_large.py", 250),
|
||||
# Disabled because it hangs on the CI.
|
||||
# TestFile("test_moe_ep.py", 181),
|
||||
TestFile("test_disaggregation.py", 270),
|
||||
TestFile("test_disaggregation_different_tp.py", 155),
|
||||
TestFile("test_full_deepseek_v3.py", 463),
|
||||
],
|
||||
"per-commit-8-gpu-deepep": [
|
||||
TestFile("test_deepep_large.py", 485),
|
||||
],
|
||||
"per-commit-8-gpu-amd": [
|
||||
TestFile("test_full_deepseek_v3.py", 250),
|
||||
],
|
||||
|
||||
145
test/srt/test_deepep_large.py
Normal file
145
test/srt/test_deepep_large.py
Normal file
@@ -0,0 +1,145 @@
|
||||
import unittest
|
||||
from types import SimpleNamespace
|
||||
|
||||
import requests
|
||||
|
||||
from sglang.srt.utils import kill_process_tree
|
||||
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_DEEPPEP_MODEL_NAME_FOR_TEST,
|
||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
popen_launch_server,
|
||||
)
|
||||
|
||||
|
||||
class TestDeepseek(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_DEEPPEP_MODEL_NAME_FOR_TEST
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"8",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"8",
|
||||
"--moe-dense-tp-size",
|
||||
"1",
|
||||
"--enable-dp-lm-head",
|
||||
"--enable-deepep-moe",
|
||||
"--enable-two-batch-overlap",
|
||||
"--ep-num-redundant-experts",
|
||||
"32",
|
||||
"--ep-dispatch-algorithm",
|
||||
"dynamic",
|
||||
"--eplb-algorithm",
|
||||
"deepseek",
|
||||
"--cuda-graph-bs",
|
||||
"256",
|
||||
"--max-running-requests",
|
||||
"2048",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=8,
|
||||
data_path=None,
|
||||
num_questions=1250,
|
||||
parallel=1250,
|
||||
max_new_tokens=512,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(f"Eval accuracy of GSM8K: {metrics=}")
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.93)
|
||||
|
||||
|
||||
class TestDeepseekMTP(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_DEEPPEP_MODEL_NAME_FOR_TEST
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"8",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"8",
|
||||
"--moe-dense-tp-size",
|
||||
"1",
|
||||
"--enable-dp-lm-head",
|
||||
"--enable-deepep-moe",
|
||||
"--enable-two-batch-overlap",
|
||||
"--ep-num-redundant-experts",
|
||||
"32",
|
||||
"--ep-dispatch-algorithm",
|
||||
"dynamic",
|
||||
"--eplb-algorithm",
|
||||
"deepseek",
|
||||
"--cuda-graph-bs",
|
||||
"64", # TODO: increase it to 128 when TBO is supported in draft_extend
|
||||
"--max-running-requests",
|
||||
"512",
|
||||
"--speculative-algorithm",
|
||||
"NEXTN",
|
||||
"--speculative-num-steps",
|
||||
"1",
|
||||
"--speculative-eagle-topk",
|
||||
"1",
|
||||
"--speculative-num-draft-tokens",
|
||||
"2",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=8,
|
||||
data_path=None,
|
||||
num_questions=1250,
|
||||
parallel=1250,
|
||||
max_new_tokens=512,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(f"Eval accuracy of GSM8K: {metrics=}")
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.93)
|
||||
|
||||
server_info = requests.get(self.base_url + "/get_server_info")
|
||||
avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
"avg_spec_accept_length"
|
||||
]
|
||||
print(
|
||||
f"###test_gsm8k:\n"
|
||||
f"accuracy={metrics['accuracy']=:.3f}\n"
|
||||
f"{avg_spec_accept_length=:.3f}\n"
|
||||
)
|
||||
self.assertGreater(avg_spec_accept_length, 1.9)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
384
test/srt/test_deepep_small.py
Normal file
384
test/srt/test_deepep_small.py
Normal file
@@ -0,0 +1,384 @@
|
||||
import unittest
|
||||
from types import SimpleNamespace
|
||||
|
||||
import requests
|
||||
|
||||
from sglang.srt.utils import kill_process_tree
|
||||
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_MODEL_NAME_FOR_TEST_MLA,
|
||||
DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
|
||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
popen_launch_server,
|
||||
)
|
||||
|
||||
|
||||
class TestPureDP(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"4",
|
||||
"--enable-deepep-moe",
|
||||
"--cuda-graph-max-bs",
|
||||
"128",
|
||||
"--max-running-requests",
|
||||
"128",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.62)
|
||||
|
||||
|
||||
class TestHybridDPTP(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"2",
|
||||
"--enable-deepep-moe",
|
||||
"--cuda-graph-max-bs",
|
||||
"128",
|
||||
"--max-running-requests",
|
||||
"128",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.62)
|
||||
|
||||
|
||||
class TestTP(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--enable-deepep-moe",
|
||||
"--cuda-graph-max-bs",
|
||||
"128",
|
||||
"--max-running-requests",
|
||||
"128",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.62)
|
||||
|
||||
|
||||
@unittest.skip("covered in test_deepep_large.py")
|
||||
class TestNoGatherdBuffer(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"4",
|
||||
"--moe-dense-tp-size",
|
||||
"1",
|
||||
"--enable-dp-lm-head",
|
||||
"--enable-deepep-moe",
|
||||
"--cuda-graph-max-bs",
|
||||
"32",
|
||||
"--max-running-requests",
|
||||
"128",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.62)
|
||||
|
||||
|
||||
class TestTBO(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"4",
|
||||
"--moe-dense-tp-size",
|
||||
"1",
|
||||
"--enable-deepep-moe",
|
||||
"--enable-two-batch-overlap",
|
||||
"--cuda-graph-max-bs",
|
||||
"128",
|
||||
"--max-running-requests",
|
||||
"128",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.62)
|
||||
|
||||
|
||||
@unittest.skip("covered in TestMTPWithTBO")
|
||||
class TestMTP(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--trust-remote-code",
|
||||
"--tp",
|
||||
"4",
|
||||
"--enable-dp-attention",
|
||||
"--dp",
|
||||
"2",
|
||||
"--enable-dp-lm-head",
|
||||
"--enable-deepep-moe",
|
||||
"--speculative-algo",
|
||||
"NEXTN",
|
||||
"--speculative-draft",
|
||||
DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
|
||||
"--speculative-num-steps",
|
||||
"2",
|
||||
"--speculative-eagle-topk",
|
||||
"3",
|
||||
"--speculative-num-draft-tokens",
|
||||
"3",
|
||||
"--cuda-graph-max-bs",
|
||||
"32",
|
||||
"--max-running-requests",
|
||||
"32",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
requests.get(self.base_url + "/flush_cache")
|
||||
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.60)
|
||||
|
||||
server_info = requests.get(self.base_url + "/get_server_info")
|
||||
avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
"avg_spec_accept_length"
|
||||
]
|
||||
print(
|
||||
f"###test_gsm8k (deepseek-v3 mtp + dp + tbo):\n"
|
||||
f"accuracy={metrics['accuracy']=:.3f}\n"
|
||||
f"{avg_spec_accept_length=:.3f}\n"
|
||||
)
|
||||
self.assertGreater(avg_spec_accept_length, 2.1)
|
||||
|
||||
|
||||
class TestMTPWithTBO(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
import os
|
||||
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=[
|
||||
"--tp-size",
|
||||
"4",
|
||||
"--enable-dp-attention",
|
||||
"--dp-size",
|
||||
"4",
|
||||
"--enable-two-batch-overlap",
|
||||
"--enable-deepep-moe",
|
||||
"--trust-remote-code",
|
||||
"--speculative-algorithm",
|
||||
"NEXTN",
|
||||
"--speculative-num-steps",
|
||||
"2",
|
||||
"--speculative-eagle-topk",
|
||||
"3",
|
||||
"--speculative-num-draft-tokens",
|
||||
"3",
|
||||
"--speculative-draft",
|
||||
DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
|
||||
"--chunked-prefill-size",
|
||||
"256",
|
||||
"--cuda-graph-max-bs",
|
||||
"32",
|
||||
"--max-running-requests",
|
||||
"32",
|
||||
],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_gsm8k(self):
|
||||
requests.get(self.base_url + "/flush_cache")
|
||||
|
||||
args = SimpleNamespace(
|
||||
num_shots=5,
|
||||
data_path=None,
|
||||
num_questions=200,
|
||||
max_new_tokens=512,
|
||||
parallel=128,
|
||||
host="http://127.0.0.1",
|
||||
port=int(self.base_url.split(":")[-1]),
|
||||
)
|
||||
metrics = run_eval_few_shot_gsm8k(args)
|
||||
print(metrics)
|
||||
|
||||
self.assertGreater(metrics["accuracy"], 0.60)
|
||||
|
||||
server_info = requests.get(self.base_url + "/get_server_info")
|
||||
avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
"avg_spec_accept_length"
|
||||
]
|
||||
print(
|
||||
f"###test_gsm8k (deepseek-v3 mtp + dp + tbo):\n"
|
||||
f"accuracy={metrics['accuracy']=:.3f}\n"
|
||||
f"{avg_spec_accept_length=:.3f}\n"
|
||||
)
|
||||
self.assertGreater(avg_spec_accept_length, 2.1)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -137,86 +137,5 @@ class TestDPAttentionDP2TP2DeepseekV3MTP(CustomTestCase):
|
||||
self.assertGreater(avg_spec_accept_length, 2.5)
|
||||
|
||||
|
||||
# TODO: enable this test later
|
||||
# class TestDPAttentionDP2TP2DeepseekV3MTPTBO(CustomTestCase):
|
||||
# @classmethod
|
||||
# def setUpClass(cls):
|
||||
# import os
|
||||
|
||||
# # print debug log for tbo
|
||||
# os.environ["SGLANG_TBO_DEBUG"] = "1"
|
||||
# cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
|
||||
# cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
# other_args = [
|
||||
# "--trust-remote-code",
|
||||
# "--disable-radix",
|
||||
# "--speculative-algorithm",
|
||||
# "EAGLE",
|
||||
# "--speculative-num-steps",
|
||||
# "2",
|
||||
# "--speculative-eagle-topk",
|
||||
# "4",
|
||||
# "--speculative-num-draft-tokens",
|
||||
# "4",
|
||||
# "--speculative-draft",
|
||||
# DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
|
||||
# "--tp-size",
|
||||
# "2",
|
||||
# "--enable-dp-attention",
|
||||
# "--dp-size",
|
||||
# "2",
|
||||
# "--enable-two-batch-overlap",
|
||||
# "--enable-deepep-moe",
|
||||
# "--deepep-mode",
|
||||
# "low_latency",
|
||||
# "--chunked-prefill-size",
|
||||
# "256",
|
||||
# "--cuda-graph-max-bs",
|
||||
# "32",
|
||||
# "--max-running-requests",
|
||||
# "32",
|
||||
# ]
|
||||
# if not is_in_amd_ci():
|
||||
# other_args += ["--mem-frac", "0.7"]
|
||||
# cls.process = popen_launch_server(
|
||||
# cls.model,
|
||||
# cls.base_url,
|
||||
# timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
# other_args=other_args,
|
||||
# )
|
||||
|
||||
# @classmethod
|
||||
# def tearDownClass(cls):
|
||||
# kill_process_tree(cls.process.pid)
|
||||
|
||||
# def test_gsm8k(self):
|
||||
# requests.get(self.base_url + "/flush_cache")
|
||||
|
||||
# args = SimpleNamespace(
|
||||
# num_shots=5,
|
||||
# data_path=None,
|
||||
# num_questions=200,
|
||||
# max_new_tokens=512,
|
||||
# parallel=128,
|
||||
# host="http://127.0.0.1",
|
||||
# port=int(self.base_url.split(":")[-1]),
|
||||
# )
|
||||
# metrics = run_eval_few_shot_gsm8k(args)
|
||||
# print(metrics)
|
||||
|
||||
# self.assertGreater(metrics["accuracy"], 0.60)
|
||||
|
||||
# server_info = requests.get(self.base_url + "/get_server_info")
|
||||
# avg_spec_accept_length = server_info.json()["internal_states"][0][
|
||||
# "avg_spec_accept_length"
|
||||
# ]
|
||||
# print(
|
||||
# f"###test_gsm8k (deepseek-v3 mtp + dp + tbo):\n"
|
||||
# f"accuracy={metrics['accuracy']=:.3f}\n"
|
||||
# f"{avg_spec_accept_length=:.3f}\n"
|
||||
# )
|
||||
# self.assertGreater(avg_spec_accept_length, 2.3)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
|
||||
Reference in New Issue
Block a user