Unverified Commit 56405076 authored by Yineng Zhang's avatar Yineng Zhang Committed by GitHub
Browse files

fix: update dsv3 fp4 ut (#10584)

parent b73ac629
...@@ -141,8 +141,8 @@ class TestDeepseekV3FP4MTP(CustomTestCase): ...@@ -141,8 +141,8 @@ class TestDeepseekV3FP4MTP(CustomTestCase):
f'{metrics["accuracy"]=:.3f}\n' f'{metrics["accuracy"]=:.3f}\n'
f"{avg_spec_accept_length=:.2f}\n" f"{avg_spec_accept_length=:.2f}\n"
) )
self.assertGreater(metrics["accuracy"], 0.935) self.assertGreater(metrics["accuracy"], 0.94)
self.assertGreater(avg_spec_accept_length, 2.9) self.assertGreater(avg_spec_accept_length, 2.04)
def test_bs_1_speed(self): def test_bs_1_speed(self):
args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048) args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
...@@ -156,8 +156,8 @@ class TestDeepseekV3FP4MTP(CustomTestCase): ...@@ -156,8 +156,8 @@ class TestDeepseekV3FP4MTP(CustomTestCase):
f"{acc_length=:.2f}\n" f"{acc_length=:.2f}\n"
f"{speed=:.2f} token/s\n" f"{speed=:.2f} token/s\n"
) )
self.assertGreater(acc_length, 2.9) self.assertGreater(acc_length, 2.04)
self.assertGreater(speed, 130) self.assertGreater(speed, 150)
class TestDeepseekV3FP4CutlassMoE(CustomTestCase): class TestDeepseekV3FP4CutlassMoE(CustomTestCase):
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment