chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Add EAGLE3 test with MMLU dataset. (#15945)

Browse Source
This commit is contained in:
Liangsheng Yin
2025-12-28 16:07:58 +08:00
committed by GitHub
parent bf90ea9c5b
commit c7e7bfa32c
8 changed files with 63 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/registered/attention/test_fa3.py
View File

@@ -115,8 +115,8 @@ class BaseFlashAttentionTest(CustomTestCase):
self.assertGreater(metrics[metric_key], self.accuracy_threshold)
if self.speculative_decode:
server_info = requests.get(self.base_url + "/get_server_info")
avg_spec_accept_length = server_info.json()["internal_states"][0][
server_info = requests.get(self.base_url + "/server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")

5
test/registered/mla/test_flashmla.py
View File

@@ -125,9 +125,8 @@ class TestFlashMLAMTP(CustomTestCase):
self.assertGreater(metrics["accuracy"], 0.60)
server_info = requests.get(self.base_url + "/get_server_info")
print(f"{server_info=}")
avg_spec_accept_length = server_info.json()["internal_states"][0][
server_info = requests.get(self.base_url + "/server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")

5
test/registered/mla/test_mla_flashinfer.py
View File

@@ -115,9 +115,8 @@ class TestFlashinferMLAMTP(CustomTestCase):
self.assertGreater(metrics["accuracy"], 0.60)
server_info = requests.get(self.base_url + "/get_server_info")
print(f"{server_info=}")
avg_spec_accept_length = server_info.json()["internal_states"][0][
server_info = requests.get(self.base_url + "/get_server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")

4
test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_stage_b.py
View File

@@ -79,8 +79,8 @@ class TestDeepseekV3FP4MTP(CustomTestCase):
metrics = run_eval_few_shot_gsm8k(args)
print(f"{metrics=}")
server_info = requests.get(self.base_url + "/get_server_info")
avg_spec_accept_length = server_info.json()["internal_states"][0][
server_info = requests.get(self.base_url + "/server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")

50
test/registered/spec/eagle/test_eagle3_basic.py Normal file
View File

@@ -0,0 +1,50 @@
import unittest
from types import SimpleNamespace
import requests
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.run_eval import run_eval
from sglang.test.server_fixtures.eagle_fixture import EagleServerBase
from sglang.test.test_utils import (
DEFAULT_DRAFT_MODEL_EAGLE3,
DEFAULT_TARGET_MODEL_EAGLE3,
)
register_cuda_ci(est_time=50, suite="stage-b-test-small-1-gpu")
class TestEagle3Basic(EagleServerBase):
target_model = DEFAULT_TARGET_MODEL_EAGLE3
draft_model = DEFAULT_DRAFT_MODEL_EAGLE3
spec_algo = "EAGLE3"
spec_steps = 2
spec_topk = 1
spec_tokens = 3
extra_args = ["--dtype=float16", "--chunked-prefill-size", 1024]
def test_mmlu(self):
"""Override to add EAGLE-specific assertions"""
args = SimpleNamespace(
base_url=self.base_url,
model=self.target_model,
eval_name="mmlu",
num_examples=64,
num_threads=32,
)
metrics = run_eval(args)
self.assertGreaterEqual(metrics["score"], 0.72)
server_info = requests.get(self.base_url + "/server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")
self.assertGreater(avg_spec_accept_length, 2.26)
if __name__ == "__main__":
unittest.main()

2
test/registered/spec/eagle/test_eagle_infer_b.py
View File

@@ -75,7 +75,7 @@ class TestEAGLEServerBasic(EagleServerBase):
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.20)
server_info = requests.get(self.base_url + "/get_server_info").json()
server_info = requests.get(self.base_url + "/server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]

6
test/registered/spec/eagle/test_eagle_infer_beta_dp_attention_large.py
View File

@@ -34,10 +34,8 @@ def test_gsm8k(base_url: str):
port=int(base_url.split(":")[-1]),
)
metrics = run_eval_few_shot_gsm8k(args)
server_info = requests.get(base_url + "/get_server_info")
avg_spec_accept_length = server_info.json()["internal_states"][0][
"avg_spec_accept_length"
]
server_info = requests.get(base_url + "/server_info").json()
avg_spec_accept_length = server_info["internal_states"][0]["avg_spec_accept_length"]
print(f"{metrics=}")
print(f"{avg_spec_accept_length=}")

5
test/srt/hicache/test_hicache_variants.py
View File

@@ -152,9 +152,8 @@ class TestHiCacheEagle(HiCacheBaseServer, HiCacheEvalMixin):
self.assertGreaterEqual(metrics["score"], self.expected_mmlu_score)
# EAGLE-specific check
server_info = requests.get(self.base_url + "/get_server_info")
print(f"{server_info=}")
avg_spec_accept_length = server_info.json()["internal_states"][0][
server_info = requests.get(self.base_url + "/get_server_info").json()
avg_spec_accept_length = server_info["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")