A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI] Enable encoder model compilation test (#24442)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Jiangyun Zhu committed
Sep 9, 2025 at 02:48 UTC
7be141b2c52366f0cc4e731c36819aed178d2258
1 file changed
+17
-14
tests/compile/test_basic_correctness.py
+17
-14
@@ -62,8 +62,12 @@ class TestSetting:
62
TestSetting(
63
model="BAAI/bge-multilingual-gemma2",
64
model_args=[
65
- "--runner", "pooling", "--dtype", "bfloat16",
66
- "--max-model-len", "2048"
65
+ "--runner",
66
+ "pooling",
67
+ "--dtype",
68
+ "bfloat16",
69
+ "--max-model-len",
70
+ "2048",
71
],
72
pp_size=1,
73
tp_size=1,
@@ -71,17 +75,15 @@ class TestSetting:
75
method="encode",
76
fullgraph=True,
77
),
74
- # TODO: bert models are not supported in V1 yet
75
- # # encoder-based embedding model (BERT)
76
- # TestSetting(
77
- # model="BAAI/bge-base-en-v1.5",
78
- # model_args=["--runner", "pooling"],
79
- # pp_size=1,
80
- # tp_size=1,
81
- # attn_backend="XFORMERS",
82
- # method="encode",
83
- # fullgraph=True,
84
- # ),
78
+ TestSetting(
79
+ model="BAAI/bge-base-en-v1.5",
80
+ model_args=["--runner", "pooling"],
81
+ pp_size=1,
82
+ tp_size=1,
83
+ attn_backend="FLASH_ATTN",
84
+ method="encode",
85
+ fullgraph=True,
86
+ ),
87
# vision language model
88
TestSetting(
89
model="microsoft/Phi-3.5-vision-instruct",
@@ -92,7 +94,8 @@ class TestSetting:
94
method="generate_with_image",
95
fullgraph=False,
96
),
95
- ])
97
+ ],
98
+)
99
def test_compile_correctness(
100
monkeypatch: pytest.MonkeyPatch,
101
test_setting: TestSetting,