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 5, 11:00 UTC
[CI Failure] pin nomic-embed-text-v1 revision (#39292)
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
wang.yuqi committed
Apr 8, 2026 at 19:43 UTC
4e2ab1861d17d6df3d0220f7bb82df7864bbd218
5 files changed
+50
-8
tests/models/language/pooling/test_nomic_max_model_len.py
+39
-7
@@ -8,7 +8,13 @@ import pytest
8
from ...utils import EmbedModelInfo
9
10
MODELS = [
11
- EmbedModelInfo("nomic-ai/nomic-embed-text-v1"),
11
+ EmbedModelInfo(
12
+ "nomic-ai/nomic-embed-text-v1",
13
+ # Fixme:
14
+ # Update nomic-embed code to support the latest
15
+ # HF version and remove revision set.
16
+ revision="720244025c1a7e15661a174c63cce63c8218e52b",
17
+ ),
18
# EmbedModelInfo("nomic-ai/nomic-embed-text-v1.5"),
19
# EmbedModelInfo("nomic-ai/CodeRankEmbed"),
20
EmbedModelInfo("nomic-ai/nomic-embed-text-v2-moe"),
@@ -24,7 +30,10 @@ max_model_len = int(original_max_position_embeddings * factor)
30
@pytest.mark.parametrize("model_info", MODELS)
31
def test_default(model_info, vllm_runner):
32
with vllm_runner(
27
- model_info.name, runner="pooling", max_model_len=None
33
+ model_info.name,
34
+ revision=model_info.revision,
35
+ runner="pooling",
36
+ max_model_len=None,
37
) as vllm_model:
38
model_config = vllm_model.llm.llm_engine.model_config
39
if model_info.name == "nomic-ai/nomic-embed-text-v2-moe":
@@ -39,7 +48,10 @@ def test_default(model_info, vllm_runner):
48
def test_set_max_model_len_legal(model_info, vllm_runner):
49
# set max_model_len <= 512
50
with vllm_runner(
42
- model_info.name, runner="pooling", max_model_len=256
51
+ model_info.name,
52
+ revision=model_info.revision,
53
+ runner="pooling",
54
+ max_model_len=256,
55
) as vllm_model:
56
model_config = vllm_model.llm.llm_engine.model_config
57
assert model_config.max_model_len == 256
@@ -49,11 +61,19 @@ def test_set_max_model_len_legal(model_info, vllm_runner):
61
# For nomic-embed-text-v2-moe the length is set to 512
62
# by sentence_bert_config.json.
63
with pytest.raises(ValueError):
52
- with vllm_runner(model_info.name, runner="pooling", max_model_len=1024):
64
+ with vllm_runner(
65
+ model_info.name,
66
+ revision=model_info.revision,
67
+ runner="pooling",
68
+ max_model_len=1024,
69
+ ):
70
pass
71
else:
72
with vllm_runner(
56
- model_info.name, runner="pooling", max_model_len=1024
73
+ model_info.name,
74
+ revision=model_info.revision,
75
+ runner="pooling",
76
+ max_model_len=1024,
77
) as vllm_model:
78
model_config = vllm_model.llm.llm_engine.model_config
79
assert model_config.max_model_len == 1024
@@ -63,7 +83,12 @@ def test_set_max_model_len_legal(model_info, vllm_runner):
83
def test_set_max_model_len_illegal(model_info, vllm_runner):
84
# set max_model_len > 2048
85
with pytest.raises(ValueError):
66
- with vllm_runner(model_info.name, runner="pooling", max_model_len=4096):
86
+ with vllm_runner(
87
+ model_info.name,
88
+ revision=model_info.revision,
89
+ runner="pooling",
90
+ max_model_len=4096,
91
+ ):
92
pass
93
94
# set max_model_len > 2048 by hf_overrides
@@ -71,6 +96,7 @@ def test_set_max_model_len_illegal(model_info, vllm_runner):
96
with pytest.raises(ValueError):
97
with vllm_runner(
98
model_info.name,
99
+ revision=model_info.revision,
100
runner="pooling",
101
max_model_len=None,
102
hf_overrides=hf_overrides,
@@ -91,7 +117,11 @@ def test_use_rope_scaling_legal(model_info, vllm_runner):
117
}
118
119
with vllm_runner(
94
- model_info.name, runner="pooling", max_model_len=None, hf_overrides=hf_overrides
120
+ model_info.name,
121
+ revision=model_info.revision,
122
+ runner="pooling",
123
+ max_model_len=None,
124
+ hf_overrides=hf_overrides,
125
):
126
pass
127
@@ -110,6 +140,7 @@ def test_use_rope_scaling_illegal(model_info, vllm_runner):
140
with pytest.raises(ValueError):
141
with vllm_runner(
142
model_info.name,
143
+ revision=model_info.revision,
144
runner="pooling",
145
max_model_len=max_model_len + 1,
146
hf_overrides=hf_overrides,
@@ -129,6 +160,7 @@ def test_use_rope_scaling_illegal(model_info, vllm_runner):
160
with pytest.raises(ValueError):
161
with vllm_runner(
162
model_info.name,
163
+ revision=model_info.revision,
164
runner="pooling",
165
max_model_len=None,
166
hf_overrides=hf_overrides,
tests/models/language/pooling_mteb_test/mteb_embed_utils.py
+2
@@ -151,6 +151,7 @@ def mteb_test_embed_models(
151
152
with vllm_runner(
153
model_info.name,
154
+ revision=model_info.revision,
155
runner="pooling",
156
max_model_len=model_info.max_model_len,
157
**vllm_extra_kwargs,
@@ -201,6 +202,7 @@ def mteb_test_embed_models(
202
if model_info.mteb_score is None:
203
with hf_runner(
204
model_info.name,
205
+ revision=model_info.revision,
206
is_sentence_transformer=True,
207
dtype=ci_envs.VLLM_CI_HF_DTYPE or model_info.hf_dtype,
208
) as hf_model:
tests/models/language/pooling_mteb_test/mteb_score_utils.py
+4
-1
@@ -241,6 +241,7 @@ def mteb_test_rerank_models(
241
242
with vllm_runner(
243
model_info.name,
244
+ revision=model_info.revision,
245
runner="pooling",
246
max_model_len=None,
247
max_num_seqs=8,
@@ -286,7 +287,9 @@ def mteb_test_rerank_models(
287
# Accelerate mteb test by setting
288
# SentenceTransformers mteb score to a constant
289
if model_info.mteb_score is None:
289
- with hf_runner(model_info.name, dtype=model_info.hf_dtype) as hf_model:
290
+ with hf_runner(
291
+ model_info.name, revision=model_info.revision, dtype=model_info.hf_dtype
292
+ ) as hf_model:
293
hf_model.chat_template = chat_template
294
st_main_score = run_mteb_rerank(
295
hf_model,
tests/models/language/pooling_mteb_test/test_nomic.py
+4
@@ -12,6 +12,10 @@ MODELS = [
12
EmbedModelInfo(
13
"nomic-ai/nomic-embed-text-v1",
14
architecture="NomicBertModel",
15
+ # Fixme:
16
+ # Update nomic-embed code to support the latest
17
+ # HF version and remove revision set.
18
+ revision="720244025c1a7e15661a174c63cce63c8218e52b",
19
mteb_score=0.737568559,
20
enable_test=True,
21
seq_pooling_type="MEAN",
tests/models/utils.py
+1
@@ -375,6 +375,7 @@ def softmax(data):
375
@dataclass
376
class ModelInfo:
377
name: str
378
+ revision: str | None = None
379
architecture: str = ""
380
dtype: str = "auto"
381
max_model_len: int | None = None