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 7, 22:45 UTC
[CI] Bump mteb version to `mteb[bm25s]>=2, <3` for pooling model unit tests (#34961)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Feb 21, 2026 at 23:23 UTC
d24bdd7c4b3eb61a8a025068b6b3ad4d8041abf7
4 files changed
+6
-3
requirements/nightly_torch_test.txt
+1
-1
@@ -28,7 +28,7 @@ num2words # required for smolvlm test
28
opencv-python-headless >= 4.13.0 # required for video test
29
datamodel_code_generator # required for minicpm3 test
30
lm-eval[api]>=0.4.11 # required for model evaluation test
31
-mteb>=1.38.11, <2 # required for mteb test
31
+mteb[bm25s]>=2, <3 # required for mteb test
32
transformers==4.57.5
33
tokenizers==0.22.0
34
schemathesis>=3.39.15 # Required for openai schema test.
requirements/rocm-test.txt
+1
-1
@@ -70,7 +70,7 @@ ray[cgraph,default]>=2.48.0
70
torchgeo==0.7.0
71
# via terratorch
72
# MTEB Benchmark Test
73
-mteb==2.1.2
73
+mteb[bm25s]>=2, <3
74
75
# Utilities
76
num2words==0.5.14
requirements/test.txt
+1
-1
@@ -491,7 +491,7 @@ msgpack==1.1.0
491
# via
492
# librosa
493
# ray
494
-mteb==2.1.2
494
+mteb==2.8.3
495
# via -r requirements/test.in
496
multidict==6.1.0
497
# via
tests/models/language/pooling_mteb_test/mteb_score_utils.py
+3
@@ -191,6 +191,9 @@ def run_mteb_rerank(cross_encoder: mteb.CrossEncoderProtocol, tasks, languages):
191
mteb_tasks: list[mteb.abstasks.AbsTaskRetrieval] = mteb.get_tasks(
192
tasks=tasks, languages=languages, eval_splits=eval_splits
193
)
194
+ for task in mteb_tasks:
195
+ if not task.data_loaded:
196
+ task.load_data()
197
198
mteb.evaluate(
199
bm25s,