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 6, 14:30 UTC
Fix Phi3 test that fails with Transformers v5 (#37298)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Mar 17, 2026 at 14:29 UTC
ecfcdd2ce47e2216eee11550645b1bb3cfa44d7b
1 file changed
+7
tests/models/multimodal/pooling/test_phi3v.py
+7
@@ -3,6 +3,7 @@
3
4
import pytest
5
import torch.nn.functional as F
6
+import transformers.utils
7
from PIL import Image
8
9
from vllm.assets.base import get_vllm_public_assets
@@ -12,6 +13,12 @@ from ....conftest import IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner
13
from ....utils import large_gpu_test
14
from ...utils import check_embeddings_close
15
16
+# BC for method that was deleted in Transformers v5.
17
+# Only needed for generating the HF reference.
18
+transformers.utils.is_flash_attn_greater_or_equal_2_10 = (
19
+ lambda: transformers.utils.is_flash_attn_greater_or_equal("2.1.0")
20
+)
21
+
22
HF_TEXT_PROMPTS = [
23
# T -> X
24
"Find me an everyday image that matches the given caption: The label of the object is stop sign", # noqa: E501