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 15, 19:45 UTC
[Bugfix] Fix Aria test loading (#14823)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Mar 15, 2025 at 00:11 UTC
613c5bb9458f9ced270fd98c9d877e00c0ad165b
1 file changed
+3
-1
tests/models/decoder_only/vision_language/test_models.py
+3
-1
@@ -9,7 +9,8 @@ from pathlib import PosixPath
9
10
import pytest
11
from packaging.version import Version
12
-from transformers import AutoModelForPreTraining, AutoModelForVision2Seq
12
+from transformers import (AutoModelForImageTextToText, AutoModelForPreTraining,
13
+ AutoModelForVision2Seq)
14
from transformers import __version__ as TRANSFORMERS_VERSION
15
16
from vllm.platforms import current_platform
@@ -163,6 +164,7 @@ VLM_TEST_SETTINGS = {
164
img_idx_to_prompt=lambda idx: "<fim_prefix><|img|><fim_suffix>\n",
165
max_model_len=4096,
166
max_num_seqs=2,
167
+ auto_cls=AutoModelForImageTextToText,
168
single_image_prompts=IMAGE_ASSETS.prompts({
169
"stop_sign": "<vlm_image>Please describe the image shortly.",
170
"cherry_blossom": "<vlm_image>Please infer the season with reason.",