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
[Misc] Avoid use of deprecated `AutoModelForVision2Seq` (#25065)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Sep 17, 2025 at 20:19 UTC
e120533d7ae3bc8c3ef39f215de274f1280bb454
1 file changed
+7
-7
tests/models/multimodal/generation/test_common.py
+7
-7
@@ -10,7 +10,7 @@ from pathlib import PosixPath
10
11
import pytest
12
from transformers import (AutoModel, AutoModelForImageTextToText,
13
- AutoModelForTextToWaveform, AutoModelForVision2Seq)
13
+ AutoModelForTextToWaveform)
14
15
from vllm.platforms import current_platform
16
from vllm.utils import identity
@@ -137,7 +137,7 @@ VLM_TEST_SETTINGS = {
137
video_idx_to_prompt=lambda idx: "<|vision_start|><|video_pad|><|vision_end|>", # noqa: E501
138
max_model_len=4096,
139
max_num_seqs=2,
140
- auto_cls=AutoModelForVision2Seq,
140
+ auto_cls=AutoModelForImageTextToText,
141
vllm_output_post_proc=model_utils.qwen2_vllm_to_hf_output,
142
image_size_factors=[(), (0.25,), (0.25, 0.25, 0.25), (0.25, 0.2, 0.15)],
143
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
@@ -502,7 +502,7 @@ VLM_TEST_SETTINGS = {
502
num_video_frames=16,
503
max_model_len=16384,
504
hf_model_kwargs=model_utils.llava_onevision_hf_model_kwargs("llava-hf/llava-onevision-qwen2-0.5b-ov-hf"), # noqa: E501
505
- auto_cls=AutoModelForVision2Seq,
505
+ auto_cls=AutoModelForImageTextToText,
506
vllm_output_post_proc=model_utils.llava_onevision_vllm_to_hf_output,
507
custom_test_opts=[CustomTestOptions(
508
inputs=custom_inputs.multi_video_multi_aspect_ratio_inputs(
@@ -518,7 +518,7 @@ VLM_TEST_SETTINGS = {
518
num_video_frames=16,
519
max_model_len=4096,
520
max_num_seqs=2,
521
- auto_cls=AutoModelForVision2Seq,
521
+ auto_cls=AutoModelForImageTextToText,
522
vllm_output_post_proc=model_utils.llava_video_vllm_to_hf_output,
523
),
524
"mantis": VLMTestInfo(
@@ -680,7 +680,7 @@ VLM_TEST_SETTINGS = {
680
multi_image_prompt="Picture 1: <vlm_image>\nPicture 2: <vlm_image>\nDescribe these two images with one paragraph respectively.", # noqa: E501
681
max_model_len=4096,
682
max_num_seqs=2,
683
- auto_cls=AutoModelForVision2Seq,
683
+ auto_cls=AutoModelForImageTextToText,
684
vllm_output_post_proc=model_utils.qwen2_vllm_to_hf_output,
685
image_size_factors=[(), (0.25,), (0.25, 0.25, 0.25), (0.25, 0.2, 0.15)],
686
marks=[pytest.mark.cpu_model],
@@ -784,7 +784,7 @@ VLM_TEST_SETTINGS = {
784
test_type=VLMTestType.CUSTOM_INPUTS,
785
max_model_len=16384,
786
max_num_seqs=2,
787
- auto_cls=AutoModelForVision2Seq,
787
+ auto_cls=AutoModelForImageTextToText,
788
hf_model_kwargs=model_utils.llava_onevision_hf_model_kwargs("llava-hf/llava-onevision-qwen2-0.5b-ov-hf"), # noqa: E501
789
vllm_output_post_proc=model_utils.llava_onevision_vllm_to_hf_output,
790
custom_test_opts=[CustomTestOptions(
@@ -800,7 +800,7 @@ VLM_TEST_SETTINGS = {
800
test_type=VLMTestType.CUSTOM_INPUTS,
801
max_model_len=4096,
802
max_num_seqs=2,
803
- auto_cls=AutoModelForVision2Seq,
803
+ auto_cls=AutoModelForImageTextToText,
804
vllm_output_post_proc=model_utils.qwen2_vllm_to_hf_output,
805
custom_test_opts=[CustomTestOptions(
806
inputs=custom_inputs.windows_attention_image_qwen2_5_vl(),