[CI/Build] Bump test transformers version

requirements-test.txt

@@ -550,7 +550,7 @@ tqdm==4.66.6
     #   transformers
 tqdm-multiprocess==0.0.11
     # via lm-eval
-transformers==4.45.2
+transformers==4.46.3
     # via
     #   lm-eval
     #   peft

tests/models/decoder_only/vision_language/test_models.py

@@ -6,7 +6,6 @@
 from typing import Type
 
 import pytest
-import transformers
 from transformers import AutoModelForVision2Seq
 from transformers.utils import is_flash_attn_2_available
 
@@ -187,12 +186,6 @@
         comparator=check_outputs_equal,
         max_tokens=8,
         dtype="bfloat16",
-        marks=[
-            pytest.mark.skipif(
-                transformers.__version__ < "4.46.2",
-                reason="Model broken in HF, see huggingface/transformers#34379"
-            ),
-        ]
     ),
     "fuyu": VLMTestInfo(
         models=["adept/fuyu-8b"],
@@ -243,13 +236,7 @@
         max_model_len=8192,
         max_num_seqs=2,
         auto_cls=AutoModelForVision2Seq,
-        marks=[
-            pytest.mark.skipif(
-                transformers.__version__ < "4.46.0",
-                reason="Model introduced in HF >= 4.46.0"
-            ),
-            large_gpu_mark(min_gb=48),
-        ],
+        marks=[large_gpu_mark(min_gb=48)],
     ),
     "intern_vl": VLMTestInfo(
         models=[
@@ -318,12 +305,6 @@
         auto_cls=AutoModelForVision2Seq,
         vllm_output_post_proc=model_utils.llava_video_vllm_to_hf_output,
         image_sizes=[((1669, 2560), (2560, 1669), (183, 488), (488, 183))],
-        marks=[
-            pytest.mark.skipif(
-                transformers.__version__ < "4.46.2",
-                reason="Model broken with changes in transformers 4.46"
-            )
-        ],
     ),
     "minicpmv_25": VLMTestInfo(
         models=["openbmb/MiniCPM-Llama3-V-2_5"],
@@ -404,10 +385,6 @@
                 cuda_device_count_stateless() < 2,
                 reason="Need at least 2 GPUs to run the test.",
             ),
-            pytest.mark.skipif(
-                transformers.__version__ < "4.46.2",
-                reason="Model broken in HF, see huggingface/transformers#34379"
-            )
         ],
         **COMMON_BROADCAST_SETTINGS # type: ignore
     ),

tests/models/decoder_only/vision_language/test_pixtral.py

@@ -228,7 +228,7 @@ def test_model_engine(vllm_runner, model: str, dtype: str) -> None:
                          name_1="output")
 
 
-@large_gpu_test(min_gb=24)
+@large_gpu_test(min_gb=48)
 @pytest.mark.parametrize(
     "prompt,expected_ranges",
     [(_create_engine_inputs_hf(IMG_URLS[:1]), [{

tests/models/embedding/vision_language/test_llava_next.py

@@ -2,7 +2,6 @@
 
 import pytest
 import torch.nn.functional as F
-import transformers
 from transformers import AutoModelForVision2Seq
 
 from ....conftest import IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner
@@ -86,9 +85,6 @@ def _run_test(
     )
 
 
-@pytest.mark.skipif(transformers.__version__.startswith("4.46"),
-                    reason="Model broken with changes in transformers 4.46")
-@pytest.mark.core_model
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("dtype", ["half"])
 def test_models_text(

tests/models/test_initialization.py

@@ -1,7 +1,6 @@
 from unittest.mock import patch
 
 import pytest
-import transformers
 from transformers import PretrainedConfig
 
 from vllm import LLM
@@ -11,10 +10,6 @@
 
 @pytest.mark.parametrize("model_arch", HF_EXAMPLE_MODELS.get_supported_archs())
 def test_can_initialize(model_arch):
-    if (model_arch in {"Idefics3ForConditionalGeneration", "GlmForCausalLM"}
-            and transformers.__version__ < "4.46.0"):
-        pytest.skip(reason="Model introduced in HF >= 4.46.0")
-
     model_info = HF_EXAMPLE_MODELS.get_hf_info(model_arch)
     if not model_info.is_available_online:
         pytest.skip("Model is not available online")