[Misc] Update config loading for Qwen2-VL and remove Granite (vllm-project#8837)

Author: ywang96

docs/source/models/supported_models.rst

@@ -280,7 +280,7 @@ Multimodal Language Models
     - :code:`Qwen/Qwen-VL`, :code:`Qwen/Qwen-VL-Chat`, etc.
     -
   * - :code:`Qwen2VLForConditionalGeneration`
-    - Qwen2-VL (see note)
+    - Qwen2-VL
     - Image\ :sup:`+` / Video\ :sup:`+`
     - :code:`Qwen/Qwen2-VL-2B-Instruct`, :code:`Qwen/Qwen2-VL-7B-Instruct`, :code:`Qwen/Qwen2-VL-72B-Instruct`, etc.
     -
@@ -297,15 +297,6 @@ Multimodal Language Models
   For :code:`openbmb/MiniCPM-V-2`, the official repo doesn't work yet, so we need to use a fork (:code:`HwwwH/MiniCPM-V-2`) for now.
   For more details, please see: https://github.com/vllm-project/vllm/pull/4087#issuecomment-2250397630
 
-.. note::
-  For :code:`Qwen2-VL`, the latest release of :code:`huggingface/transformers` doesn't work yet, so we need to use a developer version (:code:`21fac7abba2a37fae86106f87fcf9974fd1e3830`) for now.
-  This can be installed by running the following command: 
-
-  .. code-block:: bash
-    
-    pip install git+https://github.com/huggingface/transformers.git@21fac7abba2a37fae86106f87fcf9974fd1e3830
-
-----
 
 If your model uses one of the above model architectures, you can seamlessly run your model with vLLM.
 Otherwise, please refer to :ref:`Adding a New Model <adding_a_new_model>` and :ref:`Enabling Multimodal Inputs <enabling_multimodal_inputs>` 

vllm/model_executor/models/granite.py

@@ -25,6 +25,7 @@
 
 import torch
 from torch import nn
+from transformers import GraniteConfig
 
 from vllm.attention import Attention, AttentionMetadata
 from vllm.config import CacheConfig, LoRAConfig
@@ -48,7 +49,6 @@
     default_weight_loader, kv_cache_scales_loader, maybe_remap_kv_scale_name)
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
-from vllm.transformers_utils.configs.granite import GraniteConfig
 from vllm.utils import is_hip
 
 from .interfaces import SupportsLoRA

vllm/model_executor/models/qwen2_vl.py

@@ -31,12 +31,9 @@
 import torch.nn.functional as F
 from einops import rearrange, repeat
 from PIL import Image
-from transformers import Qwen2VLConfig
 from transformers.image_utils import (get_image_size,
                                       infer_channel_dimension_format,
                                       to_numpy_array)
-from transformers.models.qwen2_vl.configuration_qwen2_vl import (
-    Qwen2VLVisionConfig)
 from transformers.models.qwen2_vl.image_processing_qwen2_vl import (
     make_batched_images, make_batched_videos, smart_resize)
 
@@ -66,6 +63,8 @@
 from vllm.multimodal.image import cached_get_image_processor
 from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors, SequenceData
+from vllm.transformers_utils.configs.qwen2vl import (Qwen2VLConfig,
+                                                     Qwen2VLVisionConfig)
 from vllm.transformers_utils.processor import get_processor
 from vllm.utils import is_cpu
 

vllm/transformers_utils/config.py

@@ -20,10 +20,10 @@
 # yapf: disable
 from vllm.transformers_utils.configs import (ChatGLMConfig, DbrxConfig,
                                              EAGLEConfig, ExaoneConfig,
-                                             GraniteConfig, InternVLChatConfig,
-                                             JAISConfig, MedusaConfig,
-                                             MllamaConfig, MLPSpeculatorConfig,
-                                             MPTConfig, NemotronConfig,
+                                             InternVLChatConfig, JAISConfig,
+                                             MedusaConfig, MllamaConfig,
+                                             MLPSpeculatorConfig, MPTConfig,
+                                             NemotronConfig, Qwen2VLConfig,
                                              RWConfig, SolarConfig,
                                              UltravoxConfig)
 # yapf: enable
@@ -57,9 +57,7 @@
     "nemotron": NemotronConfig,
     "solar": SolarConfig,
     "ultravox": UltravoxConfig,
-    # Granite can be removed from here once we have upgraded to
-    # transformers 4.45+
-    "granite": GraniteConfig,
+    "qwen2_vl": Qwen2VLConfig,
     **_CONFIG_REGISTRY_OVERRIDE_HF
 }
 

vllm/transformers_utils/configs/__init__.py

@@ -6,14 +6,15 @@
 # tiiuae/falcon-7b(-instruct) models. Newer Falcon models will use the
 # `FalconConfig` class from the official HuggingFace transformers library.
 from vllm.transformers_utils.configs.falcon import RWConfig
-from vllm.transformers_utils.configs.granite import GraniteConfig
 from vllm.transformers_utils.configs.internvl import InternVLChatConfig
 from vllm.transformers_utils.configs.jais import JAISConfig
 from vllm.transformers_utils.configs.medusa import MedusaConfig
 from vllm.transformers_utils.configs.mllama import MllamaConfig
 from vllm.transformers_utils.configs.mlp_speculator import MLPSpeculatorConfig
 from vllm.transformers_utils.configs.mpt import MPTConfig
 from vllm.transformers_utils.configs.nemotron import NemotronConfig
+from vllm.transformers_utils.configs.qwen2vl import (Qwen2VLConfig,
+                                                     Qwen2VLVisionConfig)
 from vllm.transformers_utils.configs.solar import SolarConfig
 from vllm.transformers_utils.configs.ultravox import UltravoxConfig
 
@@ -32,7 +33,6 @@
     "NemotronConfig",
     "SolarConfig",
     "UltravoxConfig",
-    # Granite can be removed from here once we have upgraded to
-    # transformers 4.45+
-    "GraniteConfig",
+    "Qwen2VLConfig",
+    "Qwen2VLVisionConfig",
 ]