[Model] Initialize support for InternVL2 series models (#6514)

Co-authored-by: Roger Wang <ywang@roblox.com>

vllm/transformers_utils/configs/__init__.py

@@ -4,6 +4,7 @@ from vllm.transformers_utils.configs.dbrx import DbrxConfig
 # 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.internvl import InternVLChatConfig
 from vllm.transformers_utils.configs.jais import JAISConfig
 from vllm.transformers_utils.configs.medusa import MedusaConfig
 from vllm.transformers_utils.configs.mlp_speculator import MLPSpeculatorConfig
@@ -15,6 +16,7 @@ __all__ = [
     "DbrxConfig",
     "MPTConfig",
     "RWConfig",
+    "InternVLChatConfig",
     "JAISConfig",
     "MedusaConfig",
     "MLPSpeculatorConfig",

vllm/transformers_utils/configs/internvl.py

@@ -0,0 +1,51 @@
+# Adapted from
+# https://huggingface.co/OpenGVLab/InternVL2-1B/blob/main/configuration_internvl_chat.py
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2024 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+from transformers.configuration_utils import PretrainedConfig
+
+
+class InternVLChatConfig(PretrainedConfig):
+    model_type = 'internvl_chat'
+    is_composition = True
+
+    def __init__(self,
+                 vision_config=None,
+                 llm_config=None,
+                 use_backbone_lora=0,
+                 use_llm_lora=0,
+                 select_layer=-1,
+                 force_image_size=None,
+                 downsample_ratio=0.5,
+                 template=None,
+                 dynamic_image_size=False,
+                 use_thumbnail=False,
+                 ps_version='v1',
+                 min_dynamic_patch=1,
+                 max_dynamic_patch=6,
+                 **kwargs):
+        super().__init__(**kwargs)
+
+        if vision_config is None:
+            vision_config = {}
+
+        if llm_config is None:
+            llm_config = {}
+
+        self.vision_config = PretrainedConfig(**vision_config)
+        self.text_config = PretrainedConfig(**llm_config)
+
+        self.use_backbone_lora = use_backbone_lora
+        self.use_llm_lora = use_llm_lora
+        self.select_layer = select_layer
+        self.force_image_size = force_image_size
+        self.downsample_ratio = downsample_ratio
+        self.template = template
+        self.dynamic_image_size = dynamic_image_size
+        self.use_thumbnail = use_thumbnail
+        self.ps_version = ps_version  # pixel shuffle version
+        self.min_dynamic_patch = min_dynamic_patch
+        self.max_dynamic_patch = max_dynamic_patch