Switch to modular

Author: lkhl

src/transformers/models/videollama3/configuration_videollama3.py

@@ -1,12 +1,12 @@
-"""VideoLLaMA3 model configuration"""
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/videollama3/modular_videollama3.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_videollama3.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
 
 from ...configuration_utils import PretrainedConfig
-from ...utils import logging
-from ..auto import CONFIG_MAPPING, AutoConfig
-from ..qwen2 import Qwen2Config
-
-
-logger = logging.get_logger(__name__)
+from ..qwen2.configuration_qwen2 import Qwen2Config
 
 
 class Videollama3VisionConfig(PretrainedConfig):
@@ -31,19 +31,21 @@ class Videollama3VisionConfig(PretrainedConfig):
             The epsilon used by the layer normalization layers.
         attention_dropout (`float`, *optional*, defaults to 0.0):
             The dropout ratio for the attention probabilities.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
     """
 
     model_type = "videollama3_vision"
     base_config_key = "vision_config"
 
     def __init__(
         self,
-        hidden_size=1152,
-        intermediate_size=4304,
-        num_hidden_layers=27,
-        num_attention_heads=16,
+        hidden_size=768,
+        intermediate_size=3072,
+        num_hidden_layers=12,
+        num_attention_heads=12,
         num_channels=3,
-        patch_size=14,
+        patch_size=16,
         hidden_act="gelu_pytorch_tanh",
         layer_norm_eps=1e-6,
         attention_dropout=0.0,
@@ -58,9 +60,10 @@ def __init__(
         self.num_attention_heads = num_attention_heads
         self.num_channels = num_channels
         self.patch_size = patch_size
-        self.hidden_act = hidden_act
-        self.layer_norm_eps = layer_norm_eps
         self.attention_dropout = attention_dropout
+        self.layer_norm_eps = layer_norm_eps
+        self.hidden_act = hidden_act
+
         self.initializer_range = initializer_range
 
 
@@ -71,60 +74,42 @@ class Videollama3Config(PretrainedConfig):
             The config object or dictionary of the text backbone.
         vision_config (`Union[PreTrainedConfig, dict]`,  *optional*, defaults to `Videollama3VisionConfig`):
             The config object or dictionary of the vision backbone.
-        use_token_compression (`bool`, *optional*, defaults to `False`):
-            Whether to use temporal token compression to reduce the number of video tokens.
         image_token_id (`int`, *optional*, defaults to -1):
             The image token index to encode the image prompt.
         video_token_id (`int`, *optional*, defaults to -1):
             The video token index to encode the image prompt.
+        use_token_compression (`bool`, *optional*, defaults to `False`):
+            Whether to use temporal token compression to reduce the number of video tokens.
     """
 
     model_type = "videollama3"
-    sub_configs = {"vision_config": Videollama3VisionConfig, "text_config": AutoConfig}
+    sub_configs = {"vision_config": Videollama3VisionConfig, "text_config": Qwen2Config}
     keys_to_ignore_at_inference = ["past_key_values"]
 
     def __init__(
         self,
         text_config=None,
         vision_config=None,
+        image_token_id=151655,
+        video_token_id=151656,
         use_token_compression=False,
-        image_token_id=-1,
-        video_token_id=-1,
         **kwargs,
     ):
-        if text_config is None:
-            self.text_config = Qwen2Config(**kwargs)
-            logger.info("text_config is None, using default qwen2 config")
-        elif isinstance(text_config, dict):
-            assert "model_type" in text_config, "text_config must contain 'model_type' key"
-            self.text_config = CONFIG_MAPPING[text_config["model_type"]](**text_config)
-        elif isinstance(text_config, PretrainedConfig):
-            self.text_config = text_config
-        else:
-            raise ValueError(
-                "text_config must be a dictionary, PretrainedConfig instance, or None. "
-                f"Got {type(text_config)} instead."
-            )
+        super().__init__(**kwargs)
+        if isinstance(vision_config, dict):
+            self.vision_config = self.sub_configs["vision_config"](**vision_config)
+        elif vision_config is None:
+            self.vision_config = self.sub_configs["vision_config"]()
 
-        if vision_config is None:
-            self.vision_config = Videollama3VisionConfig()
-            logger.info("vision_config is None, using default vision config")
-        elif isinstance(vision_config, dict):
-            assert "model_type" in vision_config, "vision_config must contain 'model_type' key"
-            self.vision_config = CONFIG_MAPPING[vision_config["model_type"]](**vision_config)
-        elif isinstance(vision_config, PretrainedConfig):
-            self.vision_config = vision_config
-        else:
-            raise ValueError(
-                "vision_config must be a dictionary, PretrainedConfig instance, or None. "
-                f"Got {type(vision_config)} instead."
-            )
+        if isinstance(text_config, dict):
+            self.text_config = self.sub_configs["text_config"](**text_config)
+        elif text_config is None:
+            # For BC use all kwargs to init `TextConfig`
+            self.text_config = self.sub_configs["text_config"](**kwargs)
 
-        self.use_token_compression = use_token_compression
         self.image_token_id = image_token_id
         self.video_token_id = video_token_id
-
-        super().__init__(**kwargs)
+        self.use_token_compression = use_token_compression
 
 
-__all__ = ["Videollama3Config", "Videollama3VisionConfig"]
+__all__ = ["Videollama3VisionConfig", "Videollama3Config"]