Update composition flag usage (#36263)

* update composition flag usage

* remove print

* fix tests

* actually fix

* oh c'mon

* now should be fixed right?

* fix copies

src/transformers/cache_utils.py

@@ -1195,9 +1195,7 @@ class StaticCache(Cache):
         self.max_cache_len = config.max_position_embeddings if max_cache_len is None else max_cache_len
 
         # Some model define a custom `head_dim` != config.hidden_size // config.num_attention_heads
-        self.head_dim = (
+        self.head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
-            config.head_dim if hasattr(config, "head_dim") else config.hidden_size // config.num_attention_heads
-        )
 
         self._dtype = dtype
         self.num_key_value_heads = (

src/transformers/configuration_utils.py

@@ -61,9 +61,10 @@ class PretrainedConfig(PushToHubMixin):
 
     - **model_type** (`str`) -- An identifier for the model type, serialized into the JSON file, and used to recreate
       the correct object in [`~transformers.AutoConfig`].
-    - **is_composition** (`bool`) -- Whether the config class is composed of multiple sub-configs. In this case the
+    - **has_no_defaults_at_init** (`bool`) -- Whether the config class can be initialized without providing input arguments.
-      config has to be initialized from two or more configs of type [`~transformers.PretrainedConfig`] like:
+      Some configurations requires inputs to be defined at init and have no default values, usually these are composite configs,
-      [`~transformers.EncoderDecoderConfig`] or [`~RagConfig`].
+      (but not necessarily) such as [`~transformers.EncoderDecoderConfig`] or [`~RagConfig`]. They have to be initialized from
+      two or more configs of type [`~transformers.PretrainedConfig`].
     - **keys_to_ignore_at_inference** (`List[str]`) -- A list of keys to ignore by default when looking at dictionary
       outputs of the model during inference.
     - **attribute_map** (`Dict[str, str]`) -- A dict that maps model specific attribute names to the standardized
@@ -193,7 +194,7 @@ class PretrainedConfig(PushToHubMixin):
     model_type: str = ""
     base_config_key: str = ""
     sub_configs: dict[str, "PretrainedConfig"] = {}
-    is_composition: bool = False
+    has_no_defaults_at_init: bool = False
     attribute_map: dict[str, str] = {}
     base_model_tp_plan: Optional[dict[str, Any]] = None
     base_model_pp_plan: Optional[dict[str, tuple[list[str]]]] = None
@@ -813,8 +814,8 @@ class PretrainedConfig(PushToHubMixin):
         # Get the default config dict (from a fresh PreTrainedConfig instance)
         default_config_dict = PretrainedConfig().to_dict()
 
-        # Get class-specific config dict if not part of a composition
+        # get class specific config dict
-        class_config_dict = self.__class__().to_dict() if not self.is_composition else {}
+        class_config_dict = self.__class__().to_dict() if not self.has_no_defaults_at_init else {}
 
         serializable_config_dict = {}
 

src/transformers/modeling_rope_utils.py

@@ -121,7 +121,7 @@ def _compute_default_rope_parameters(
     elif config is not None:
         base = config.rope_theta
         partial_rotary_factor = config.partial_rotary_factor if hasattr(config, "partial_rotary_factor") else 1.0
-        head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
         dim = int(head_dim * partial_rotary_factor)
 
     attention_factor = 1.0  # Unused in this type of RoPE

src/transformers/models/bark/generation_configuration_bark.py

@@ -240,7 +240,6 @@ class BarkFineGenerationConfig(GenerationConfig):
 
 class BarkGenerationConfig(GenerationConfig):
     model_type = "bark"
-    is_composition = True
 
     # TODO (joao): nested from_dict
 

src/transformers/models/encoder_decoder/configuration_encoder_decoder.py

@@ -72,7 +72,7 @@ class EncoderDecoderConfig(PretrainedConfig):
 
     model_type = "encoder-decoder"
     sub_configs = {"encoder": AutoConfig, "decoder": AutoConfig}
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)

src/transformers/models/llava/configuration_llava.py

@@ -76,7 +76,6 @@ class LlavaConfig(PretrainedConfig):
 
     model_type = "llava"
     sub_configs = {"text_config": AutoConfig, "vision_config": AutoConfig}
-    is_composition = True
 
     def __init__(
         self,

src/transformers/models/mistral/configuration_mistral.py

@@ -143,7 +143,7 @@ class MistralConfig(PretrainedConfig):
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
         self.sliding_window = sliding_window
-        self.head_dim = head_dim or hidden_size // num_attention_heads
+        self.head_dim = head_dim
 
         # for backward compatibility
         if num_key_value_heads is None:

src/transformers/models/mistral/modeling_mistral.py

@@ -139,7 +139,7 @@ class MistralAttention(nn.Module):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
-        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
         self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
         self.scaling = self.head_dim**-0.5
         self.attention_dropout = config.attention_dropout

src/transformers/models/mistral/modular_mistral.py

@@ -42,6 +42,7 @@ class MistralMLP(LlamaMLP):
 class MistralAttention(LlamaAttention):
     def __init__(self, config: MistralConfig, layer_idx: int):
         super().__init__()
+        self.head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
         self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=False)
         self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
         self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)

src/transformers/models/mixtral/configuration_mixtral.py

@@ -172,7 +172,7 @@ class MixtralConfig(PretrainedConfig):
         self.use_cache = use_cache
         self.rope_theta = rope_theta
         self.attention_dropout = attention_dropout
-        self.head_dim = head_dim if head_dim is not None else self.hidden_size // self.num_attention_heads
+        self.head_dim = head_dim
 
         self.num_experts_per_tok = num_experts_per_tok
         self.num_local_experts = num_local_experts

src/transformers/models/mixtral/modeling_mixtral.py

@@ -251,7 +251,7 @@ class MixtralAttention(nn.Module):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
-        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
         self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
         self.scaling = self.head_dim**-0.5
         self.attention_dropout = config.attention_dropout

src/transformers/models/musicgen/configuration_musicgen.py

@@ -195,7 +195,7 @@ class MusicgenConfig(PretrainedConfig):
         "audio_encoder": AutoConfig,
         "decoder": MusicgenDecoderConfig,
     }
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)

src/transformers/models/musicgen_melody/configuration_musicgen_melody.py

@@ -201,7 +201,7 @@ class MusicgenMelodyConfig(PretrainedConfig):
         "audio_encoder": AutoConfig,
         "decoder": MusicgenMelodyDecoderConfig,
     }
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(
         self,

src/transformers/models/pixtral/convert_pixtral_weights_to_hf.py

@@ -158,7 +158,6 @@ def convert_mistral_model(input_dir, output_dir):
             hidden_act="silu",
             sliding_window=None,
             tie_word_embeddings=False,
-            is_composition=True,
             rms_norm_eps=1e-5,
         )
     else:

src/transformers/models/rag/configuration_rag.py

@@ -79,7 +79,7 @@ RAG_CONFIG_DOC = r"""
 @add_start_docstrings(RAG_CONFIG_DOC)
 class RagConfig(PretrainedConfig):
     model_type = "rag"
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(
         self,

src/transformers/models/speech_encoder_decoder/configuration_speech_encoder_decoder.py

@@ -72,7 +72,7 @@ class SpeechEncoderDecoderConfig(PretrainedConfig):
 
     model_type = "speech-encoder-decoder"
     sub_configs = {"encoder": AutoConfig, "decoder": AutoConfig}
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)

src/transformers/models/starcoder2/modeling_starcoder2.py

@@ -158,7 +158,7 @@ class Starcoder2Attention(nn.Module):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
-        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
         self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
         self.scaling = self.head_dim**-0.5
         self.attention_dropout = config.attention_dropout

src/transformers/models/vision_encoder_decoder/configuration_vision_encoder_decoder.py

@@ -79,7 +79,7 @@ class VisionEncoderDecoderConfig(PretrainedConfig):
 
     model_type = "vision-encoder-decoder"
     sub_configs = {"encoder": AutoConfig, "decoder": AutoConfig}
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)

src/transformers/models/vision_text_dual_encoder/configuration_vision_text_dual_encoder.py

@@ -76,7 +76,7 @@ class VisionTextDualEncoderConfig(PretrainedConfig):
 
     model_type = "vision-text-dual-encoder"
     sub_configs = {"vision_config": AutoConfig, "text_config": AutoConfig}
-    is_composition = True
+    has_no_defaults_at_init = True
 
     def __init__(self, projection_dim=512, logit_scale_init_value=2.6592, **kwargs):
         super().__init__(**kwargs)

tests/generation/test_utils.py

@@ -1755,9 +1755,7 @@ class GenerationTesterMixin:
 
             text_config = model.config.get_text_config()
             head_dim = (
-                text_config.head_dim
+                getattr(text_config, "head_dim", None) or text_config.hidden_size // text_config.num_attention_heads
-                if hasattr(text_config, "head_dim")
-                else text_config.hidden_size // text_config.num_attention_heads
             )
             num_key_value_heads = (
                 text_config.num_attention_heads
@@ -2008,9 +2006,8 @@ class GenerationTesterMixin:
                 max_cache_len = seq_length + max_new_tokens - 1  # cache len = gen len - 1, the last token has no cache
                 text_config = config.text_config if hasattr(config, "text_config") else config
                 head_dim = (
-                    text_config.head_dim
+                    getattr(text_config, "head_dim", None)
-                    if hasattr(text_config, "head_dim")
+                    or text_config.hidden_size // text_config.num_attention_heads
-                    else text_config.hidden_size // text_config.num_attention_heads
                 )
                 num_key_value_heads = (
                     text_config.num_attention_heads

tests/models/llava/test_modeling_llava.py

@@ -184,13 +184,6 @@ class LlavaForConditionalGenerationModelTest(ModelTesterMixin, GenerationTesterM
         )
 
     def test_config(self):
-        # overwritten from `tests/test_configuration_common.py::ConfigTester` after #36077
-        # TODO: avoid overwritten once there is a better fix for #36077
-        def check_config_can_be_init_without_params():
-            config = self.config_tester.config_class()
-            self.config_tester.parent.assertIsNotNone(config)
-
-        self.config_tester.check_config_can_be_init_without_params = check_config_can_be_init_without_params
         self.config_tester.run_common_tests()
 
     # overwrite inputs_embeds tests because we need to delete "pixel values" for LVLMs

tests/test_configuration_common.py

@@ -163,7 +163,7 @@ class ConfigTester:
         self.parent.assertEqual(len(config.label2id), 3)
 
     def check_config_can_be_init_without_params(self):
-        if self.config_class.is_composition:
+        if self.config_class.has_no_defaults_at_init:
             with self.parent.assertRaises(ValueError):
                 config = self.config_class()
         else: