Add more missing arguments

src/transformers/models/bamba/modular_bamba.py

@@ -708,7 +708,7 @@ class BambaRMSNorm(LlamaRMSNorm):
 
 class BambaDecoderLayer(JambaAttentionDecoderLayer):
     def __init__(self, config: BambaConfig, layer_idx: int, layer_type: str = "mamba"):
-        super().__init__()
+        super().__init__(config, layer_idx)
 
         del self.self_attn
 

src/transformers/models/cohere2/modular_cohere2.py

@@ -271,7 +271,7 @@ class Cohere2LayerNorm(CohereLayerNorm):
     pass
 
 
-class Cohere2Attention(CohereAttention, nn.Module):
+class Cohere2Attention(CohereAttention):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
     def __init__(self, config: Cohere2Config, layer_idx: Optional[int] = None):

src/transformers/models/d_fine/modular_d_fine.py

@@ -898,7 +898,7 @@ def __init__(self, config: DFineConfig):
 
 class DFineForObjectDetection(RTDetrForObjectDetection, DFinePreTrainedModel):
     def __init__(self, config: DFineConfig):
-        DFinePreTrainedModel.__init__(config)
+        DFinePreTrainedModel.__init__(self, config)
 
         # D-FINE encoder-decoder model
         self.eval_idx = config.eval_idx if config.eval_idx >= 0 else config.decoder_layers + config.eval_idx

src/transformers/models/data2vec/modular_data2vec_audio.py

@@ -112,7 +112,7 @@ def forward(self, hidden_states):
         return hidden_states
 
 
-class Data2VecAudioFeatureEncoder(Wav2Vec2FeatureEncoder, nn.Module):
+class Data2VecAudioFeatureEncoder(Wav2Vec2FeatureEncoder):
     def __init__(self, config):
         nn.Module.__init__(self)
         self.conv_layers = nn.ModuleList(
@@ -183,7 +183,7 @@ def load_adapter(self):
 
 class Data2VecAudioModel(Data2VecAudioPreTrainedModel, Wav2Vec2Model):
     def __init__(self, config: Data2VecAudioConfig):
-        Data2VecAudioPreTrainedModel.__init__(config)
+        Data2VecAudioPreTrainedModel.__init__(self, config)
         self.config = config
         self.feature_extractor = Data2VecAudioFeatureEncoder(config)
         self.feature_projection = Data2VecAudioFeatureProjection(config)
@@ -215,7 +215,7 @@ def forward(self, **super_kwargs):
 
 class Data2VecAudioForCTC(Data2VecAudioPreTrainedModel, Wav2Vec2ForCTC):
     def __init__(self, config):
-        Data2VecAudioPreTrainedModel.__init__(config)
+        Data2VecAudioPreTrainedModel.__init__(self, config)
 
         self.data2vec_audio = Data2VecAudioModel(config)
         self.dropout = nn.Dropout(config.final_dropout)

src/transformers/models/dia/modular_dia.py

@@ -107,7 +107,7 @@ class DiaRotaryEmbedding(LlamaRotaryEmbedding):
     pass
 
 
-class DiaSelfAttention(LlamaAttention, nn.Module):
+class DiaSelfAttention(LlamaAttention):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
     def __init__(self, config: Union[DiaEncoderConfig, DiaDecoderConfig], layer_idx: int, is_causal: bool = False):

src/transformers/models/dots1/modular_dots1.py

@@ -62,7 +62,7 @@ class Dots1TopkRouter(DeepseekV3TopkRouter):
 
 class Dots1DecoderLayer(DeepseekV3DecoderLayer):
     def __init__(self, config: Dots1Config, layer_idx: int):
-        super().__init__()
+        super().__init__(config, layer_idx)
         self.attention_type = config.layer_types[layer_idx]
 
 

src/transformers/models/ernie4_5/modular_ernie4_5.py

@@ -84,7 +84,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
 
 class Ernie4_5MLP(LlamaMLP):
     def __init__(self, config: Ernie4_5Config):
-        super().__init__()
+        super().__init__(config)
 
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.use_bias)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.use_bias)

src/transformers/models/esm/openfold_utils/residue_constants.py

@@ -541,7 +541,6 @@ def make_bond_key(atom1_name: str, atom2_name: str) -> str:
 
 # A compact atom encoding with 14 columns
 # pylint: disable=line-too-long
-# pylint: disable=bad-whitespace
 restype_name_to_atom14_names: dict[str, list[str]] = {
     "ALA": ["N", "CA", "C", "O", "CB", "", "", "", "", "", "", "", "", ""],
     "ARG": ["N", "CA", "C", "O", "CB", "CG", "CD", "NE", "CZ", "NH1", "NH2", "", "", ""],
@@ -566,7 +565,6 @@ def make_bond_key(atom1_name: str, atom2_name: str) -> str:
     "UNK": ["", "", "", "", "", "", "", "", "", "", "", "", "", ""],
 }
 # pylint: enable=line-too-long
-# pylint: enable=bad-whitespace
 
 
 # This is the standard residue order when coding AA type as a number.

src/transformers/models/evolla/modular_evolla.py

@@ -65,7 +65,7 @@
 
 class EvollaSaProtEmbeddings(EsmEmbeddings):
     def __init__(self, config):
-        super().__init__()
+        super().__init__(config)
         # remove the position_ids in EsmEmbeddings
         self.position_ids = None
 
@@ -127,7 +127,7 @@ def forward(self, q: torch.Tensor, k: torch.Tensor) -> tuple[torch.Tensor, torch
         )
 
 
-class EvollaSaProtSelfAttention(EsmSelfAttention, nn.Module):
+class EvollaSaProtSelfAttention(EsmSelfAttention):
     def __init__(self, config, position_embedding_type=None, layer_idx=None, is_cross_attention=False):
         nn.Module.__init__(self)
         self.config = config

src/transformers/models/falcon_h1/modeling_falcon_h1.py

@@ -1007,7 +1007,7 @@ def forward(
 
 
 class FalconH1MLP(nn.Module):
-    def __init__(self, config: FalconH1Config = None):
+    def __init__(self, config: FalconH1Config):
         super().__init__()
         self.config = config
         self.hidden_size = config.hidden_size

src/transformers/models/falcon_h1/modular_falcon_h1.py

@@ -252,7 +252,7 @@ def forward(
 
 class FalconH1RMSNormGated(MambaRMSNormGated):
     def __init__(self, hidden_size, eps=1e-6, n_groups=1, norm_before_gate=True):
-        super().__init__()
+        super().__init__(hidden_size=hidden_size, eps=eps)
         self.weight = nn.Parameter(torch.ones(hidden_size))
         self.variance_epsilon = eps
         self.n_groups = n_groups
@@ -812,8 +812,8 @@ def forward(
 
 
 class FalconH1MLP(LlamaMLP):
-    def __init__(self, config: FalconH1Config = None):
-        super().__init__()
+    def __init__(self, config: FalconH1Config):
+        super().__init__(config)
         self.gate_multiplier, self.down_multiplier = config.mlp_multipliers
 
     def forward(self, x):

src/transformers/models/falcon_mamba/modular_falcon_mamba.py

@@ -517,7 +517,7 @@ class FalconMambaCausalLMOutput(MambaCausalLMOutput):
 
 class FalconMambaModel(MambaModel, FalconMambaPreTrainedModel):
     def __init__(self, config):
-        FalconMambaPreTrainedModel.__init__(config)
+        FalconMambaPreTrainedModel.__init__(self, config)
 
         self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size)
         self.layers = nn.ModuleList(

src/transformers/models/florence2/modular_florence2.py

@@ -1065,7 +1065,7 @@ def forward(self, seq_embeds: torch.Tensor) -> torch.Tensor:
 
 class Florence2VisionMLP(Llama4VisionMLP):
     def __init__(self, config: Florence2VisionConfig, stage_idx: int):
-        super().__init__()
+        super().__init__(config)
         self.fc1 = nn.Linear(config.embed_dim[stage_idx], int(config.embed_dim[stage_idx] * config.mlp_ratio))
         self.activation_fn = ACT2FN[config.activation_function]
         self.fc2 = nn.Linear(int(config.embed_dim[stage_idx] * config.mlp_ratio), config.embed_dim[stage_idx])

src/transformers/models/gemma/modular_gemma.py

@@ -360,7 +360,7 @@ def extra_repr(self):
 
 class GemmaMLP(LlamaMLP):
     def __init__(self, config):
-        super().__init__()
+        super().__init__(config)
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)

src/transformers/models/gemma2/modular_gemma2.py

@@ -209,7 +209,7 @@ class Gemma2RMSNorm(GemmaRMSNorm):
 
 class Gemma2MLP(GemmaMLP):
     def __init__(self, config):
-        super().__init__()
+        super().__init__(config)
         self.act_fn = ACT2FN[config.hidden_activation]
 
 

src/transformers/models/gemma3/modular_gemma3.py

@@ -383,7 +383,7 @@ def __init__(self, config: Gemma3TextConfig):
 
 class Gemma3RMSNorm(Gemma2RMSNorm):
     def __init__(self, dim: int, eps: float = 1e-6):
-        super().__init__()
+        super().__init__(dim=dim, eps=eps)
 
 
 class Gemma3RotaryEmbedding(Gemma2RotaryEmbedding):
@@ -396,7 +396,7 @@ class Gemma3Attention(Gemma2Attention):
     def __init__(self, config: Gemma3TextConfig, layer_idx: int):
         self.is_sliding = config.layer_types[layer_idx] == "sliding_attention"
 
-        super().__init__()
+        super().__init__(config, layer_idx)
         self.sliding_window = config.sliding_window if self.is_sliding else None
 
         self.q_norm = Gemma3RMSNorm(dim=config.head_dim, eps=config.rms_norm_eps)

src/transformers/models/gemma3n/modular_gemma3n.py

@@ -1739,7 +1739,7 @@ def apply_rotary_pos_emb(
 
 class Gemma3nTextAttention(Gemma3Attention):
     def __init__(self, config: Gemma3nTextConfig, layer_idx: int):
-        super().__init__()
+        super().__init__(config, layer_idx)
         del self.attn_logit_softcapping
         del self.scaling
         self.v_norm = Gemma3nRMSNorm(dim=config.head_dim, eps=config.rms_norm_eps, with_scale=False)
@@ -2234,7 +2234,7 @@ class Gemma3nModel(PaliGemmaModel):
     _checkpoint_conversion_mapping = {}
 
     def __init__(self, config: Gemma3nConfig):
-        super().__init__()
+        super().__init__(config)
         del self.multi_modal_projector  # Replaced by Gemma3nVisionEmbedder
         self.vocab_size_per_layer_input = config.text_config.vocab_size_per_layer_input
         self.audio_tower = AutoModel.from_config(config.audio_config)

src/transformers/models/glm4_moe/modular_glm4_moe.py

@@ -255,7 +255,7 @@ def __init__(
         )
 
 
-class Glm4MoeAttention(CohereAttention, nn.Module):
+class Glm4MoeAttention(CohereAttention):
     def __init__(self, config: Glm4MoeConfig, layer_idx: Optional[int] = None):
         nn.Module.__init__(self)
         self.config = config
@@ -287,7 +287,7 @@ class Glm4MoeMLP(DeepseekV3MLP):
     pass
 
 
-class Glm4MoeTopkRouter(DeepseekV3TopkRouter, nn.Module):
+class Glm4MoeTopkRouter(DeepseekV3TopkRouter):
     def __init__(self, config: Glm4MoeConfig):
         nn.Module.__init__(self)
         self.config = config

src/transformers/models/glm4v/modular_glm4v.py

@@ -507,15 +507,15 @@ def forward(self, embeddings, lengths, image_shapes, h_coords, w_coords) -> torc
 
 class Glm4vVisionAttention(Qwen2_5_VLVisionAttention):
     def __init__(self, config: Glm4vVisionConfig) -> None:
-        super().__init__()
+        super().__init__(config)
         self.attention_dropout = config.attention_dropout
         self.qkv = nn.Linear(config.hidden_size, config.hidden_size * 3, bias=config.attention_bias)
         self.proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
 
 
 class Glm4vVisionBlock(Qwen2_5_VLVisionBlock):
     def __init__(self, config) -> None:
-        super().__init__()
+        super().__init__(config)
         self.norm1 = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.norm2 = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.attn = Glm4vVisionAttention(config)

src/transformers/models/got_ocr2/modular_got_ocr2.py

@@ -240,7 +240,7 @@ class GotOcr2VisionAttention(SamVisionAttention):
 
 class GotOcr2VisionLayer(SamVisionLayer):
     def __init__(self, config, window_size):
-        super().__init__()
+        super().__init__(config, window_size)
         self.layer_norm1 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
         self.attn = GotOcr2VisionAttention(config, window_size)
         self.layer_norm2 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

src/transformers/models/helium/modular_helium.py

@@ -104,7 +104,7 @@ def __init__(self, config: HeliumConfig, layer_idx: Optional[int] = None):
 
 class HeliumDecoderLayer(LlamaDecoderLayer):
     def __init__(self, config: HeliumConfig, layer_idx: Optional[int] = None):
-        super().__init__()
+        super().__init__(config, layer_idx)
 
         self.mlp = HeliumMLP(config)
         self.input_layernorm = HeliumRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

src/transformers/models/hunyuan_v1_dense/modular_hunyuan_v1_dense.py

@@ -114,7 +114,7 @@ def forward(
 
 class HunYuanDenseV1DecoderLayer(LlamaDecoderLayer):
     def __init__(self, config: HunYuanDenseV1Config, layer_idx: int):
-        super().__init__()
+        super().__init__(config, layer_idx)
         self.layer_idx = layer_idx
 
 

src/transformers/models/hunyuan_v1_moe/modular_hunyuan_v1_moe.py

@@ -187,7 +187,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
 
 class HunYuanMoEV1DecoderLayer(LlamaDecoderLayer):
     def __init__(self, config: HunYuanMoEV1Config, layer_idx: int):
-        super().__init__()
+        super().__init__(config, layer_idx)
         self.hidden_size = config.hidden_size
         self.self_attn = HunYuanMoEV1Attention(config=config, layer_idx=layer_idx)
         self.mlp = HunYuanMoEV1Moe(config, layer_idx=layer_idx)

src/transformers/models/internvl/modular_internvl.py

@@ -79,7 +79,7 @@ class InternVLVisionRMSNorm(LlamaRMSNorm):
 
 class InternVLVisionAttention(JanusVisionAttention):
     def __init__(self, config: InternVLVisionConfig):
-        super().__init__()
+        super().__init__(config)
         del self.num_key_value_groups
 
         # Needed for flash attention

src/transformers/models/janus/modular_janus.py

@@ -536,7 +536,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
 
 class JanusVisionEncoderLayer(SiglipEncoderLayer):
     def __init__(self, config: JanusVisionConfig):
-        super().__init__()
+        super().__init__(config)
         self.config = config
         self.embed_dim = config.hidden_size
         self.self_attn = JanusVisionAttention(config)

src/transformers/models/mistral/modular_mistral.py

@@ -44,7 +44,7 @@ def __init__(self, config):
 
 class MistralAttention(LlamaAttention):
     def __init__(self, config: MistralConfig, layer_idx: int):
-        super().__init__()
+        super().__init__(config, layer_idx)
         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)

src/transformers/models/mm_grounding_dino/modular_mm_grounding_dino.py

@@ -337,7 +337,7 @@ class MMGroundingDinoDecoder(GroundingDinoDecoder):
 
 class MMGroundingDinoModel(GroundingDinoModel, MMGroundingDinoPreTrainedModel):
     def __init__(self, config: MMGroundingDinoConfig):
-        MMGroundingDinoPreTrainedModel.__init__(config)
+        MMGroundingDinoPreTrainedModel.__init__(self, config)
 
         # Create backbone + positional encoding
         backbone = MMGroundingDinoConvEncoder(config)
@@ -400,7 +400,7 @@ class MMGroundingDinoForObjectDetection(GroundingDinoForObjectDetection, MMGroun
     ]
 
     def __init__(self, config: MMGroundingDinoConfig):
-        MMGroundingDinoPreTrainedModel.__init__(config)
+        MMGroundingDinoPreTrainedModel.__init__(self, config)
 
         self.model = MMGroundingDinoModel(config)
 

src/transformers/models/ovis2/modular_ovis2.py

@@ -60,7 +60,7 @@ class Ovis2VisionMLP(LlamaMLP):
 
 class Ovis2VisionEmbeddings(SiglipVisionEmbeddings):
     def __init__(self, config: Ovis2VisionConfig):
-        super().__init__()
+        super().__init__(config)
         self.rms_norm = Ovis2RMSNorm(config.hidden_size, config.rms_norm_eps)
 
     def interpolate_pos_encoding(self):
@@ -87,7 +87,7 @@ class Ovis2VisionEncoderLayer(Aimv2EncoderLayer):
 
 class Ovis2VisionEncoder(SiglipEncoder):
     def __init__(self, config: Ovis2VisionConfig):
-        super().__init__()
+        super().__init__(config)
         self.layers = nn.ModuleList([Ovis2VisionEncoderLayer(config) for _ in range(config.num_hidden_layers)])
 
 

src/transformers/models/phi4_multimodal/modular_phi4_multimodal.py

@@ -528,7 +528,7 @@ def __init__(self, config: Phi4MultimodalVisionConfig):
 
 class Phi4MultimodalVisionEncoder(SiglipEncoder):
     def __init__(self, config: Phi4MultimodalVisionConfig):
-        super().__init__()
+        super().__init__(config)
         self.layers = nn.ModuleList(
             [Phi4MultimodalVisionEncoderLayer(config) for _ in range(config.num_hidden_layers)]
         )
@@ -582,7 +582,7 @@ def _init_weights(self, module):
             module.weight.data.fill_(1.0)
 
 
-class Phi4MultimodalVisionEmbeddings(SiglipVisionEmbeddings, nn.Module):
+class Phi4MultimodalVisionEmbeddings(SiglipVisionEmbeddings):
     def __init__(self, config: Phi4MultimodalVisionConfig):
         nn.Module.__init__(self)
         self.config = config
@@ -1455,7 +1455,7 @@ def _init_weights(self, module):
             module.sub_img_feature_extensor.data.zero_()
 
 
-class Phi4MultimodalModel(Phi3Model, nn.Module):
+class Phi4MultimodalModel(Phi3Model):
     def __init__(self, config: Phi4MultimodalConfig):
         super().__init__(config)
         self.padding_idx = config.pad_token_id
@@ -1570,7 +1570,7 @@ def forward(
         )
 
 
-class Phi4MultimodalForCausalLM(Phi3ForCausalLM, nn.Module):
+class Phi4MultimodalForCausalLM(Phi3ForCausalLM):
     _tied_weights_keys = ["lm_head.weight"]
 
     def __init__(self, config):

src/transformers/models/qwen2_5_omni/modular_qwen2_5_omni.py

@@ -2061,7 +2061,7 @@ def __init__(self, config: Qwen2_5OmniThinkerConfig, device=None):
 
 # It's same as `Qwen2_5_VLAttention`, but talker model's hidden_size isn't divisible by num_heads.
 # Removes the value error as a workaround.
-class Qwen2_5OmniAttention(Qwen2_5_VLAttention, nn.Module):
+class Qwen2_5OmniAttention(Qwen2_5_VLAttention):
     def __init__(self, config: Qwen2_5OmniConfig, layer_idx: Optional[int] = None):
         nn.Module.__init__(self)
         self.config = config