Document check copies (#25291)

* Document check copies better and add tests

* Include header in check for copies

* Manual fixes

* Try autofix

* Fixes

* Clean tests

* Finalize doc

* Remove debug print

* More fixes

src/transformers/models/efficientformer/modeling_efficientformer.py

@@ -246,7 +246,7 @@ class EfficientFormerConvMlp(nn.Module):
 
 
 # Copied from transformers.models.convnext.modeling_convnext.drop_path
-def drop_path(input, drop_prob: float = 0.0, training: bool = False):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/esm/modeling_tf_esm.py

@@ -667,7 +667,7 @@ class TFEsmEncoder(Layer):
 
 
 # Copied from transformers.models.bert.modeling_tf_bert.TFBertPooler with Bert->Esm
-class TFEsmPooler(Layer):
+class TFEsmPooler(tf.keras.layers.Layer):
     def __init__(self, config: EsmConfig, **kwargs):
         super().__init__(**kwargs)
 

src/transformers/models/focalnet/modeling_focalnet.py

@@ -286,7 +286,7 @@ class FocalNetPatchEmbeddings(nn.Module):
 
 
 # Copied from transformers.models.beit.modeling_beit.drop_path
-def drop_path(input, drop_prob=0.0, training=False, scale_by_keep=True):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/glpn/modeling_glpn.py

@@ -52,8 +52,8 @@ GLPN_PRETRAINED_MODEL_ARCHIVE_LIST = [
 ]
 
 
-# Copied from transformers.models.segformer.modeling_segformer.drop_path
-def drop_path(input, drop_prob: float = 0.0, training: bool = False):
+# Copied from transformers.models.beit.modeling_beit.drop_path
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py

@@ -272,7 +272,7 @@ class GPTBigCodeMLP(nn.Module):
         self.dropout = nn.Dropout(config.resid_pdrop)
 
     # Copied from transformers.models.gpt2.modeling_gpt2.GPT2MLP.forward
-    def forward(self, hidden_states: Optional[Tuple[torch.Tensor]]) -> torch.Tensor:
+    def forward(self, hidden_states: Optional[Tuple[torch.FloatTensor]]) -> torch.FloatTensor:
         hidden_states = self.c_fc(hidden_states)
         hidden_states = self.act(hidden_states)
         hidden_states = self.c_proj(hidden_states)

src/transformers/models/llama/tokenization_llama.py

@@ -30,7 +30,8 @@ from ...utils import logging
 
 
 if TYPE_CHECKING:
-    from transformers.pipelines.conversational import Conversation
+    from ...pipelines.conversational import Conversation
+    from ...tokenization_utils_base import TextInput
 
 logger = logging.get_logger(__name__)
 
@@ -168,7 +169,7 @@ class LlamaTokenizer(PreTrainedTokenizer):
         return vocab
 
     # Copied from transformers.models.t5.tokenization_t5.T5Tokenizer.tokenize
-    def tokenize(self, text, **kwargs) -> List[str]:
+    def tokenize(self, text: "TextInput", **kwargs) -> List[str]:
         # Replace the SPIECE_UNDERLINE with a space to make sure SPIECE_UNDERLINE is only used at
         # the beginning of the text
         if not self.legacy:
@@ -176,7 +177,7 @@ class LlamaTokenizer(PreTrainedTokenizer):
         return super().tokenize(text, **kwargs)
 
     # Copied from transformers.models.t5.tokenization_t5.T5Tokenizer._tokenize
-    def _tokenize(self, text):
+    def _tokenize(self, text, **kwargs):
         """
         Returns a tokenized string.
 

src/transformers/models/longt5/modeling_flax_longt5.py

@@ -56,7 +56,7 @@ remat = nn_partitioning.remat
 
 
 # Copied from transformers.models.bart.modeling_flax_bart.shift_tokens_right
-def shift_tokens_right(input_ids: np.array, pad_token_id: int, decoder_start_token_id: int) -> np.ndarray:
+def shift_tokens_right(input_ids: jnp.array, pad_token_id: int, decoder_start_token_id: int) -> jnp.ndarray:
     """
     Shift input ids one token to the right.
     """

src/transformers/models/marian/modeling_flax_marian.py

@@ -227,7 +227,7 @@ def create_sinusoidal_positions(n_pos, dim):
 
 
 # Copied from transformers.models.bart.modeling_flax_bart.shift_tokens_right
-def shift_tokens_right(input_ids: jnp.ndarray, pad_token_id: int, decoder_start_token_id: int) -> jnp.ndarray:
+def shift_tokens_right(input_ids: jnp.array, pad_token_id: int, decoder_start_token_id: int) -> jnp.ndarray:
     """
     Shift input ids one token to the right.
     """

src/transformers/models/maskformer/modeling_maskformer_swin.py

@@ -123,7 +123,7 @@ def window_reverse(windows, window_size, height, width):
 
 
 # Copied from transformers.models.swin.modeling_swin.drop_path
-def drop_path(input, drop_prob=0.0, training=False, scale_by_keep=True):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/mgp_str/modeling_mgp_str.py

@@ -51,7 +51,7 @@ MGP_STR_PRETRAINED_MODEL_ARCHIVE_LIST = [
 
 
 # Copied from transformers.models.beit.modeling_beit.drop_path
-def drop_path(input, drop_prob: float = 0.0, training: bool = False):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/nat/modeling_nat.py

@@ -263,7 +263,7 @@ class NatDownsampler(nn.Module):
 
 
 # Copied from transformers.models.beit.modeling_beit.drop_path
-def drop_path(input, drop_prob=0.0, training=False, scale_by_keep=True):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/owlvit/image_processing_owlvit.py

@@ -47,7 +47,6 @@ if is_torch_available():
 logger = logging.get_logger(__name__)
 
 
-# Copied from transformers.models.detr.modeling_detr._upcast
 def _upcast(t):
     # Protects from numerical overflows in multiplications by upcasting to the equivalent higher type
     if t.is_floating_point():

src/transformers/models/owlvit/modeling_owlvit.py

@@ -22,7 +22,7 @@ from typing import Any, Dict, Optional, Tuple, Union
 import numpy as np
 import torch
 import torch.utils.checkpoint
-from torch import nn
+from torch import Tensor, nn
 
 from ...activations import ACT2FN
 from ...modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
@@ -120,7 +120,7 @@ class OwlViTOutput(ModelOutput):
 
 
 # Copied from transformers.models.detr.modeling_detr._upcast
-def _upcast(t: torch.Tensor) -> torch.Tensor:
+def _upcast(t: Tensor) -> Tensor:
     # Protects from numerical overflows in multiplications by upcasting to the equivalent higher type
     if t.is_floating_point():
         return t if t.dtype in (torch.float32, torch.float64) else t.float()
@@ -129,7 +129,7 @@ def _upcast(t: torch.Tensor) -> torch.Tensor:
 
 
 # Copied from transformers.models.detr.modeling_detr.box_area
-def box_area(boxes: torch.Tensor) -> torch.Tensor:
+def box_area(boxes: Tensor) -> Tensor:
     """
     Computes the area of a set of bounding boxes, which are specified by its (x1, y1, x2, y2) coordinates.
 
@@ -146,7 +146,7 @@ def box_area(boxes: torch.Tensor) -> torch.Tensor:
 
 
 # Copied from transformers.models.detr.modeling_detr.box_iou
-def box_iou(boxes1: torch.Tensor, boxes2: torch.Tensor) -> torch.Tensor:
+def box_iou(boxes1, boxes2):
     area1 = box_area(boxes1)
     area2 = box_area(boxes2)
 

src/transformers/models/pegasus/modeling_flax_pegasus.py

@@ -210,7 +210,7 @@ PEGASUS_DECODE_INPUTS_DOCSTRING = r"""
 
 
 # Copied from transformers.models.bart.modeling_flax_bart.shift_tokens_right
-def shift_tokens_right(input_ids: np.array, pad_token_id: int, decoder_start_token_id: int) -> np.ndarray:
+def shift_tokens_right(input_ids: jnp.array, pad_token_id: int, decoder_start_token_id: int) -> jnp.ndarray:
     """
     Shift input ids one token to the right.
     """
@@ -223,7 +223,7 @@ def shift_tokens_right(input_ids: np.array, pad_token_id: int, decoder_start_tok
 
 
 # Copied from transformers.models.marian.modeling_flax_marian.create_sinusoidal_positions
-def create_sinusoidal_positions(n_pos, dim, dtype):
+def create_sinusoidal_positions(n_pos, dim):
     position_enc = np.array([[pos / np.power(10000, 2 * (j // 2) / dim) for j in range(dim)] for pos in range(n_pos)])
     sentinel = dim // 2 + dim % 2
     out = np.zeros_like(position_enc)
@@ -686,9 +686,7 @@ class FlaxPegasusEncoder(nn.Module):
         self.max_source_positions = self.config.max_position_embeddings
         self.embed_scale = math.sqrt(embed_dim) if self.config.scale_embedding else 1.0
 
-        self.embed_positions = create_sinusoidal_positions(
-            self.config.max_position_embeddings, embed_dim, dtype=self.dtype
-        )
+        self.embed_positions = create_sinusoidal_positions(self.config.max_position_embeddings, embed_dim)
         self.layers = FlaxPegasusEncoderLayerCollection(self.config, self.dtype)
         self.layer_norm = nn.LayerNorm(dtype=self.dtype, epsilon=1e-05)
 
@@ -755,9 +753,7 @@ class FlaxPegasusDecoder(nn.Module):
         self.max_target_positions = self.config.max_position_embeddings
         self.embed_scale = math.sqrt(self.config.d_model) if self.config.scale_embedding else 1.0
 
-        self.embed_positions = create_sinusoidal_positions(
-            self.config.max_position_embeddings, embed_dim, dtype=self.dtype
-        )
+        self.embed_positions = create_sinusoidal_positions(self.config.max_position_embeddings, embed_dim)
 
         self.layers = FlaxPegasusDecoderLayerCollection(self.config, self.dtype)
         self.layer_norm = nn.LayerNorm(dtype=self.dtype, epsilon=1e-05)

src/transformers/models/poolformer/modeling_poolformer.py

@@ -50,7 +50,7 @@ POOLFORMER_PRETRAINED_MODEL_ARCHIVE_LIST = [
 
 
 # Copied from transformers.models.beit.modeling_beit.drop_path
-def drop_path(input, drop_prob: float = 0.0, training: bool = False):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/pvt/modeling_pvt.py

@@ -55,8 +55,8 @@ PVT_PRETRAINED_MODEL_ARCHIVE_LIST = [
 ]
 
 
-# Copied from transformers.models.convnext.modeling_convnext.drop_path
-def drop_path(input, drop_prob: float = 0.0, training: bool = False, scale_by_keep=True):
+# Copied from transformers.models.beit.modeling_beit.drop_path
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/segformer/modeling_segformer.py

@@ -84,8 +84,8 @@ class SegFormerImageClassifierOutput(ImageClassifierOutput):
     attentions: Optional[Tuple[torch.FloatTensor]] = None
 
 
-# Copied from transformers.models.convnext.modeling_convnext.drop_path
-def drop_path(input, drop_prob: float = 0.0, training: bool = False, scale_by_keep=True):
+# Copied from transformers.models.beit.modeling_beit.drop_path
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/swiftformer/modeling_swiftformer.py

@@ -86,7 +86,7 @@ class SwiftFormerPatchEmbedding(nn.Module):
 
 
 # Copied from transformers.models.beit.modeling_beit.drop_path
-def drop_path(x, drop_prob: float = 0.0, training: bool = False):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/swin/modeling_swin.py

@@ -380,7 +380,7 @@ class SwinPatchMerging(nn.Module):
 
 
 # Copied from transformers.models.beit.modeling_beit.drop_path
-def drop_path(input, drop_prob=0.0, training=False, scale_by_keep=True):
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
 

src/transformers/models/swin2sr/modeling_swin2sr.py

@@ -105,8 +105,8 @@ def window_reverse(windows, window_size, height, width):
     return windows
 
 
-# Copied from transformers.models.swin.modeling_swin.drop_path
-def drop_path(input, drop_prob=0.0, training=False, scale_by_keep=True):
+# Copied from transformers.models.beit.modeling_beit.drop_path
+def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = False) -> torch.Tensor:
     """
     Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).