SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Use Python 3.9 syntax in examples (#37279)

Browse Source 
Signed-off-by: cyy <cyyever@outlook.com>
This commit is contained in:
cyyever
2025-04-07 19:52:21 +08:00
committed by GitHub
parent 08f36771b3
commit 0fb8d49e88
123 changed files with 358 additions and 451 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
examples/modular-transformers/image_processing_new_imgproc_model.py
View File

@@ -4,7 +4,7 @@
# the file from the modular. If any change should be done, please apply the change to the
# modular_new_imgproc_model.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from typing import Dict, List, Optional, Union
from typing import Optional, Union
import numpy as np
import torch
@@ -74,13 +74,13 @@ class ImgprocModelImageProcessor(BaseImageProcessor):
def __init__(
self,
do_resize: bool = True,
size: Dict[str, int] = None,
size: dict[str, int] = None,
resample: PILImageResampling = PILImageResampling.BICUBIC,
do_rescale: bool = True,
rescale_factor: Union[int, float] = 1 / 255,
do_normalize: bool = True,
image_mean: Optional[Union[float, List[float]]] = None,
image_std: Optional[Union[float, List[float]]] = None,
image_mean: Optional[Union[float, list[float]]] = None,
image_std: Optional[Union[float, list[float]]] = None,
do_convert_rgb: bool = True,
**kwargs,
) -> None:
@@ -101,7 +101,7 @@ class ImgprocModelImageProcessor(BaseImageProcessor):
def resize(
self,
image: np.ndarray,
size: Dict[str, int],
size: dict[str, int],
resample: PILImageResampling = PILImageResampling.BICUBIC,
data_format: Optional[Union[str, ChannelDimension]] = None,
input_data_format: Optional[Union[str, ChannelDimension]] = None,
@@ -151,13 +151,13 @@ class ImgprocModelImageProcessor(BaseImageProcessor):
self,
images: ImageInput,
do_resize: Optional[bool] = None,
size: Optional[Dict[str, int]] = None,
size: Optional[dict[str, int]] = None,
resample: PILImageResampling = None,
do_rescale: Optional[bool] = None,
rescale_factor: Optional[float] = None,
do_normalize: Optional[bool] = None,
image_mean: Optional[Union[float, List[float]]] = None,
image_std: Optional[Union[float, List[float]]] = None,
image_mean: Optional[Union[float, list[float]]] = None,
image_std: Optional[Union[float, list[float]]] = None,
return_tensors: Optional[Union[str, TensorType]] = None,
do_convert_rgb: bool = None,
data_format: ChannelDimension = ChannelDimension.FIRST,

4
examples/modular-transformers/modeling_add_function.py
View File

@@ -5,7 +5,7 @@
# modular_add_function.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Note that zamba does not have the `apply_rotary_pos_emb` function!
from typing import Optional, Tuple
from typing import Optional
import torch
from torch import nn
@@ -62,5 +62,5 @@ class TestAttention(nn.Module):
def __init__(self):
pass
def forward(self) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
def forward(self) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
_ = apply_rotary_pos_emb(1, 1, 1, 1)

12
examples/modular-transformers/modeling_dummy.py
View File

@@ -5,7 +5,7 @@
# modular_dummy.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from functools import partial
from typing import Callable, Optional, Tuple, Union
from typing import Callable, Optional, Union
import torch
from torch import nn
@@ -223,12 +223,12 @@ class DummyAttention(nn.Module):
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: Tuple[torch.Tensor, torch.Tensor],
position_embeddings: tuple[torch.Tensor, torch.Tensor],
attention_mask: Optional[torch.Tensor],
past_key_value: Optional[Cache] = None,
cache_position: Optional[torch.LongTensor] = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
input_shape = hidden_states.shape[:-1]
hidden_shape = (*input_shape, -1, self.head_dim)
@@ -290,9 +290,9 @@ class DummyDecoderLayer(nn.Module):
output_attentions: Optional[bool] = False,
use_cache: Optional[bool] = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
@@ -494,7 +494,7 @@ class DummyModel(DummyPreTrainedModel):
return_dict: Optional[bool] = None,
cache_position: Optional[torch.LongTensor] = None,
**flash_attn_kwargs: Unpack[FlashAttentionKwargs],
) -> Union[Tuple, BaseModelOutputWithPast]:
) -> Union[tuple, BaseModelOutputWithPast]:
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states

26
examples/modular-transformers/modeling_dummy_bert.py
View File

@@ -6,7 +6,7 @@
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
import math
import os
from typing import List, Optional, Tuple, Union
from typing import Optional, Union
import torch
from packaging import version
@@ -136,9 +136,9 @@ class DummyBertSelfAttention(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
mixed_query_layer = self.query(hidden_states)
# If this is instantiated as a cross-attention module, the keys
@@ -245,9 +245,9 @@ class DummyBertSdpaSelfAttention(DummyBertSelfAttention):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
if self.position_embedding_type != "absolute" or output_attentions or head_mask is not None:
# TODO: Improve this warning with e.g. `model.config._attn_implementation = "manual"` once implemented.
logger.warning_once(
@@ -386,9 +386,9 @@ class DummyBertAttention(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
self_outputs = self.self(
hidden_states,
attention_mask,
@@ -454,9 +454,9 @@ class DummyBertLayer(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
# decoder uni-directional self-attention cached key/values tuple is at positions 1,2
self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
self_attention_outputs = self.attention(
@@ -532,12 +532,12 @@ class DummyBertEncoder(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_values: Optional[tuple[tuple[torch.FloatTensor]]] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = False,
output_hidden_states: Optional[bool] = False,
return_dict: Optional[bool] = True,
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
) -> Union[tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
all_hidden_states = () if output_hidden_states else None
all_self_attentions = () if output_attentions else None
all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
@@ -858,12 +858,12 @@ class DummyBertModel(DummyBertPreTrainedModel):
inputs_embeds: Optional[torch.Tensor] = None,
encoder_hidden_states: Optional[torch.Tensor] = None,
encoder_attention_mask: Optional[torch.Tensor] = None,
past_key_values: Optional[List[torch.FloatTensor]] = None,
past_key_values: Optional[list[torch.FloatTensor]] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
) -> Union[tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
r"""
encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if

10
examples/modular-transformers/modeling_from_uppercase_model.py
View File

@@ -4,7 +4,7 @@
# the file from the modular. If any change should be done, please apply the change to the
# modular_from_uppercase_model.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from typing import Optional, Tuple
from typing import Optional
import torch
from torch import nn
@@ -53,7 +53,7 @@ class FromUppercaseModelAttention(nn.Module):
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
"""Input shape: Batch x Time x Channel"""
bsz, tgt_len, embed_dim = hidden_states.size()
@@ -148,7 +148,7 @@ class FromUppercaseModelFlashAttention2(FromUppercaseModelAttention):
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
output_attentions = False
batch_size, q_len, _ = hidden_states.size()
@@ -226,7 +226,7 @@ class FromUppercaseModelSdpaAttention(FromUppercaseModelAttention):
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
if output_attentions:
# TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
logger.warning_once(
@@ -322,7 +322,7 @@ class FromUppercaseModelEncoderLayer(nn.Module):
attention_mask: torch.Tensor,
causal_attention_mask: torch.Tensor,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.FloatTensor]:
) -> tuple[torch.FloatTensor]:
"""
Args:
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`

12
examples/modular-transformers/modeling_multimodal1.py
View File

@@ -5,7 +5,7 @@
# modular_multimodal1.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from functools import partial
from typing import Callable, Optional, Tuple, Union
from typing import Callable, Optional, Union
import torch
from torch import nn
@@ -223,12 +223,12 @@ class Multimodal1TextAttention(nn.Module):
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: Tuple[torch.Tensor, torch.Tensor],
position_embeddings: tuple[torch.Tensor, torch.Tensor],
attention_mask: Optional[torch.Tensor],
past_key_value: Optional[Cache] = None,
cache_position: Optional[torch.LongTensor] = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
input_shape = hidden_states.shape[:-1]
hidden_shape = (*input_shape, -1, self.head_dim)
@@ -290,9 +290,9 @@ class Multimodal1TextDecoderLayer(nn.Module):
output_attentions: Optional[bool] = False,
use_cache: Optional[bool] = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
@@ -494,7 +494,7 @@ class Multimodal1TextModel(Multimodal1TextPreTrainedModel):
return_dict: Optional[bool] = None,
cache_position: Optional[torch.LongTensor] = None,
**flash_attn_kwargs: Unpack[FlashAttentionKwargs],
) -> Union[Tuple, BaseModelOutputWithPast]:
) -> Union[tuple, BaseModelOutputWithPast]:
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states

16
examples/modular-transformers/modeling_multimodal2.py
View File

@@ -5,7 +5,7 @@
# modular_multimodal2.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from typing import Optional, Tuple, Union
from typing import Optional, Union
import torch
from torch import nn
@@ -65,7 +65,7 @@ class Multimodal2VisionAttention(nn.Module):
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
"""Input shape: Batch x Time x Channel"""
bsz, tgt_len, embed_dim = hidden_states.size()
@@ -152,7 +152,7 @@ class Multimodal2VisionSdpaAttention(Multimodal2VisionAttention):
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
if output_attentions:
# TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
logger.warning_once(
@@ -233,7 +233,7 @@ class Multimodal2VisionFlashAttention2(Multimodal2VisionAttention):
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
output_attentions = False
batch_size, q_len, _ = hidden_states.size()
@@ -334,7 +334,7 @@ class Multimodal2VisionEncoderLayer(nn.Module):
attention_mask: torch.Tensor,
causal_attention_mask: torch.Tensor,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.FloatTensor]:
) -> tuple[torch.FloatTensor]:
"""
Args:
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
@@ -392,7 +392,7 @@ class Multimodal2VisionEncoder(nn.Module):
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple, BaseModelOutput]:
) -> Union[tuple, BaseModelOutput]:
r"""
Args:
inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
@@ -587,7 +587,7 @@ class Multimodal2VisionTransformer(nn.Module):
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
interpolate_pos_encoding: Optional[bool] = False,
) -> Union[Tuple, BaseModelOutputWithPooling]:
) -> Union[tuple, BaseModelOutputWithPooling]:
r"""
Returns:
@@ -671,7 +671,7 @@ class Multimodal2VisionModel(Multimodal2VisionPreTrainedModel):
output_hidden_states: Optional[bool] = None,
interpolate_pos_encoding: bool = False,
return_dict: Optional[bool] = None,
) -> Union[Tuple, BaseModelOutputWithPooling]:
) -> Union[tuple, BaseModelOutputWithPooling]:
r"""
Returns:

18
examples/modular-transformers/modeling_my_new_model2.py
View File

@@ -4,7 +4,7 @@
# the file from the modular. If any change should be done, please apply the change to the
# modular_my_new_model2.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from typing import Callable, List, Optional, Tuple, Union
from typing import Callable, Optional, Union
import torch
from torch import nn
@@ -222,12 +222,12 @@ class MyNewModel2Attention(nn.Module):
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: Tuple[torch.Tensor, torch.Tensor],
position_embeddings: tuple[torch.Tensor, torch.Tensor],
attention_mask: Optional[torch.Tensor],
past_key_value: Optional[Cache] = None,
cache_position: Optional[torch.LongTensor] = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
input_shape = hidden_states.shape[:-1]
hidden_shape = (*input_shape, -1, self.head_dim)
@@ -289,9 +289,9 @@ class MyNewModel2DecoderLayer(nn.Module):
output_attentions: Optional[bool] = False,
use_cache: Optional[bool] = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
@@ -485,7 +485,7 @@ class MyNewModel2Model(MyNewModel2PreTrainedModel):
input_ids: torch.LongTensor = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
past_key_values: Optional[Union[Cache, list[torch.FloatTensor]]] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
@@ -493,7 +493,7 @@ class MyNewModel2Model(MyNewModel2PreTrainedModel):
return_dict: Optional[bool] = None,
cache_position: Optional[torch.LongTensor] = None,
**kwargs, # NOOP kwarg for now
) -> Union[Tuple, BaseModelOutputWithPast]:
) -> Union[tuple, BaseModelOutputWithPast]:
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
@@ -753,14 +753,14 @@ class MyNewModel2ForSequenceClassification(MyNewModel2PreTrainedModel):
input_ids: Optional[torch.LongTensor] = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
past_key_values: Optional[Union[Cache, list[torch.FloatTensor]]] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
labels: Optional[torch.LongTensor] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple, SequenceClassifierOutputWithPast]:
) -> Union[tuple, SequenceClassifierOutputWithPast]:
r"""
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,

12
examples/modular-transformers/modeling_new_task_model.py
View File

@@ -5,7 +5,7 @@
# modular_new_task_model.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from dataclasses import dataclass
from typing import ClassVar, List, Optional, Tuple, Union
from typing import ClassVar, Optional, Union
import torch
from torch import nn
@@ -61,9 +61,9 @@ class NewTaskModelCausalLMOutputWithPast(ModelOutput):
loss: Optional[torch.FloatTensor] = None
logits: torch.FloatTensor = None
past_key_values: Optional[Union[List[torch.FloatTensor], Cache]] = None
hidden_states: Optional[Tuple[torch.FloatTensor]] = None
attentions: Optional[Tuple[torch.FloatTensor]] = None
past_key_values: Optional[Union[list[torch.FloatTensor], Cache]] = None
hidden_states: Optional[tuple[torch.FloatTensor]] = None
attentions: Optional[tuple[torch.FloatTensor]] = None
image_hidden_states: Optional[torch.FloatTensor] = None
@@ -337,7 +337,7 @@ class NewTaskModelForNewTask(NewTaskModelPreTrainedModel, GenerationMixin):
pixel_values: torch.FloatTensor = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Union[List[torch.FloatTensor], Cache]] = None,
past_key_values: Optional[Union[list[torch.FloatTensor], Cache]] = None,
token_type_ids: Optional[torch.LongTensor] = None,
cache_position: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
@@ -347,7 +347,7 @@ class NewTaskModelForNewTask(NewTaskModelPreTrainedModel, GenerationMixin):
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
num_logits_to_keep: int = 0,
) -> Union[Tuple, NewTaskModelCausalLMOutputWithPast]:
) -> Union[tuple, NewTaskModelCausalLMOutputWithPast]:
r"""
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,

26
examples/modular-transformers/modeling_roberta.py
View File

@@ -6,7 +6,7 @@
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
import math
import os
from typing import List, Optional, Tuple, Union
from typing import Optional, Union
import torch
import torch.nn as nn
@@ -139,9 +139,9 @@ class RobertaSelfAttention(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
mixed_query_layer = self.query(hidden_states)
# If this is instantiated as a cross-attention module, the keys
@@ -248,9 +248,9 @@ class RobertaSdpaSelfAttention(RobertaSelfAttention):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
if self.position_embedding_type != "absolute" or output_attentions or head_mask is not None:
# TODO: Improve this warning with e.g. `model.config._attn_implementation = "manual"` once implemented.
logger.warning_once(
@@ -389,9 +389,9 @@ class RobertaAttention(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
self_outputs = self.self(
hidden_states,
attention_mask,
@@ -457,9 +457,9 @@ class RobertaLayer(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
) -> tuple[torch.Tensor]:
# decoder uni-directional self-attention cached key/values tuple is at positions 1,2
self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
self_attention_outputs = self.attention(
@@ -535,12 +535,12 @@ class RobertaEncoder(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
past_key_values: Optional[tuple[tuple[torch.FloatTensor]]] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = False,
output_hidden_states: Optional[bool] = False,
return_dict: Optional[bool] = True,
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
) -> Union[tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
all_hidden_states = () if output_hidden_states else None
all_self_attentions = () if output_attentions else None
all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
@@ -861,12 +861,12 @@ class RobertaModel(RobertaPreTrainedModel):
inputs_embeds: Optional[torch.Tensor] = None,
encoder_hidden_states: Optional[torch.Tensor] = None,
encoder_attention_mask: Optional[torch.Tensor] = None,
past_key_values: Optional[List[torch.FloatTensor]] = None,
past_key_values: Optional[list[torch.FloatTensor]] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
) -> Union[tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
r"""
encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if

14
examples/modular-transformers/modeling_super.py
View File

@@ -4,7 +4,7 @@
# the file from the modular. If any change should be done, please apply the change to the
# modular_super.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from typing import Callable, List, Optional, Tuple, Union
from typing import Callable, Optional, Union
import torch
from torch import nn
@@ -222,12 +222,12 @@ class SuperAttention(nn.Module):
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: Tuple[torch.Tensor, torch.Tensor],
position_embeddings: tuple[torch.Tensor, torch.Tensor],
attention_mask: Optional[torch.Tensor],
past_key_value: Optional[Cache] = None,
cache_position: Optional[torch.LongTensor] = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
input_shape = hidden_states.shape[:-1]
hidden_shape = (*input_shape, -1, self.head_dim)
@@ -289,9 +289,9 @@ class SuperDecoderLayer(nn.Module):
output_attentions: Optional[bool] = False,
use_cache: Optional[bool] = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
@@ -485,14 +485,14 @@ class SuperModel(SuperPreTrainedModel):
input_ids: torch.LongTensor = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
past_key_values: Optional[Union[Cache, list[torch.FloatTensor]]] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
cache_position: Optional[torch.LongTensor] = None,
) -> Union[Tuple, BaseModelOutputWithPast]:
) -> Union[tuple, BaseModelOutputWithPast]:
out = super().forward(
input_ids,
attention_mask,

6
examples/modular-transformers/modeling_switch_function.py
View File

@@ -5,7 +5,7 @@
# modular_switch_function.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Note that llama and cohere have different definitions for rotate_half
from typing import Callable, Optional, Tuple
from typing import Callable, Optional
import torch
from torch import nn
@@ -123,12 +123,12 @@ class SwitchFunctionAttention(nn.Module):
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: Tuple[torch.Tensor, torch.Tensor],
position_embeddings: tuple[torch.Tensor, torch.Tensor],
attention_mask: Optional[torch.Tensor],
past_key_value: Optional[Cache] = None,
cache_position: Optional[torch.LongTensor] = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
input_shape = hidden_states.shape[:-1]
hidden_shape = (*input_shape, -1, self.head_dim)

6
examples/modular-transformers/modular_dummy_bert.py
View File

@@ -1,4 +1,4 @@
from typing import List, Optional, Tuple, Union
from typing import Optional, Union
import torch
@@ -18,10 +18,10 @@ class DummyBertModel(BertModel):
inputs_embeds: Optional[torch.Tensor] = None,
encoder_hidden_states: Optional[torch.Tensor] = None,
encoder_attention_mask: Optional[torch.Tensor] = None,
past_key_values: Optional[List[torch.FloatTensor]] = None,
past_key_values: Optional[list[torch.FloatTensor]] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
) -> Union[tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
return super().forward(input_ids)

4
examples/modular-transformers/modular_new_task_model.py
View File

@@ -1,4 +1,4 @@
from typing import ClassVar, List, Optional, Union
from typing import ClassVar, Optional, Union
import torch
import torch.utils.checkpoint
@@ -29,7 +29,7 @@ class NewTaskModelForNewTask(PaliGemmaForConditionalGeneration):
pixel_values: torch.FloatTensor = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Union[List[torch.FloatTensor], Cache]] = None,
past_key_values: Optional[Union[list[torch.FloatTensor], Cache]] = None,
token_type_ids: Optional[torch.LongTensor] = None,
cache_position: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,

6
examples/modular-transformers/modular_super.py
View File

@@ -1,4 +1,4 @@
from typing import List, Optional, Tuple, Union
from typing import Optional, Union
import torch
@@ -15,14 +15,14 @@ class SuperModel(LlamaModel):
input_ids: torch.LongTensor = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
past_key_values: Optional[Union[Cache, list[torch.FloatTensor]]] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
cache_position: Optional[torch.LongTensor] = None,
) -> Union[Tuple, CausalLMOutputWithPast]:
) -> Union[tuple, CausalLMOutputWithPast]:
out = super().forward(
input_ids,
attention_mask,