File size: 8,446 Bytes
e082821 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 | # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# This file was automatically generated from examples/modular-transformers/modular_from_uppercase_model.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_from_uppercase_model.py file directly. One of our CI enforces this.
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
from typing import Callable, Optional, Union
import torch
from torch import nn
from ...activations import ACT2FN
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS
from ...utils import logging
from .configuration_from_uppercase_model import FromUppercaseModelTextConfig, FromUppercaseModelVisionConfig
logger = logging.get_logger(__name__)
def eager_attention_forward(
module: nn.Module,
query: torch.Tensor,
key: torch.Tensor,
value: torch.Tensor,
attention_mask: Optional[torch.Tensor],
scaling: float,
dropout: float = 0.0,
output_attentions: bool = True,
**kwargs,
):
attn_weights = torch.matmul(query, key.transpose(-1, -2)) * scaling
if attention_mask is not None:
attn_weights = attn_weights + attention_mask
attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
attn_output = torch.matmul(attn_weights, value)
attn_output = attn_output.transpose(1, 2).contiguous()
if not output_attentions:
attn_weights = None
return attn_output, attn_weights
class FromUppercaseModelAttention(nn.Module):
"""Multi-headed attention from 'Attention Is All You Need' paper"""
def __init__(self, config: Union[FromUppercaseModelVisionConfig, FromUppercaseModelTextConfig]):
super().__init__()
self.config = config
self.embed_dim = config.hidden_size
self.num_heads = config.num_attention_heads
self.head_dim = self.embed_dim // self.num_heads
if self.head_dim * self.num_heads != self.embed_dim:
raise ValueError(
f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
f" {self.num_heads})."
)
self.scale = self.head_dim**-0.5
self.dropout = config.attention_dropout
self.is_causal = False
self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)
def forward(
self,
hidden_states: torch.Tensor,
attention_mask: Optional[torch.Tensor] = None,
causal_attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = False,
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
"""Input shape: Batch x Time x Channel"""
batch_size, seq_length, embed_dim = hidden_states.shape
queries = self.q_proj(hidden_states)
keys = self.k_proj(hidden_states)
values = self.v_proj(hidden_states)
queries = queries.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
keys = keys.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
values = values.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
# FROM_UPPERCASE_MODEL text model uses both `causal_attention_mask` and `attention_mask`
# in case FA2 kernel is called, `is_causal` should be inferred from `causal_attention_mask`
if self.config._attn_implementation == "flash_attention_2":
self.is_causal = causal_attention_mask is not None
else:
if attention_mask is not None and causal_attention_mask is not None:
attention_mask = attention_mask + causal_attention_mask
elif causal_attention_mask is not None:
attention_mask = causal_attention_mask
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and output_attentions:
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,
queries,
keys,
values,
attention_mask,
is_causal=self.is_causal,
scaling=self.scale,
dropout=0.0 if not self.training else self.dropout,
output_attentions=output_attentions,
)
attn_output = attn_output.reshape(batch_size, seq_length, embed_dim).contiguous()
attn_output = self.out_proj(attn_output)
if not output_attentions:
attn_weights = None
return attn_output, attn_weights
class FromUppercaseModelMLP(nn.Module):
def __init__(self, config):
super().__init__()
self.config = config
self.activation_fn = ACT2FN[config.hidden_act]
self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
hidden_states = self.fc1(hidden_states)
hidden_states = self.activation_fn(hidden_states)
hidden_states = self.fc2(hidden_states)
return hidden_states
class FromUppercaseModelEncoderLayer(GradientCheckpointingLayer):
def __init__(self, config: Union[FromUppercaseModelVisionConfig, FromUppercaseModelTextConfig]):
super().__init__()
self.embed_dim = config.hidden_size
self.self_attn = FromUppercaseModelAttention(config)
self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
self.mlp = FromUppercaseModelMLP(config)
self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
def forward(
self,
hidden_states: torch.Tensor,
attention_mask: torch.Tensor,
causal_attention_mask: torch.Tensor,
output_attentions: Optional[bool] = False,
) -> tuple[torch.FloatTensor]:
"""
Args:
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
attention_mask (`torch.FloatTensor`): attention mask of size
`(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
`(config.encoder_attention_heads,)`.
output_attentions (`bool`, *optional*):
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
returned tensors for more detail.
"""
residual = hidden_states
hidden_states = self.layer_norm1(hidden_states)
hidden_states, attn_weights = self.self_attn(
hidden_states=hidden_states,
attention_mask=attention_mask,
causal_attention_mask=causal_attention_mask,
output_attentions=output_attentions,
)
hidden_states = residual + hidden_states
residual = hidden_states
hidden_states = self.layer_norm2(hidden_states)
hidden_states = self.mlp(hidden_states)
hidden_states = residual + hidden_states
outputs = (hidden_states,)
if output_attentions:
outputs += (attn_weights,)
return outputs
|