vilsonrodrigues commited on
Commit
68fa802
1 Parent(s): 359d2fb

add official modeling by hf

Browse files
Files changed (1) hide show
  1. modeling_falcon.py +1242 -0
modeling_falcon.py ADDED
@@ -0,0 +1,1242 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2023 the Falcon authors and HuggingFace Inc. team. All rights reserved.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+ """PyTorch Falcon model."""
16
+
17
+ import math
18
+ from typing import Optional, Tuple, Union
19
+
20
+ import torch
21
+ import torch.utils.checkpoint
22
+ from torch import nn
23
+ from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, LayerNorm, MSELoss
24
+ from torch.nn import functional as F
25
+
26
+ from transformers.modeling_outputs import (
27
+ BaseModelOutputWithPastAndCrossAttentions,
28
+ CausalLMOutputWithCrossAttentions,
29
+ QuestionAnsweringModelOutput,
30
+ SequenceClassifierOutputWithPast,
31
+ TokenClassifierOutput,
32
+ )
33
+ from transformers.modeling_utils import PreTrainedModel
34
+ from transformers.utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging
35
+ from .configuration_falcon import FalconConfig
36
+
37
+
38
+ logger = logging.get_logger(__name__)
39
+
40
+ FALCON_PRETRAINED_MODEL_ARCHIVE_LIST = [
41
+ "tiiuae/falcon-40b",
42
+ "tiiuae/falcon-40b-instruct",
43
+ "tiiuae/falcon-7b",
44
+ "tiiuae/falcon-7b-instruct",
45
+ "tiiuae/falcon-rw-7b",
46
+ "tiiuae/falcon-rw-1b",
47
+ ]
48
+ _CHECKPOINT_FOR_DOC = "Rocketknight1/falcon-rw-1b"
49
+ _CONFIG_FOR_DOC = "FalconConfig"
50
+
51
+
52
+ # NOTE(Hesslow): Unfortunately we did not fuse matmul and bias during training, this means that there's one additional quantization to bfloat16 between the operations.
53
+ # In order not to degrade the quality of our HF-port, we keep these characteristics in the final model.
54
+ class FalconLinear(nn.Linear):
55
+ def forward(self, input: torch.Tensor) -> torch.Tensor:
56
+ hidden_states = input @ self.weight.T
57
+ if self.bias is None:
58
+ return hidden_states
59
+ return hidden_states + self.bias
60
+
61
+
62
+ # rotary pos emb helpers (torch.jit.script does not seem to support staticmethod...)
63
+ def rotate_half(x):
64
+ x1, x2 = x[..., : x.shape[-1] // 2], x[..., x.shape[-1] // 2 :]
65
+ return torch.cat((-x2, x1), dim=-1)
66
+
67
+
68
+ class FalconRotaryEmbedding(nn.Module):
69
+ """Implementation of RotaryEmbedding from GPT-NeoX.
70
+ This implementation is designed to operate on queries and keys that are compatible with `[batch_size,
71
+ n_heads_per_partition, seq_len, head_dim]` (e.g. MinGPTAttention format).
72
+ """
73
+
74
+ def __init__(self, head_dim: int, base=10000):
75
+ super().__init__()
76
+ inv_freq = 1.0 / (base ** (torch.arange(0, head_dim, 2).float() / head_dim))
77
+ self.register_buffer("inv_freq", inv_freq, persistent=False)
78
+ self.head_dim = head_dim
79
+ self.seq_len_cached = -1
80
+ self.cos_cached: torch.Tensor | None = None
81
+ self.sin_cached: torch.Tensor | None = None
82
+
83
+ def cos_sin(self, seq_len: int, past_key_values_length: int, device="cpu", dtype=torch.bfloat16) -> torch.Tensor:
84
+ total_length = seq_len + past_key_values_length
85
+ if total_length > self.seq_len_cached:
86
+ self.seq_len_cached = total_length
87
+ t = torch.arange(total_length, device=device, dtype=self.inv_freq.dtype)
88
+ freqs = torch.einsum("i,j->ij", t, self.inv_freq)
89
+ emb = torch.cat((freqs, freqs), dim=-1).to(device)
90
+
91
+ if dtype in [torch.float16, torch.bfloat16]:
92
+ emb = emb.float()
93
+
94
+ self.cos_cached = emb.cos()[None, :, :]
95
+ self.sin_cached = emb.sin()[None, :, :]
96
+
97
+ self.cos_cached = self.cos_cached.type(dtype)
98
+ self.sin_cached = self.sin_cached.type(dtype)
99
+
100
+ return (
101
+ self.cos_cached[:, past_key_values_length : seq_len + past_key_values_length],
102
+ self.sin_cached[:, past_key_values_length : seq_len + past_key_values_length],
103
+ )
104
+
105
+ def forward(self, query, key, past_key_values_length=0):
106
+ batch, seq_len, head_dim = query.shape
107
+ cos, sin = self.cos_sin(seq_len, past_key_values_length, query.device, query.dtype)
108
+ return (query * cos) + (rotate_half(query) * sin), (key * cos) + (rotate_half(key) * sin)
109
+
110
+
111
+ def _make_causal_mask(
112
+ input_ids_shape: torch.Size, device: torch.device, past_key_values_length: int
113
+ ) -> torch.BoolTensor:
114
+ """
115
+ Make causal mask used for self-attention. This mask does not take the existing attention mask into account - it
116
+ just blocks tokens from attending forwards in the sequence. The output shape will be `[batch_size, 1,
117
+ target_length, target_length+past_key_values_length]`.
118
+ """
119
+ batch_size, target_length = input_ids_shape
120
+
121
+ mask = torch.triu(torch.ones((target_length, target_length), dtype=torch.bool, device=device), diagonal=1)
122
+ # If past_key_values_length is 0 this is an empty tensor and the concatenation is a no-op.
123
+ # This code style is an unfortunate consequence of getting your TF engineer to port models; doing it this
124
+ # way avoids a data-dependent conditional, which will help me when I have to port this to XLA later.
125
+ past_mask = torch.zeros((target_length, past_key_values_length), dtype=torch.bool, device=device)
126
+ mask = torch.cat([past_mask, mask], dim=-1)
127
+ expanded_mask = mask[None, None, :, :].expand(batch_size, 1, target_length, target_length + past_key_values_length)
128
+ return expanded_mask
129
+
130
+
131
+ def _expand_mask(mask: torch.Tensor, past_key_values_length: int) -> torch.BoolTensor:
132
+ """
133
+ Expands attention_mask from `[batch_size, seq_length]` to `[batch_size, 1, seq_length, seq_length + past_length]`.
134
+ """
135
+ batch_size, total_length = mask.shape
136
+ seq_length = total_length - past_key_values_length if past_key_values_length is not None else total_length
137
+
138
+ expanded_mask = ~(mask[:, None, None, :].to(torch.bool))
139
+ return expanded_mask.expand(batch_size, 1, seq_length, total_length)
140
+
141
+
142
+ def build_alibi_tensor(attention_mask: torch.Tensor, num_heads: int, dtype: torch.dtype) -> torch.Tensor:
143
+ batch_size, seq_length = attention_mask.shape
144
+ closest_power_of_2 = 2 ** math.floor(math.log2(num_heads))
145
+ base = torch.tensor(
146
+ 2 ** (-(2 ** -(math.log2(closest_power_of_2) - 3))), device=attention_mask.device, dtype=torch.float32
147
+ )
148
+ powers = torch.arange(1, 1 + closest_power_of_2, device=attention_mask.device, dtype=torch.int32)
149
+ slopes = torch.pow(base, powers)
150
+
151
+ if closest_power_of_2 != num_heads:
152
+ extra_base = torch.tensor(
153
+ 2 ** (-(2 ** -(math.log2(2 * closest_power_of_2) - 3))), device=attention_mask.device, dtype=torch.float32
154
+ )
155
+ num_remaining_heads = min(closest_power_of_2, num_heads - closest_power_of_2)
156
+ extra_powers = torch.arange(1, 1 + 2 * num_remaining_heads, 2, device=attention_mask.device, dtype=torch.int32)
157
+ slopes = torch.cat([slopes, torch.pow(extra_base, extra_powers)], dim=0)
158
+
159
+ # Note: alibi will added to the attention bias that will be applied to the query, key product of attention
160
+ # => therefore alibi will have to be of shape (batch_size, num_heads, query_length, key_length)
161
+ # => here we set (batch_size=1, num_heads=num_heads, query_length=1, key_length=max_length)
162
+ # => the query_length dimension will then be broadcasted correctly
163
+ # This is more or less identical to T5's relative position bias:
164
+ # https://github.com/huggingface/transformers/blob/f681437203baa7671de3174b0fa583c349d9d5e1/src/transformers/models/t5/modeling_t5.py#L527
165
+ arange_tensor = ((attention_mask.cumsum(dim=-1) - 1) * attention_mask)[:, None, :]
166
+ alibi = slopes[..., None].bfloat16() * arange_tensor
167
+ return alibi.reshape(batch_size * num_heads, 1, seq_length).to(dtype)
168
+
169
+
170
+ # Copied from transformers.models.bloom.modeling_bloom.dropout_add
171
+ def dropout_add(x: torch.Tensor, residual: torch.Tensor, prob: float, training: bool) -> torch.Tensor:
172
+ """
173
+ Dropout add function
174
+ Args:
175
+ x (`torch.tensor`, *required*):
176
+ input tensor
177
+ residual (`torch.tensor`, *required*):
178
+ residual tensor
179
+ prob (`float`, *required*):
180
+ dropout probability
181
+ training (`bool`, *required*):
182
+ training mode
183
+ """
184
+ out = F.dropout(x, p=prob, training=training)
185
+ out = residual + out
186
+ return out
187
+
188
+
189
+ class FalconAttention(nn.Module):
190
+ def __init__(self, config: FalconConfig):
191
+ super().__init__()
192
+
193
+ self.hidden_size = config.hidden_size
194
+ self.num_heads = config.num_attention_heads
195
+ self.head_dim = self.hidden_size // self.num_heads
196
+ self.split_size = self.hidden_size
197
+ self.hidden_dropout = config.hidden_dropout
198
+
199
+ if self.head_dim * self.num_heads != self.hidden_size:
200
+ raise ValueError(
201
+ f"`hidden_size` must be divisible by num_heads (got `hidden_size`: {self.hidden_size} and `num_heads`:"
202
+ f" {self.num_heads})."
203
+ )
204
+
205
+ self.maybe_rotary = FalconRotaryEmbedding(config.head_dim) if config.rotary else lambda q, k, t: (q, k)
206
+
207
+ # Layer-wise attention scaling
208
+ self.inv_norm_factor = 1.0 / math.sqrt(self.head_dim)
209
+ self.beta = self.inv_norm_factor
210
+ if config.new_decoder_architecture:
211
+ qkv_out_dim = (config.num_kv_heads * 2 + config.num_attention_heads) * self.head_dim
212
+ elif config.multi_query:
213
+ qkv_out_dim = self.hidden_size + 2 * self.head_dim
214
+ else:
215
+ qkv_out_dim = 3 * self.hidden_size
216
+ self.query_key_value = FalconLinear(self.hidden_size, qkv_out_dim, bias=config.bias)
217
+ self.new_decoder_architecture = config.new_decoder_architecture
218
+ self.multi_query = config.multi_query
219
+ self.dense = FalconLinear(self.hidden_size, self.hidden_size, bias=config.bias)
220
+ self.attention_dropout = nn.Dropout(config.attention_dropout)
221
+ self.num_kv_heads = config.num_kv_heads if (self.new_decoder_architecture or not self.multi_query) else 1
222
+
223
+ def _split_heads(self, fused_qkv: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
224
+ """
225
+ Split the last dimension into (num_heads, head_dim), results share same memory storage as `fused_qkv`
226
+ Args:
227
+ fused_qkv (`torch.tensor`, *required*): [batch_size, seq_length, num_heads * 3 * head_dim]
228
+ Returns:
229
+ query: [batch_size, seq_length, num_heads, head_dim] key: [batch_size, seq_length, num_heads, head_dim]
230
+ value: [batch_size, seq_length, num_heads, head_dim]
231
+ """
232
+ if self.new_decoder_architecture:
233
+ batch, seq_len, _ = fused_qkv.shape
234
+ qkv = fused_qkv.view(batch, seq_len, -1, self.num_heads // self.num_kv_heads + 2, self.head_dim)
235
+ query = qkv[:, :, :, :-2]
236
+ key = qkv[:, :, :, [-2]]
237
+ value = qkv[:, :, :, [-1]]
238
+ key = torch.broadcast_to(key, query.shape)
239
+ value = torch.broadcast_to(value, query.shape)
240
+
241
+ query, key, value = [x.flatten(2, 3) for x in (query, key, value)]
242
+ return query, key, value
243
+ elif not self.multi_query:
244
+ batch_size, seq_length, three_times_hidden_size = fused_qkv.shape
245
+ fused_qkv = fused_qkv.view(batch_size, seq_length, self.num_heads, 3, self.head_dim)
246
+ return fused_qkv[..., 0, :], fused_qkv[..., 1, :], fused_qkv[..., 2, :]
247
+ else:
248
+ batch_size, seq_length, three_times_hidden_size = fused_qkv.shape
249
+ fused_qkv = fused_qkv.view(batch_size, seq_length, self.num_heads + 2, self.head_dim)
250
+ return fused_qkv[..., :-2, :], fused_qkv[..., [-2], :], fused_qkv[..., [-1], :]
251
+
252
+ # Copied from transformers.models.bloom.modeling_bloom.BloomAttention._merge_heads
253
+ def _merge_heads(self, x: torch.Tensor) -> torch.Tensor:
254
+ """
255
+ Merge heads together over the last dimenstion
256
+ Args:
257
+ x (`torch.tensor`, *required*): [batch_size * num_heads, seq_length, head_dim]
258
+ Returns:
259
+ torch.tensor: [batch_size, seq_length, num_heads * head_dim]
260
+ """
261
+ # What we want to achieve is:
262
+ # batch_size * num_heads, seq_length, head_dim -> batch_size, seq_length, num_heads * head_dim
263
+ batch_size_and_num_heads, seq_length, _ = x.shape
264
+ batch_size = batch_size_and_num_heads // self.num_heads
265
+
266
+ # First view to decompose the batch size
267
+ # batch_size * num_heads, seq_length, head_dim -> batch_size, num_heads, seq_length, head_dim
268
+ x = x.view(batch_size, self.num_heads, seq_length, self.head_dim)
269
+
270
+ # batch_size, num_heads, seq_length, head_dim -> batch_size, seq_length, num_heads, head_dim
271
+ x = x.permute(0, 2, 1, 3)
272
+
273
+ # batch_size, seq_length, num_heads, head_dim -> batch_size, seq_length, num_heads * head_dim
274
+ return x.reshape(batch_size, seq_length, self.num_heads * self.head_dim)
275
+
276
+ def forward(
277
+ self,
278
+ hidden_states: torch.Tensor,
279
+ alibi: Optional[torch.Tensor],
280
+ attention_mask: torch.Tensor,
281
+ layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
282
+ head_mask: Optional[torch.Tensor] = None,
283
+ use_cache: bool = False,
284
+ output_attentions: bool = False,
285
+ ):
286
+ fused_qkv = self.query_key_value(hidden_states) # [batch_size, seq_length, 3 x hidden_size]
287
+ num_kv_heads = self.num_heads if self.new_decoder_architecture else self.num_kv_heads
288
+ # 3 x [batch_size, seq_length, num_heads, head_dim]
289
+ (query_layer, key_layer, value_layer) = self._split_heads(fused_qkv)
290
+
291
+ batch_size, query_length, _, _ = query_layer.shape
292
+
293
+ query_layer = query_layer.transpose(1, 2).reshape(batch_size * self.num_heads, query_length, self.head_dim)
294
+ key_layer = key_layer.transpose(1, 2).reshape(
295
+ batch_size * num_kv_heads,
296
+ query_length,
297
+ self.head_dim,
298
+ )
299
+ value_layer = value_layer.transpose(1, 2).reshape(batch_size * num_kv_heads, query_length, self.head_dim)
300
+
301
+ past_kv_length = 0 if layer_past is None else layer_past[0].shape[1]
302
+ query_layer, key_layer = self.maybe_rotary(query_layer, key_layer, past_kv_length)
303
+
304
+ if layer_past is not None:
305
+ past_key, past_value = layer_past
306
+ # concatenate along seq_length dimension:
307
+ # - key: [batch_size * self.num_heads, kv_length, head_dim]
308
+ # - value: [batch_size * self.num_heads, kv_length, head_dim]
309
+ key_layer = torch.cat((past_key, key_layer), dim=1)
310
+ value_layer = torch.cat((past_value, value_layer), dim=1)
311
+
312
+ _, kv_length, _ = key_layer.shape
313
+ if use_cache:
314
+ present = (key_layer, value_layer)
315
+ else:
316
+ present = None
317
+
318
+ attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, float("-1e9")).to(query_layer.dtype)
319
+
320
+ query_layer_ = query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
321
+ key_layer_ = key_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
322
+ value_layer_ = value_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
323
+
324
+ if alibi is None:
325
+ if output_attentions:
326
+ # F.scaled_dot_product_attention doesn't return the attention weights, so we have
327
+ # to do it by hand if we want them
328
+ attention_scores = query_layer_ @ key_layer_.transpose(-1, -2)
329
+ attention_scores /= math.sqrt(self.head_dim)
330
+
331
+ attention_scores = F.softmax(
332
+ attention_scores + attention_mask_float, dim=-1, dtype=hidden_states.dtype
333
+ )
334
+ attn_output = attention_scores @ value_layer_
335
+ else:
336
+ attn_output = F.scaled_dot_product_attention(
337
+ query_layer_, key_layer_, value_layer_, attention_mask_float, 0.0, is_causal=False
338
+ )
339
+ attention_scores = None
340
+
341
+ attn_output = attn_output.view(batch_size, self.num_heads, query_length, self.head_dim)
342
+ attn_output = attn_output.permute(0, 2, 1, 3)
343
+ attn_output = attn_output.reshape(batch_size, query_length, self.num_heads * self.head_dim)
344
+
345
+ output_tensor = self.dense(attn_output)
346
+
347
+ if output_attentions:
348
+ return output_tensor, present, attention_scores
349
+ else:
350
+ return output_tensor, present
351
+
352
+ else:
353
+ matmul_result = query_layer_ @ key_layer_.transpose(-1, -2)
354
+
355
+ # change view to [batch_size, num_heads, q_length, kv_length]
356
+ attention_scores = matmul_result.view(batch_size, self.num_heads, query_length, kv_length)
357
+
358
+ # cast attention scores to fp32, compute scaled softmax and cast back to initial dtype - [batch_size, num_heads, q_length, kv_length]
359
+ input_dtype = attention_scores.dtype
360
+ # `float16` has a minimum value of -65504.0, whereas `bfloat16` and `float32` have a minimum value of `-3.4e+38`
361
+ if input_dtype == torch.float16 or input_dtype == torch.bfloat16:
362
+ attention_scores = attention_scores.to(torch.float32)
363
+ # Matt (HF) note: We could possibly use F.scaled_dot_product_attention here too, by
364
+ # adding (alibi * self.inv_norm_factor) to attention_mask_float. I think this would be mathematically
365
+ # equivalent and more performant, but there might be a numerical difference. If you're reading this
366
+ # and you'd like to experiment and maybe file a PR, feel free!
367
+ attention_logits = attention_scores + alibi.view(batch_size, self.num_heads, 1, -1)
368
+ attention_logits *= self.inv_norm_factor
369
+ attention_probs = F.softmax(attention_logits + attention_mask_float, dim=-1, dtype=hidden_states.dtype)
370
+ # [batch_size, num_heads, q_length, kv_length]
371
+ attention_probs = self.attention_dropout(attention_probs)
372
+
373
+ if head_mask is not None:
374
+ attention_probs = attention_probs * head_mask
375
+
376
+ # change view [batch_size, num_heads, q_length, kv_length]
377
+ attention_probs_reshaped = attention_probs.view(batch_size, self.num_heads, query_length, kv_length)
378
+
379
+ # matmul: [batch_size * num_heads, q_length, head_dim]
380
+ context_layer = (attention_probs_reshaped @ value_layer_).flatten(0, 1)
381
+
382
+ # change view [batch_size, num_heads, q_length, head_dim]
383
+ context_layer = self._merge_heads(context_layer)
384
+
385
+ output_tensor = self.dense(context_layer)
386
+
387
+ if output_attentions:
388
+ return output_tensor, present, attention_probs
389
+ else:
390
+ return output_tensor, present
391
+
392
+
393
+ class FalconMLP(nn.Module):
394
+ def __init__(self, config: FalconConfig):
395
+ super().__init__()
396
+ hidden_size = config.hidden_size
397
+
398
+ self.dense_h_to_4h = FalconLinear(hidden_size, 4 * hidden_size, bias=config.bias)
399
+ self.act = nn.GELU()
400
+ self.dense_4h_to_h = FalconLinear(4 * hidden_size, hidden_size, bias=config.bias)
401
+ self.hidden_dropout = config.hidden_dropout
402
+
403
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
404
+ x = self.act(self.dense_h_to_4h(x))
405
+ x = self.dense_4h_to_h(x)
406
+ return x
407
+
408
+
409
+ class FalconDecoderLayer(nn.Module):
410
+ def __init__(self, config: FalconConfig):
411
+ super().__init__()
412
+ hidden_size = config.hidden_size
413
+ self.num_heads = config.num_attention_heads
414
+ self.self_attention = FalconAttention(config)
415
+ self.mlp = FalconMLP(config)
416
+ self.hidden_dropout = config.hidden_dropout
417
+ self.config = config
418
+
419
+ if config.new_decoder_architecture:
420
+ # The layer norm before self-attention
421
+ self.ln_attn = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
422
+ # The layer norm before the MLP
423
+ self.ln_mlp = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
424
+ else:
425
+ self.input_layernorm = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
426
+ if not config.parallel_attn:
427
+ self.post_attention_layernorm = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
428
+
429
+ def forward(
430
+ self,
431
+ hidden_states: torch.Tensor,
432
+ alibi: Optional[torch.Tensor],
433
+ attention_mask: torch.Tensor,
434
+ layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
435
+ head_mask: Optional[torch.Tensor] = None,
436
+ use_cache: bool = False,
437
+ output_attentions: bool = False,
438
+ ):
439
+ residual = hidden_states
440
+
441
+ if self.config.new_decoder_architecture:
442
+ attention_layernorm_out = self.ln_attn(hidden_states)
443
+ mlp_layernorm_out = self.ln_mlp(hidden_states)
444
+ else:
445
+ attention_layernorm_out = self.input_layernorm(hidden_states)
446
+
447
+ # Self attention.
448
+ attn_outputs = self.self_attention(
449
+ attention_layernorm_out,
450
+ layer_past=layer_past,
451
+ attention_mask=attention_mask,
452
+ alibi=alibi,
453
+ head_mask=head_mask,
454
+ use_cache=use_cache,
455
+ output_attentions=output_attentions,
456
+ )
457
+
458
+ attention_output = attn_outputs[0]
459
+
460
+ if not self.config.new_decoder_architecture:
461
+ if self.config.parallel_attn:
462
+ mlp_layernorm_out = attention_layernorm_out
463
+ else:
464
+ residual = dropout_add(
465
+ attention_output, residual, self.config.attention_dropout, training=self.training
466
+ )
467
+ mlp_layernorm_out = self.post_attention_layernorm(residual)
468
+
469
+ outputs = attn_outputs[1:]
470
+
471
+ # MLP.
472
+ mlp_output = self.mlp(mlp_layernorm_out)
473
+
474
+ if self.config.new_decoder_architecture or self.config.parallel_attn:
475
+ mlp_output += attention_output
476
+
477
+ output = dropout_add(mlp_output, residual, self.config.hidden_dropout, training=self.training)
478
+
479
+ if use_cache:
480
+ outputs = (output,) + outputs
481
+ else:
482
+ outputs = (output,) + outputs[1:]
483
+
484
+ return outputs # hidden_states, present, attentions
485
+
486
+
487
+ FALCON_START_DOCSTRING = r"""
488
+ This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
489
+ library implements for all its model (such as downloading or saving, resizing the input embeddings etc.)
490
+ This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
491
+ Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
492
+ and behavior.
493
+ Parameters:
494
+ config ([`FalconConfig`]): Model configuration class with all the parameters of the model.
495
+ Initializing with a config file does not load the weights associated with the model, only the
496
+ configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
497
+ """
498
+
499
+ FALCON_INPUTS_DOCSTRING = r"""
500
+ Args:
501
+ input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
502
+ `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0][0].shape[2]`
503
+ (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary.
504
+ If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as
505
+ `input_ids`.
506
+ Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
507
+ [`PreTrainedTokenizer.__call__`] for details.
508
+ [What are input IDs?](../glossary#input-ids)
509
+ past_key_values (`Tuple[Tuple[torch.Tensor]]` of length `config.num_hidden_layers`):
510
+ Contains precomputed hidden-states (key and values in the attention blocks) as computed by the model (see
511
+ `past_key_values` output below). Can be used to speed up sequential decoding. The `input_ids` which have
512
+ their past given to this model should not be passed as `input_ids` as they have already been computed.
513
+ Each element of `past_key_values` is a tuple (past_key, past_value):
514
+ - past_key: [batch_size * num_heads, head_dim, kv_length]
515
+ - past_value: [batch_size * num_heads, kv_length, head_dim]
516
+ attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
517
+ Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
518
+ - 1 for tokens that are **not masked**,
519
+ - 0 for tokens that are **masked**.
520
+ [What are attention masks?](../glossary#attention-mask)
521
+ head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
522
+ Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
523
+ - 1 indicates the head is **not masked**,
524
+ - 0 indicates the head is **masked**.
525
+ inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
526
+ Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
527
+ is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
528
+ model's internal embedding lookup matrix.
529
+ If `past_key_values` is used, optionally only the last `inputs_embeds` have to be input (see
530
+ `past_key_values`).
531
+ use_cache (`bool`, *optional*):
532
+ If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
533
+ `past_key_values`).
534
+ output_attentions (`bool`, *optional*):
535
+ Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
536
+ tensors for more detail.
537
+ output_hidden_states (`bool`, *optional*):
538
+ Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
539
+ more detail.
540
+ return_dict (`bool`, *optional*):
541
+ Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
542
+ """
543
+
544
+
545
+ class FalconPreTrainedModel(PreTrainedModel):
546
+ """
547
+ An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
548
+ models.
549
+ """
550
+
551
+ config_class = FalconConfig
552
+ base_model_prefix = "transformer"
553
+ supports_gradient_checkpointing = True
554
+ _no_split_modules = ["FalconDecoderLayer"]
555
+
556
+ def __init__(self, *inputs, **kwargs):
557
+ super().__init__(*inputs, **kwargs)
558
+
559
+ def _init_weights(self, module: nn.Module):
560
+ """Initialize the weights."""
561
+ if isinstance(module, nn.Linear) or isinstance(module, FalconLinear):
562
+ # Slightly different from the TF version which uses truncated_normal for initialization
563
+ # cf https://github.com/pytorch/pytorch/pull/5617
564
+ module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
565
+ if module.bias is not None:
566
+ module.bias.data.zero_()
567
+ elif isinstance(module, nn.Embedding):
568
+ module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
569
+ if module.padding_idx is not None:
570
+ module.weight.data[module.padding_idx].zero_()
571
+ elif isinstance(module, LayerNorm):
572
+ module.bias.data.zero_()
573
+ module.weight.data.fill_(1.0)
574
+
575
+ # Copied from transformers.models.bloom.modeling_bloom.BloomPreTrainedModel._set_gradient_checkpointing with BloomModel->FalconModel
576
+ def _set_gradient_checkpointing(self, module: nn.Module, value: bool = False):
577
+ if isinstance(module, FalconModel):
578
+ module.gradient_checkpointing = value
579
+
580
+ @staticmethod
581
+ def _convert_cache_to_standard_format(
582
+ past_key_value: Tuple[Tuple[torch.Tensor, torch.Tensor]], batch_size: int
583
+ ) -> Tuple[Tuple[torch.Tensor, torch.Tensor]]:
584
+ """
585
+ Standardizes the format of the cache so as to match most implementations, i.e. to tuple(tuple([batch_size,
586
+ num_heads, ...]))
587
+ """
588
+ batch_size_times_num_heads, kv_length, head_dim = past_key_value[0][0].shape
589
+ # [batch_size * self.num_heads, kv_length, head_dim] -> [batch_size, num_heads, kv_length, head_dim]
590
+ # Note that don't want to use self.num_attention_heads because the number of heads may vary depending
591
+ # on whether we use multi_query attention.
592
+ num_heads = batch_size_times_num_heads // batch_size
593
+ return tuple(
594
+ (
595
+ layer_past[0].view(batch_size, num_heads, kv_length, head_dim),
596
+ layer_past[1].view(batch_size, num_heads, kv_length, head_dim),
597
+ )
598
+ for layer_past in past_key_value
599
+ )
600
+
601
+ @staticmethod
602
+ def _convert_to_rw_cache(
603
+ past_key_value: Tuple[Tuple[torch.Tensor, torch.Tensor]]
604
+ ) -> Tuple[Tuple[torch.Tensor, torch.Tensor]]:
605
+ batch_size, num_heads, kv_length, head_dim = past_key_value[0][0].shape
606
+ batch_size_times_num_heads = batch_size * num_heads
607
+ # [batch_size, num_heads, kv_length, head_dim] -> [batch_size * num_heads, kv_length, head_dim]
608
+ return tuple(
609
+ (
610
+ layer_past[0].view(batch_size_times_num_heads, kv_length, head_dim),
611
+ layer_past[1].view(batch_size_times_num_heads, kv_length, head_dim),
612
+ )
613
+ for layer_past in past_key_value
614
+ )
615
+
616
+
617
+ @add_start_docstrings(
618
+ "The bare Falcon Model transformer outputting raw hidden-states without any specific head on top.",
619
+ FALCON_START_DOCSTRING,
620
+ )
621
+ class FalconModel(FalconPreTrainedModel):
622
+ def __init__(self, config: FalconConfig):
623
+ super().__init__(config)
624
+
625
+ self.embed_dim = config.hidden_size
626
+ self.num_heads = config.num_attention_heads
627
+ self.use_alibi = config.alibi
628
+
629
+ # Embedding + LN Embedding
630
+ self.word_embeddings = nn.Embedding(config.vocab_size, self.embed_dim)
631
+
632
+ # Transformer blocks
633
+ self.h = nn.ModuleList([FalconDecoderLayer(config) for _ in range(config.num_hidden_layers)])
634
+
635
+ # Final Layer Norm
636
+ self.ln_f = LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
637
+
638
+ self.gradient_checkpointing = False
639
+
640
+ # Initialize weights and apply final processing
641
+ self.post_init()
642
+
643
+ def get_input_embeddings(self):
644
+ return self.word_embeddings
645
+
646
+ @staticmethod
647
+ def _prepare_attn_mask(
648
+ attention_mask: torch.Tensor, input_shape: Tuple[int, int], past_key_values_length: int
649
+ ) -> torch.BoolTensor:
650
+ # Create a causal mask
651
+ # The attention mask we receive as input should cover the whole extended sequence, including any past
652
+ # cache, so its shape should be [batch_size, seq_length + past_key_values_length]
653
+ # The output shape will be [batch_size, 1, seq_length, seq_length + past_key_values_length]
654
+ if input_shape[1] + past_key_values_length != attention_mask.shape[1]:
655
+ raise ValueError(
656
+ "Attention mask shape should be (batch_size, seq_length + past_key_values_length)"
657
+ f" but is {attention_mask.shape} with input_ids shape {input_shape} and past length"
658
+ f" {past_key_values_length}."
659
+ )
660
+ combined_attention_mask = None
661
+ device = attention_mask.device
662
+ _, seq_length = input_shape
663
+
664
+ if seq_length > 1:
665
+ combined_attention_mask = _make_causal_mask(
666
+ input_shape, device=device, past_key_values_length=past_key_values_length
667
+ )
668
+
669
+ # [batch_size, seq_length + past_key_values_length] -> [batch_size, 1, seq_length, seq_length + past_key_values_length]
670
+ expanded_attn_mask = _expand_mask(attention_mask, past_key_values_length=past_key_values_length)
671
+ combined_attention_mask = (
672
+ expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask | combined_attention_mask
673
+ )
674
+
675
+ return combined_attention_mask
676
+
677
+ def set_input_embeddings(self, new_embeddings: torch.Tensor):
678
+ self.word_embeddings = new_embeddings
679
+
680
+ @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
681
+ @add_code_sample_docstrings(
682
+ checkpoint=_CHECKPOINT_FOR_DOC,
683
+ output_type=BaseModelOutputWithPastAndCrossAttentions,
684
+ config_class=_CONFIG_FOR_DOC,
685
+ )
686
+ def forward(
687
+ self,
688
+ input_ids: Optional[torch.LongTensor] = None,
689
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
690
+ attention_mask: Optional[torch.Tensor] = None,
691
+ head_mask: Optional[torch.LongTensor] = None,
692
+ inputs_embeds: Optional[torch.LongTensor] = None,
693
+ use_cache: Optional[bool] = None,
694
+ output_attentions: Optional[bool] = None,
695
+ output_hidden_states: Optional[bool] = None,
696
+ return_dict: Optional[bool] = None,
697
+ ) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPastAndCrossAttentions]:
698
+ output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
699
+ output_hidden_states = (
700
+ output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
701
+ )
702
+ use_cache = use_cache if use_cache is not None else self.config.use_cache
703
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
704
+
705
+ if input_ids is not None and inputs_embeds is not None:
706
+ raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
707
+ elif input_ids is not None:
708
+ batch_size, seq_length = input_ids.shape
709
+ elif inputs_embeds is not None:
710
+ batch_size, seq_length, _ = inputs_embeds.shape
711
+ else:
712
+ raise ValueError("You have to specify either input_ids or inputs_embeds")
713
+
714
+ if past_key_values is None:
715
+ past_key_values = tuple([None] * len(self.h))
716
+ else:
717
+ past_key_values = self._convert_to_rw_cache(past_key_values)
718
+
719
+ # Prepare head mask if needed
720
+ # 1.0 in head_mask indicate we keep the head
721
+ # attention_probs has shape batch_size x num_heads x N x N
722
+ # head_mask has shape n_layer x batch x num_heads x N x N
723
+ head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
724
+
725
+ if inputs_embeds is None:
726
+ inputs_embeds = self.word_embeddings(input_ids)
727
+
728
+ hidden_states = inputs_embeds
729
+
730
+ presents = () if use_cache else None
731
+ all_self_attentions = () if output_attentions else None
732
+ all_hidden_states = () if output_hidden_states else None
733
+
734
+ # Compute alibi tensor: check build_alibi_tensor documentation
735
+ past_key_values_length = 0
736
+ if past_key_values[0] is not None:
737
+ past_key_values_length = past_key_values[0][0].shape[1] # 1 because RW-cache, not standard format
738
+ if attention_mask is None:
739
+ attention_mask = torch.ones((batch_size, seq_length + past_key_values_length), device=hidden_states.device)
740
+ else:
741
+ attention_mask = attention_mask.to(hidden_states.device)
742
+
743
+ if self.use_alibi:
744
+ alibi = build_alibi_tensor(attention_mask, self.num_heads, dtype=hidden_states.dtype)
745
+ else:
746
+ alibi = None
747
+
748
+ causal_mask = self._prepare_attn_mask(
749
+ attention_mask,
750
+ input_shape=(batch_size, seq_length),
751
+ past_key_values_length=past_key_values_length,
752
+ )
753
+
754
+ for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
755
+ if output_hidden_states:
756
+ all_hidden_states = all_hidden_states + (hidden_states,)
757
+
758
+ if self.gradient_checkpointing and self.training:
759
+ if use_cache:
760
+ logger.warning(
761
+ "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
762
+ )
763
+ use_cache = False
764
+
765
+ def create_custom_forward(module):
766
+ def custom_forward(*inputs):
767
+ # None for past_key_value
768
+ return module(*inputs, use_cache=use_cache, output_attentions=output_attentions)
769
+
770
+ return custom_forward
771
+
772
+ outputs = torch.utils.checkpoint.checkpoint(
773
+ create_custom_forward(block),
774
+ hidden_states,
775
+ alibi,
776
+ causal_mask,
777
+ head_mask[i],
778
+ )
779
+ else:
780
+ outputs = block(
781
+ hidden_states,
782
+ layer_past=layer_past,
783
+ attention_mask=causal_mask,
784
+ head_mask=head_mask[i],
785
+ use_cache=use_cache,
786
+ output_attentions=output_attentions,
787
+ alibi=alibi,
788
+ )
789
+
790
+ hidden_states = outputs[0]
791
+ if use_cache is True:
792
+ presents = presents + (outputs[1],)
793
+
794
+ if output_attentions:
795
+ all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
796
+
797
+ # Add last hidden state
798
+ hidden_states = self.ln_f(hidden_states)
799
+
800
+ if output_hidden_states:
801
+ all_hidden_states = all_hidden_states + (hidden_states,)
802
+
803
+ if presents is not None:
804
+ presents = self._convert_cache_to_standard_format(presents, batch_size)
805
+
806
+ if not return_dict:
807
+ return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
808
+
809
+ return BaseModelOutputWithPastAndCrossAttentions(
810
+ last_hidden_state=hidden_states,
811
+ past_key_values=presents,
812
+ hidden_states=all_hidden_states,
813
+ attentions=all_self_attentions,
814
+ )
815
+
816
+
817
+ @add_start_docstrings(
818
+ "The Falcon Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).",
819
+ FALCON_START_DOCSTRING,
820
+ )
821
+ class FalconForCausalLM(FalconPreTrainedModel):
822
+ _tied_weights_keys = ["lm_head.weight"]
823
+
824
+ def __init__(self, config: FalconConfig):
825
+ super().__init__(config)
826
+ self.transformer = FalconModel(config)
827
+ self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
828
+
829
+ # Initialize weights and apply final processing
830
+ self.post_init()
831
+
832
+ def get_output_embeddings(self):
833
+ return self.lm_head
834
+
835
+ def set_output_embeddings(self, new_embeddings: torch.Tensor):
836
+ self.lm_head = new_embeddings
837
+
838
+ def prepare_inputs_for_generation(
839
+ self,
840
+ input_ids: torch.LongTensor,
841
+ past_key_values: Optional[torch.Tensor] = None,
842
+ attention_mask: Optional[torch.Tensor] = None,
843
+ **kwargs,
844
+ ) -> dict:
845
+ if past_key_values is not None:
846
+ input_ids = input_ids[:, -1:]
847
+
848
+ return {
849
+ "input_ids": input_ids,
850
+ "past_key_values": past_key_values,
851
+ "use_cache": kwargs.get("use_cache"),
852
+ "attention_mask": attention_mask,
853
+ }
854
+
855
+ @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
856
+ @add_code_sample_docstrings(
857
+ checkpoint=_CHECKPOINT_FOR_DOC,
858
+ output_type=CausalLMOutputWithCrossAttentions,
859
+ config_class=_CONFIG_FOR_DOC,
860
+ )
861
+ def forward(
862
+ self,
863
+ input_ids: Optional[torch.LongTensor] = None,
864
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
865
+ attention_mask: Optional[torch.Tensor] = None,
866
+ head_mask: Optional[torch.Tensor] = None,
867
+ inputs_embeds: Optional[torch.Tensor] = None,
868
+ labels: Optional[torch.Tensor] = None,
869
+ use_cache: Optional[bool] = None,
870
+ output_attentions: Optional[bool] = None,
871
+ output_hidden_states: Optional[bool] = None,
872
+ return_dict: Optional[bool] = None,
873
+ ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
874
+ r"""
875
+ labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
876
+ Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
877
+ `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
878
+ are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
879
+ """
880
+
881
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
882
+
883
+ transformer_outputs = self.transformer(
884
+ input_ids,
885
+ past_key_values=past_key_values,
886
+ attention_mask=attention_mask,
887
+ head_mask=head_mask,
888
+ inputs_embeds=inputs_embeds,
889
+ use_cache=use_cache,
890
+ output_attentions=output_attentions,
891
+ output_hidden_states=output_hidden_states,
892
+ return_dict=return_dict,
893
+ )
894
+ hidden_states = transformer_outputs[0]
895
+
896
+ lm_logits = self.lm_head(hidden_states)
897
+
898
+ loss = None
899
+ if labels is not None:
900
+ # Shift so that tokens < n predict n
901
+ shift_logits = lm_logits[..., :-1, :].contiguous()
902
+ shift_labels = labels[..., 1:].contiguous()
903
+ batch_size, seq_length, vocab_size = shift_logits.shape
904
+ # Flatten the tokens
905
+ loss_fct = CrossEntropyLoss()
906
+ loss = loss_fct(
907
+ shift_logits.view(batch_size * seq_length, vocab_size), shift_labels.view(batch_size * seq_length)
908
+ )
909
+
910
+ if not return_dict:
911
+ output = (lm_logits,) + transformer_outputs[1:]
912
+ return ((loss,) + output) if loss is not None else output
913
+
914
+ return CausalLMOutputWithCrossAttentions(
915
+ loss=loss,
916
+ logits=lm_logits,
917
+ past_key_values=transformer_outputs.past_key_values,
918
+ hidden_states=transformer_outputs.hidden_states,
919
+ attentions=transformer_outputs.attentions,
920
+ )
921
+
922
+ def _reorder_cache(
923
+ self, past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
924
+ ) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
925
+ """
926
+ This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
927
+ [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
928
+ beam_idx at every generation step.
929
+ Output shares the same memory storage as `past`.
930
+ """
931
+
932
+ # Get a copy of `beam_idx` on all the devices where we need those indices.
933
+ device_to_beam_idx = {
934
+ past_state.device: beam_idx.to(past_state.device) for layer_past in past for past_state in layer_past
935
+ }
936
+ reordered_past = tuple(
937
+ (
938
+ layer_past[0].index_select(0, device_to_beam_idx[layer_past[0].device]),
939
+ layer_past[1].index_select(0, device_to_beam_idx[layer_past[0].device]),
940
+ )
941
+ for layer_past in past
942
+ )
943
+ return reordered_past
944
+
945
+
946
+ @add_start_docstrings(
947
+ """
948
+ The Falcon Model transformer with a sequence classification head on top (linear layer).
949
+ [`FalconForSequenceClassification`] uses the last token in order to do the classification, as other causal models
950
+ (e.g. GPT-1) do.
951
+ Since it does classification on the last token, it requires to know the position of the last token. If a
952
+ `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
953
+ no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
954
+ padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
955
+ each row of the batch).
956
+ """,
957
+ FALCON_START_DOCSTRING,
958
+ )
959
+ class FalconForSequenceClassification(FalconPreTrainedModel):
960
+ def __init__(self, config: FalconConfig):
961
+ super().__init__(config)
962
+ self.num_labels = config.num_labels
963
+ self.transformer = FalconModel(config)
964
+ self.score = nn.Linear(config.hidden_size, config.num_labels, bias=False)
965
+
966
+ # Initialize weights and apply final processing
967
+ self.post_init()
968
+
969
+ @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
970
+ @add_code_sample_docstrings(
971
+ checkpoint=_CHECKPOINT_FOR_DOC,
972
+ output_type=SequenceClassifierOutputWithPast,
973
+ config_class=_CONFIG_FOR_DOC,
974
+ )
975
+ def forward(
976
+ self,
977
+ input_ids: Optional[torch.LongTensor] = None,
978
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
979
+ attention_mask: Optional[torch.Tensor] = None,
980
+ head_mask: Optional[torch.Tensor] = None,
981
+ inputs_embeds: Optional[torch.Tensor] = None,
982
+ labels: Optional[torch.Tensor] = None,
983
+ use_cache: Optional[bool] = None,
984
+ output_attentions: Optional[bool] = None,
985
+ output_hidden_states: Optional[bool] = None,
986
+ return_dict: Optional[bool] = None,
987
+ ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutputWithPast]:
988
+ r"""
989
+ labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
990
+ Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
991
+ config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
992
+ `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
993
+ """
994
+
995
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
996
+
997
+ transformer_outputs = self.transformer(
998
+ input_ids,
999
+ past_key_values=past_key_values,
1000
+ attention_mask=attention_mask,
1001
+ head_mask=head_mask,
1002
+ inputs_embeds=inputs_embeds,
1003
+ use_cache=use_cache,
1004
+ output_attentions=output_attentions,
1005
+ output_hidden_states=output_hidden_states,
1006
+ return_dict=return_dict,
1007
+ )
1008
+
1009
+ hidden_states = transformer_outputs[0]
1010
+ logits = self.score(hidden_states)
1011
+
1012
+ if input_ids is not None:
1013
+ batch_size = input_ids.shape[0]
1014
+ else:
1015
+ batch_size = inputs_embeds.shape[0]
1016
+
1017
+ if self.config.pad_token_id is None and batch_size != 1:
1018
+ raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
1019
+ if self.config.pad_token_id is None:
1020
+ sequence_lengths = -1
1021
+ else:
1022
+ if input_ids is not None:
1023
+ sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(dim=-1) - 1
1024
+ else:
1025
+ sequence_lengths = -1
1026
+ logger.warning(
1027
+ f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be "
1028
+ "unexpected if using padding tokens in conjunction with `inputs_embeds.`"
1029
+ )
1030
+
1031
+ pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths]
1032
+
1033
+ loss = None
1034
+ if labels is not None:
1035
+ if self.config.problem_type is None:
1036
+ if self.num_labels == 1:
1037
+ self.config.problem_type = "regression"
1038
+ elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
1039
+ self.config.problem_type = "single_label_classification"
1040
+ else:
1041
+ self.config.problem_type = "multi_label_classification"
1042
+
1043
+ if self.config.problem_type == "regression":
1044
+ loss_fct = MSELoss()
1045
+ if self.num_labels == 1:
1046
+ loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
1047
+ else:
1048
+ loss = loss_fct(pooled_logits, labels)
1049
+ elif self.config.problem_type == "single_label_classification":
1050
+ loss_fct = CrossEntropyLoss()
1051
+ loss = loss_fct(pooled_logits, labels)
1052
+ elif self.config.problem_type == "multi_label_classification":
1053
+ loss_fct = BCEWithLogitsLoss()
1054
+ loss = loss_fct(pooled_logits, labels)
1055
+ if not return_dict:
1056
+ output = (pooled_logits,) + transformer_outputs[1:]
1057
+ return ((loss,) + output) if loss is not None else output
1058
+
1059
+ return SequenceClassifierOutputWithPast(
1060
+ loss=loss,
1061
+ logits=pooled_logits,
1062
+ past_key_values=transformer_outputs.past_key_values,
1063
+ hidden_states=transformer_outputs.hidden_states,
1064
+ attentions=transformer_outputs.attentions,
1065
+ )
1066
+
1067
+
1068
+ @add_start_docstrings(
1069
+ """
1070
+ Falcon Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for
1071
+ Named-Entity-Recognition (NER) tasks.
1072
+ """,
1073
+ FALCON_START_DOCSTRING,
1074
+ )
1075
+ class FalconForTokenClassification(FalconPreTrainedModel):
1076
+ def __init__(self, config: FalconConfig):
1077
+ super().__init__(config)
1078
+ self.num_labels = config.num_labels
1079
+
1080
+ self.transformer = FalconModel(config)
1081
+ if getattr(config, "classifier_dropout", None) is not None:
1082
+ classifier_dropout = config.classifier_dropout
1083
+ elif getattr(config, "hidden_dropout", None) is not None:
1084
+ classifier_dropout = config.hidden_dropout
1085
+ else:
1086
+ classifier_dropout = 0.1
1087
+ self.dropout = nn.Dropout(classifier_dropout)
1088
+ self.classifier = nn.Linear(config.hidden_size, config.num_labels)
1089
+
1090
+ # Initialize weights and apply final processing
1091
+ self.post_init()
1092
+
1093
+ @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
1094
+ @add_code_sample_docstrings(
1095
+ checkpoint=_CHECKPOINT_FOR_DOC,
1096
+ output_type=TokenClassifierOutput,
1097
+ config_class=_CONFIG_FOR_DOC,
1098
+ )
1099
+ def forward(
1100
+ self,
1101
+ input_ids: Optional[torch.LongTensor] = None,
1102
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
1103
+ attention_mask: Optional[torch.Tensor] = None,
1104
+ head_mask: Optional[torch.Tensor] = None,
1105
+ inputs_embeds: Optional[torch.Tensor] = None,
1106
+ labels: Optional[torch.Tensor] = None,
1107
+ use_cache: Optional[bool] = None,
1108
+ output_attentions: Optional[bool] = None,
1109
+ output_hidden_states: Optional[bool] = None,
1110
+ return_dict: Optional[bool] = None,
1111
+ ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:
1112
+ r"""
1113
+ labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
1114
+ Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
1115
+ config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
1116
+ `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
1117
+ """
1118
+
1119
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
1120
+
1121
+ transformer_outputs = self.transformer(
1122
+ input_ids,
1123
+ past_key_values=past_key_values,
1124
+ attention_mask=attention_mask,
1125
+ head_mask=head_mask,
1126
+ inputs_embeds=inputs_embeds,
1127
+ use_cache=use_cache,
1128
+ output_attentions=output_attentions,
1129
+ output_hidden_states=output_hidden_states,
1130
+ return_dict=return_dict,
1131
+ )
1132
+
1133
+ hidden_states = transformer_outputs[0]
1134
+ hidden_states = self.dropout(hidden_states)
1135
+ logits = self.classifier(hidden_states)
1136
+
1137
+ loss = None
1138
+ if labels is not None:
1139
+ batch_size, seq_length = labels.shape
1140
+ loss_fct = CrossEntropyLoss()
1141
+ loss = loss_fct(
1142
+ logits.view(batch_size * seq_length, self.num_labels), labels.view(batch_size * seq_length)
1143
+ )
1144
+
1145
+ if not return_dict:
1146
+ output = (logits,) + transformer_outputs[2:]
1147
+ return ((loss,) + output) if loss is not None else output
1148
+
1149
+ return TokenClassifierOutput(
1150
+ loss=loss,
1151
+ logits=logits,
1152
+ hidden_states=transformer_outputs.hidden_states,
1153
+ attentions=transformer_outputs.attentions,
1154
+ )
1155
+
1156
+
1157
+ @add_start_docstrings(
1158
+ """
1159
+ The Falcon Model transformer with a span classification head on top for extractive question-answering tasks like
1160
+ SQuAD (a linear layers on top of the hidden-states output to compute `span start logits` and `span end logits`).
1161
+ """,
1162
+ FALCON_START_DOCSTRING,
1163
+ )
1164
+ class FalconForQuestionAnswering(FalconPreTrainedModel):
1165
+ def __init__(self, config):
1166
+ super().__init__(config)
1167
+ self.transformer = FalconModel(config)
1168
+ self.qa_outputs = nn.Linear(config.hidden_size, 2)
1169
+
1170
+ # Initialize weights and apply final processing
1171
+ self.post_init()
1172
+
1173
+ @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
1174
+ def forward(
1175
+ self,
1176
+ input_ids: Optional[torch.LongTensor] = None,
1177
+ attention_mask: Optional[torch.FloatTensor] = None,
1178
+ head_mask: Optional[torch.FloatTensor] = None,
1179
+ inputs_embeds: Optional[torch.FloatTensor] = None,
1180
+ start_positions: Optional[torch.LongTensor] = None,
1181
+ end_positions: Optional[torch.LongTensor] = None,
1182
+ output_attentions: Optional[bool] = None,
1183
+ output_hidden_states: Optional[bool] = None,
1184
+ return_dict: Optional[bool] = None,
1185
+ ) -> Union[Tuple, QuestionAnsweringModelOutput]:
1186
+ r"""
1187
+ start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
1188
+ Labels for position (index) of the start of the labelled span for computing the token classification loss.
1189
+ Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
1190
+ are not taken into account for computing the loss.
1191
+ end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
1192
+ Labels for position (index) of the end of the labelled span for computing the token classification loss.
1193
+ Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
1194
+ are not taken into account for computing the loss.
1195
+ """
1196
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
1197
+
1198
+ outputs = self.transformer(
1199
+ input_ids,
1200
+ attention_mask=attention_mask,
1201
+ head_mask=head_mask,
1202
+ inputs_embeds=inputs_embeds,
1203
+ output_attentions=output_attentions,
1204
+ output_hidden_states=output_hidden_states,
1205
+ return_dict=return_dict,
1206
+ )
1207
+
1208
+ sequence_output = outputs[0]
1209
+
1210
+ logits = self.qa_outputs(sequence_output)
1211
+ start_logits, end_logits = logits.split(1, dim=-1)
1212
+ start_logits = start_logits.squeeze(-1).contiguous()
1213
+ end_logits = end_logits.squeeze(-1).contiguous()
1214
+
1215
+ total_loss = None
1216
+ if start_positions is not None and end_positions is not None:
1217
+ # If we are on multi-GPU, split add a dimension
1218
+ if len(start_positions.size()) > 1:
1219
+ start_positions = start_positions.squeeze(-1)
1220
+ if len(end_positions.size()) > 1:
1221
+ end_positions = end_positions.squeeze(-1)
1222
+ # sometimes the start/end positions are outside our model inputs, we ignore these terms
1223
+ ignored_index = start_logits.size(1)
1224
+ start_positions = start_positions.clamp(0, ignored_index)
1225
+ end_positions = end_positions.clamp(0, ignored_index)
1226
+
1227
+ loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
1228
+ start_loss = loss_fct(start_logits, start_positions)
1229
+ end_loss = loss_fct(end_logits, end_positions)
1230
+ total_loss = (start_loss + end_loss) / 2
1231
+
1232
+ if not return_dict:
1233
+ output = (start_logits, end_logits) + outputs[2:]
1234
+ return ((total_loss,) + output) if total_loss is not None else output
1235
+
1236
+ return QuestionAnsweringModelOutput(
1237
+ loss=total_loss,
1238
+ start_logits=start_logits,
1239
+ end_logits=end_logits,
1240
+ hidden_states=outputs.hidden_states,
1241
+ attentions=outputs.attentions,
1242
+ )