TideDra commited on
Commit
16408ed
1 Parent(s): b05c510

Upload 7 files

Browse files
configuration_qwen.py ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) Alibaba Cloud.
2
+ #
3
+ # This source code is licensed under the license found in the
4
+ # LICENSE file in the root directory of this source tree.
5
+
6
+ from transformers import PretrainedConfig
7
+
8
+
9
+ class QWenConfig(PretrainedConfig):
10
+ model_type = "qwen"
11
+ keys_to_ignore_at_inference = ["past_key_values"]
12
+
13
+ def __init__(
14
+ self,
15
+ vocab_size=151936,
16
+ hidden_size=4096,
17
+ num_hidden_layers=32,
18
+ num_attention_heads=32,
19
+ emb_dropout_prob=0.0,
20
+ attn_dropout_prob=0.0,
21
+ layer_norm_epsilon=1e-6,
22
+ initializer_range=0.02,
23
+ max_position_embeddings=8192,
24
+ scale_attn_weights=True,
25
+ use_cache=True,
26
+ bf16=False,
27
+ fp16=False,
28
+ fp32=False,
29
+ kv_channels=128,
30
+ rotary_pct=1.0,
31
+ rotary_emb_base=10000,
32
+ use_dynamic_ntk=True,
33
+ use_logn_attn=True,
34
+ use_flash_attn="auto",
35
+ intermediate_size=22016,
36
+ no_bias=True,
37
+ tie_word_embeddings=False,
38
+ **kwargs,
39
+ ):
40
+ self.vocab_size = vocab_size
41
+ self.hidden_size = hidden_size
42
+ self.intermediate_size = intermediate_size
43
+ self.num_hidden_layers = num_hidden_layers
44
+ self.num_attention_heads = num_attention_heads
45
+ self.emb_dropout_prob = emb_dropout_prob
46
+ self.attn_dropout_prob = attn_dropout_prob
47
+ self.layer_norm_epsilon = layer_norm_epsilon
48
+ self.initializer_range = initializer_range
49
+ self.scale_attn_weights = scale_attn_weights
50
+ self.use_cache = use_cache
51
+ self.max_position_embeddings = max_position_embeddings
52
+ self.bf16 = bf16
53
+ self.fp16 = fp16
54
+ self.fp32 = fp32
55
+ self.kv_channels = kv_channels
56
+ self.rotary_pct = rotary_pct
57
+ self.rotary_emb_base = rotary_emb_base
58
+ self.use_dynamic_ntk = use_dynamic_ntk
59
+ self.use_logn_attn = use_logn_attn
60
+ self.use_flash_attn = use_flash_attn
61
+ self.no_bias = no_bias
62
+ super().__init__(
63
+ tie_word_embeddings=tie_word_embeddings,
64
+ **kwargs
65
+ )
modeling_qwen.py ADDED
@@ -0,0 +1,1162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) Alibaba Cloud.
2
+ #
3
+ # This source code is licensed under the license found in the
4
+ # LICENSE file in the root directory of this source tree.
5
+
6
+ import importlib
7
+ import math
8
+ from typing import TYPE_CHECKING, Optional, Tuple, Union, Callable, List, Any, Generator
9
+
10
+ import torch
11
+ import torch.nn.functional as F
12
+ import torch.utils.checkpoint
13
+ from torch.cuda.amp import autocast
14
+
15
+ from torch.nn import CrossEntropyLoss
16
+ from transformers import PreTrainedTokenizer, GenerationConfig, StoppingCriteriaList
17
+ from transformers.generation.logits_process import LogitsProcessorList
18
+
19
+ if TYPE_CHECKING:
20
+ from transformers.generation.streamers import BaseStreamer
21
+ from transformers.generation.utils import GenerateOutput
22
+ from transformers.modeling_outputs import (
23
+ BaseModelOutputWithPast,
24
+ CausalLMOutputWithPast,
25
+ )
26
+ from transformers.modeling_utils import PreTrainedModel
27
+ from transformers.utils import logging
28
+
29
+ try:
30
+ from einops import rearrange
31
+ except ImportError:
32
+ rearrange = None
33
+ from torch import nn
34
+
35
+ SUPPORT_CUDA = torch.cuda.is_available()
36
+ SUPPORT_BF16 = SUPPORT_CUDA and torch.cuda.is_bf16_supported()
37
+ SUPPORT_FP16 = SUPPORT_CUDA and torch.cuda.get_device_capability(0)[0] >= 7
38
+
39
+ from .configuration_qwen import QWenConfig
40
+ from .qwen_generation_utils import (
41
+ HistoryType,
42
+ make_context,
43
+ decode_tokens,
44
+ get_stop_words_ids,
45
+ StopWordsLogitsProcessor,
46
+ )
47
+ from .visual import VisionTransformer
48
+
49
+
50
+ logger = logging.get_logger(__name__)
51
+
52
+ _CHECKPOINT_FOR_DOC = "qwen"
53
+ _CONFIG_FOR_DOC = "QWenConfig"
54
+
55
+ QWen_PRETRAINED_MODEL_ARCHIVE_LIST = ["qwen-7b"]
56
+
57
+ _ERROR_BAD_CHAT_FORMAT = """\
58
+ We detect you are probably using the pretrained model (rather than chat model) for chatting, since the chat_format in generation_config is not "chatml".
59
+ If you are directly using the model downloaded from Huggingface, please make sure you are using our "Qwen/Qwen-7B-Chat" Huggingface model (rather than "Qwen/Qwen-7B") when you call model.chat().
60
+ 我们检测到您可能在使用预训练模型(而非chat模型)进行多轮chat,因为您当前在generation_config指定的chat_format,并未设置为我们在对话中所支持的"chatml"格式。
61
+ 如果您在直接使用我们从Huggingface提供的模型,请确保您在调用model.chat()时,使用的是"Qwen/Qwen-7B-Chat"模型(而非"Qwen/Qwen-7B"预训练模型)。
62
+ """
63
+
64
+ _SENTINEL = object()
65
+ _ERROR_STREAM_IN_CHAT = """\
66
+ Pass argument `stream` to model.chat() is buggy, deprecated, and marked for removal. Please use model.chat_stream(...) instead of model.chat(..., stream=True).
67
+ 向model.chat()传入参数stream的用法可能存在Bug,该用法已被废弃,将在未来被移除。请使用model.chat_stream(...)代替model.chat(..., stream=True)。
68
+ """
69
+
70
+ apply_rotary_emb_func = None
71
+ rms_norm = None
72
+
73
+
74
+ # Copied from transformers.models.bart.modeling_bart._make_causal_mask
75
+ def _make_causal_mask(
76
+ input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
77
+ ):
78
+ """
79
+ Make causal mask used for bi-directional self-attention.
80
+ """
81
+ bsz, tgt_len = input_ids_shape
82
+ mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
83
+ mask_cond = torch.arange(mask.size(-1), device=device)
84
+ mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
85
+ mask = mask.to(dtype)
86
+
87
+ if past_key_values_length > 0:
88
+ mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
89
+ return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
90
+
91
+
92
+ # Copied from transformers.models.bart.modeling_bart._expand_mask
93
+ def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
94
+ """
95
+ Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
96
+ """
97
+ bsz, src_len = mask.size()
98
+ tgt_len = tgt_len if tgt_len is not None else src_len
99
+
100
+ expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
101
+
102
+ inverted_mask = 1.0 - expanded_mask
103
+
104
+ return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
105
+
106
+
107
+ class QWenAttention(nn.Module):
108
+ def __init__(self, config):
109
+ super().__init__()
110
+
111
+ self.register_buffer("masked_bias", torch.tensor(-1e4), persistent=False)
112
+ self.seq_length = config.seq_length
113
+
114
+ self.hidden_size = config.hidden_size
115
+ self.split_size = config.hidden_size
116
+ self.num_heads = config.num_attention_heads
117
+ self.head_dim = self.hidden_size // self.num_heads
118
+
119
+ self.scale_attn_weights = True
120
+
121
+ self.projection_size = config.kv_channels * config.num_attention_heads
122
+
123
+ assert self.projection_size % config.num_attention_heads == 0
124
+ self.hidden_size_per_attention_head = (
125
+ self.projection_size // config.num_attention_heads
126
+ )
127
+
128
+ self.c_attn = nn.Linear(config.hidden_size, 3 * self.projection_size)
129
+
130
+ self.c_proj = nn.Linear(
131
+ config.hidden_size, self.projection_size, bias=not config.no_bias
132
+ )
133
+
134
+ self.is_fp32 = not (config.bf16 or config.fp16)
135
+ self.bf16 = config.bf16
136
+
137
+ self.use_dynamic_ntk = config.use_dynamic_ntk
138
+ self.use_logn_attn = config.use_logn_attn
139
+
140
+ logn_list = [
141
+ math.log(i, self.seq_length) if i > self.seq_length else 1
142
+ for i in range(1, 32768)
143
+ ]
144
+ self.logn_tensor = torch.tensor(logn_list)[None, :, None, None]
145
+
146
+ self.attn_dropout = nn.Dropout(config.attn_dropout_prob)
147
+
148
+ def _attn(self, query, key, value, registered_causal_mask, attention_mask=None, head_mask=None):
149
+ attn_weights = torch.matmul(query, key.transpose(-1, -2))
150
+
151
+ if self.scale_attn_weights:
152
+ attn_weights = attn_weights / torch.full(
153
+ [],
154
+ value.size(-1) ** 0.5,
155
+ dtype=attn_weights.dtype,
156
+ device=attn_weights.device,
157
+ )
158
+
159
+ query_length, key_length = query.size(-2), key.size(-2)
160
+ # causal_mask = self.bias[
161
+ # :, :, key_length - query_length : key_length, :key_length
162
+ # ]
163
+ # mask_value = torch.finfo(attn_weights.dtype).min
164
+ # mask_value = torch.full([], mask_value, dtype=attn_weights.dtype).to(
165
+ # attn_weights.device
166
+ # )
167
+ # attn_weights = torch.where(
168
+ # causal_mask, attn_weights.to(attn_weights.dtype), mask_value
169
+ # )
170
+ attn_weights = attn_weights + attention_mask
171
+
172
+ attn_weights = nn.functional.softmax(attn_weights, dim=-1)
173
+
174
+ attn_weights = attn_weights.type(value.dtype)
175
+ attn_weights = self.attn_dropout(attn_weights)
176
+
177
+ if head_mask is not None:
178
+ attn_weights = attn_weights * head_mask
179
+
180
+ attn_output = torch.matmul(attn_weights, value)
181
+ attn_output = attn_output.transpose(1, 2)
182
+
183
+ return attn_output, attn_weights
184
+
185
+ def _upcast_and_reordered_attn(
186
+ self, query, key, value, registered_causal_mask, attention_mask=None, head_mask=None
187
+ ):
188
+ bsz, num_heads, q_seq_len, dk = query.size()
189
+ _, _, k_seq_len, _ = key.size()
190
+
191
+ attn_weights = torch.empty(
192
+ bsz * num_heads,
193
+ q_seq_len,
194
+ k_seq_len,
195
+ dtype=torch.float32,
196
+ device=query.device,
197
+ )
198
+
199
+ scale_factor = 1.0
200
+ if self.scale_attn_weights:
201
+ scale_factor /= float(value.size(-1)) ** 0.5
202
+
203
+ with autocast(enabled=False):
204
+ q, k = query.reshape(-1, q_seq_len, dk), key.transpose(-1, -2).reshape(
205
+ -1, dk, k_seq_len
206
+ )
207
+ attn_weights = torch.baddbmm(
208
+ attn_weights, q.float(), k.float(), beta=0, alpha=scale_factor
209
+ )
210
+ attn_weights = attn_weights.reshape(bsz, num_heads, q_seq_len, k_seq_len)
211
+
212
+ query_length, key_length = query.size(-2), key.size(-2)
213
+ causal_mask = registered_causal_mask[
214
+ :, :, key_length - query_length : key_length, :key_length
215
+ ]
216
+ mask_value = torch.finfo(attn_weights.dtype).min
217
+ mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(
218
+ attn_weights.device
219
+ )
220
+ attn_weights = torch.where(causal_mask, attn_weights, mask_value)
221
+
222
+ if attention_mask is not None:
223
+ attn_weights = attn_weights + attention_mask
224
+
225
+ attn_weights = nn.functional.softmax(attn_weights, dim=-1)
226
+
227
+ if attn_weights.dtype != torch.float32:
228
+ raise RuntimeError(
229
+ "Error with upcasting, attn_weights does not have dtype torch.float32"
230
+ )
231
+ attn_weights = attn_weights.type(value.dtype)
232
+ attn_weights = self.attn_dropout(attn_weights)
233
+
234
+ if head_mask is not None:
235
+ attn_weights = attn_weights * head_mask
236
+
237
+ attn_output = torch.matmul(attn_weights, value)
238
+
239
+ return attn_output, attn_weights
240
+
241
+ def _split_heads(self, tensor, num_heads, attn_head_size):
242
+ new_shape = tensor.size()[:-1] + (num_heads, attn_head_size)
243
+ tensor = tensor.view(new_shape)
244
+ return tensor
245
+
246
+ def _merge_heads(self, tensor, num_heads, attn_head_size):
247
+ tensor = tensor.contiguous()
248
+ new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,)
249
+ return tensor.view(new_shape)
250
+
251
+ def forward(
252
+ self,
253
+ hidden_states: Optional[Tuple[torch.FloatTensor]],
254
+ rotary_pos_emb: Optional[List[torch.Tensor]] = None,
255
+ registered_causal_mask: Optional[torch.Tensor] = None,
256
+ layer_past: Optional[Tuple[torch.Tensor]] = None,
257
+ attention_mask: Optional[torch.FloatTensor] = None,
258
+ head_mask: Optional[torch.FloatTensor] = None,
259
+ encoder_hidden_states: Optional[torch.Tensor] = None,
260
+ encoder_attention_mask: Optional[torch.FloatTensor] = None,
261
+ output_attentions: Optional[bool] = False,
262
+ use_cache: Optional[bool] = False,
263
+ ):
264
+
265
+ mixed_x_layer = self.c_attn(hidden_states)
266
+
267
+ query, key, value = mixed_x_layer.split(self.split_size, dim=2)
268
+
269
+ query = self._split_heads(query, self.num_heads, self.head_dim)
270
+ key = self._split_heads(key, self.num_heads, self.head_dim)
271
+ value = self._split_heads(value, self.num_heads, self.head_dim)
272
+
273
+ if rotary_pos_emb is not None:
274
+ cur_len = query.shape[1]
275
+ rotary_pos_emb = [i[:, -cur_len:, :, :] for i in rotary_pos_emb]
276
+ rotary_pos_emb = (rotary_pos_emb,) * 2
277
+ q_pos_emb, k_pos_emb = rotary_pos_emb
278
+ # Slice the pos emb for current inference
279
+ query = apply_rotary_pos_emb(query, q_pos_emb)
280
+ key = apply_rotary_pos_emb(key, k_pos_emb)
281
+
282
+ if layer_past is not None:
283
+ past_key, past_value = layer_past[0], layer_past[1]
284
+ key = torch.cat((past_key, key), dim=1)
285
+ value = torch.cat((past_value, value), dim=1)
286
+
287
+ if use_cache:
288
+ present = (key, value)
289
+ else:
290
+ present = None
291
+
292
+ if self.use_logn_attn and not self.training:
293
+ if self.logn_tensor.device != query.device or self.logn_tensor.dtype != query.dtype:
294
+ self.logn_tensor = self.logn_tensor.to(query.device).type_as(query)
295
+ seq_start = key.size(1) - query.size(1)
296
+ seq_end = key.size(1)
297
+ logn_tensor = self.logn_tensor[:, seq_start:seq_end, :, :]
298
+ query = query * logn_tensor.expand_as(query)
299
+
300
+ query = query.permute(0, 2, 1, 3)
301
+ key = key.permute(0, 2, 1, 3)
302
+ value = value.permute(0, 2, 1, 3)
303
+ attn_output, attn_weight = self._attn(
304
+ query, key, value, registered_causal_mask, attention_mask, head_mask
305
+ )
306
+ context_layer = self._merge_heads(
307
+ attn_output, self.num_heads, self.head_dim
308
+ )
309
+
310
+ attn_output = self.c_proj(context_layer)
311
+
312
+ outputs = (attn_output, present)
313
+ if output_attentions:
314
+ outputs += (attn_weight,)
315
+
316
+ return outputs
317
+
318
+
319
+ class QWenMLP(nn.Module):
320
+ def __init__(self, config):
321
+ super().__init__()
322
+ self.w1 = nn.Linear(
323
+ config.hidden_size, config.intermediate_size // 2, bias=not config.no_bias
324
+ )
325
+ self.w2 = nn.Linear(
326
+ config.hidden_size, config.intermediate_size // 2, bias=not config.no_bias
327
+ )
328
+ ff_dim_in = config.intermediate_size // 2
329
+ self.c_proj = nn.Linear(ff_dim_in, config.hidden_size, bias=not config.no_bias)
330
+
331
+ def forward(self, hidden_states):
332
+ a1 = self.w1(hidden_states)
333
+ a2 = self.w2(hidden_states)
334
+ intermediate_parallel = a1 * F.silu(a2)
335
+ output = self.c_proj(intermediate_parallel)
336
+ return output
337
+
338
+ class QWenBlock(nn.Module):
339
+ def __init__(self, config):
340
+ super().__init__()
341
+ hidden_size = config.hidden_size
342
+ self.bf16 = config.bf16
343
+
344
+ self.ln_1 = RMSNorm(
345
+ hidden_size,
346
+ eps=config.layer_norm_epsilon,
347
+ )
348
+ self.attn = QWenAttention(config)
349
+ self.ln_2 = RMSNorm(
350
+ hidden_size,
351
+ eps=config.layer_norm_epsilon,
352
+ )
353
+
354
+ self.mlp = QWenMLP(config)
355
+
356
+ def forward(
357
+ self,
358
+ hidden_states: Optional[Tuple[torch.FloatTensor]],
359
+ rotary_pos_emb: Optional[List[torch.Tensor]] = None,
360
+ registered_causal_mask: Optional[torch.Tensor] = None,
361
+ layer_past: Optional[Tuple[torch.Tensor]] = None,
362
+ attention_mask: Optional[torch.FloatTensor] = None,
363
+ head_mask: Optional[torch.FloatTensor] = None,
364
+ encoder_hidden_states: Optional[torch.Tensor] = None,
365
+ encoder_attention_mask: Optional[torch.FloatTensor] = None,
366
+ use_cache: Optional[bool] = False,
367
+ output_attentions: Optional[bool] = False,
368
+ ):
369
+ layernorm_output = self.ln_1(hidden_states)
370
+
371
+ attn_outputs = self.attn(
372
+ layernorm_output,
373
+ rotary_pos_emb,
374
+ registered_causal_mask=registered_causal_mask,
375
+ layer_past=layer_past,
376
+ attention_mask=attention_mask,
377
+ head_mask=head_mask,
378
+ use_cache=use_cache,
379
+ output_attentions=output_attentions,
380
+ )
381
+ attn_output = attn_outputs[0]
382
+
383
+ outputs = attn_outputs[1:]
384
+
385
+ residual = hidden_states
386
+ layernorm_input = attn_output + residual
387
+
388
+ layernorm_output = self.ln_2(layernorm_input)
389
+
390
+ residual = layernorm_input
391
+ mlp_output = self.mlp(layernorm_output)
392
+ hidden_states = residual + mlp_output
393
+
394
+ if use_cache:
395
+ outputs = (hidden_states,) + outputs
396
+ else:
397
+ outputs = (hidden_states,) + outputs[1:]
398
+
399
+ return outputs
400
+
401
+
402
+ class QWenPreTrainedModel(PreTrainedModel):
403
+ config_class = QWenConfig
404
+ base_model_prefix = "transformer"
405
+ is_parallelizable = False
406
+ supports_gradient_checkpointing = True
407
+ _no_split_modules = ["QWenBlock"]
408
+
409
+ def __init__(self, *inputs, **kwargs):
410
+ super().__init__(*inputs, **kwargs)
411
+
412
+ def _init_weights(self, module):
413
+ """Initialize the weights."""
414
+ if isinstance(module, nn.Linear):
415
+ module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
416
+ if module.bias is not None:
417
+ module.bias.data.zero_()
418
+ elif isinstance(module, nn.Embedding):
419
+ module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
420
+ if module.padding_idx is not None:
421
+ module.weight.data[module.padding_idx].zero_()
422
+ elif isinstance(module, RMSNorm):
423
+ module.weight.data.fill_(1.0)
424
+
425
+ for name, p in module.named_parameters():
426
+ if name == "c_proj.weight":
427
+ p.data.normal_(
428
+ mean=0.0,
429
+ std=(
430
+ self.config.initializer_range
431
+ / math.sqrt(2 * self.config.num_hidden_layers)
432
+ ),
433
+ )
434
+
435
+ def _set_gradient_checkpointing(self, module, value=False):
436
+ if isinstance(module, QWenModel):
437
+ module.gradient_checkpointing = value
438
+
439
+
440
+ class QWenModel(QWenPreTrainedModel):
441
+ _keys_to_ignore_on_load_missing = ["attn.masked_bias"]
442
+
443
+ def __init__(self, config):
444
+ super().__init__(config)
445
+ self.vocab_size = config.vocab_size
446
+ self.num_hidden_layers = config.num_hidden_layers
447
+ self.embed_dim = config.hidden_size
448
+
449
+ self.gradient_checkpointing = False
450
+ self.use_dynamic_ntk = config.use_dynamic_ntk
451
+ self.seq_length = config.seq_length
452
+
453
+ self.wte = nn.Embedding(self.vocab_size, self.embed_dim)
454
+
455
+ self.drop = nn.Dropout(config.emb_dropout_prob)
456
+
457
+ if config.rotary_pct == 1.0:
458
+ self.rotary_ndims = None
459
+ else:
460
+ assert config.rotary_pct < 1
461
+ self.rotary_ndims = int(
462
+ config.kv_channels * config.rotary_pct
463
+ )
464
+ dim = (
465
+ self.rotary_ndims
466
+ if self.rotary_ndims is not None
467
+ else config.kv_channels
468
+ )
469
+ self.rotary_emb = RotaryEmbedding(dim, base=config.rotary_emb_base)
470
+
471
+ self.use_flash_attn = config.use_flash_attn
472
+ self.is_fp32 = not (config.bf16 or config.fp16)
473
+ self.registered_causal_mask = None
474
+ # if (
475
+ # self.use_flash_attn
476
+ # and flash_attn_unpadded_func is not None
477
+ # and not self.is_fp32
478
+ # ):
479
+ # self.registered_causal_mask = None
480
+ # else:
481
+ # max_positions = config.max_position_embeddings
482
+ # self.register_buffer(
483
+ # "registered_causal_mask",
484
+ # torch.tril(
485
+ # torch.ones((max_positions, max_positions), dtype=torch.bool)
486
+ # ).view(1, 1, max_positions, max_positions),
487
+ # persistent=False,
488
+ # )
489
+
490
+ self.h = nn.ModuleList(
491
+ [
492
+ QWenBlock(
493
+ config
494
+ )
495
+ for i in range(config.num_hidden_layers)
496
+ ]
497
+ )
498
+ self.ln_f = RMSNorm(
499
+ self.embed_dim,
500
+ eps=config.layer_norm_epsilon,
501
+ )
502
+
503
+ self.visual = VisionTransformer(**config.visual)
504
+
505
+ self.post_init()
506
+
507
+ def get_input_embeddings(self):
508
+ return self.wte
509
+
510
+ def set_input_embeddings(self, new_embeddings):
511
+ self.wte = new_embeddings
512
+
513
+ # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
514
+ def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
515
+ # create causal mask
516
+ # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
517
+ combined_attention_mask = None
518
+ if input_shape[-1] > 1:
519
+ combined_attention_mask = _make_causal_mask(
520
+ input_shape,
521
+ inputs_embeds.dtype,
522
+ device=inputs_embeds.device,
523
+ past_key_values_length=past_key_values_length,
524
+ )
525
+
526
+ if attention_mask is not None:
527
+ # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
528
+ expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
529
+ inputs_embeds.device
530
+ )
531
+ combined_attention_mask = (
532
+ expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
533
+ )
534
+
535
+ return combined_attention_mask
536
+
537
+
538
+ def forward(
539
+ self,
540
+ input_ids: Optional[torch.LongTensor] = None,
541
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
542
+ attention_mask: Optional[torch.FloatTensor] = None,
543
+ token_type_ids: Optional[torch.LongTensor] = None,
544
+ position_ids: Optional[torch.LongTensor] = None,
545
+ head_mask: Optional[torch.FloatTensor] = None,
546
+ inputs_embeds: Optional[torch.FloatTensor] = None,
547
+ encoder_hidden_states: Optional[torch.Tensor] = None,
548
+ encoder_attention_mask: Optional[torch.FloatTensor] = None,
549
+ use_cache: Optional[bool] = None,
550
+ output_attentions: Optional[bool] = None,
551
+ output_hidden_states: Optional[bool] = None,
552
+ return_dict: Optional[bool] = None,
553
+ ):
554
+ if past_key_values is None and torch.any(input_ids == self.config.visual['image_start_id']):
555
+ bos_pos = torch.where(input_ids == self.config.visual['image_start_id'])
556
+ eos_pos = torch.where(input_ids == self.config.visual['image_start_id'] + 1)
557
+ assert (bos_pos[0] == eos_pos[0]).all()
558
+ img_pos = torch.stack((bos_pos[0], bos_pos[1], eos_pos[1]), dim=1)
559
+ images = []
560
+ for i, a, b in img_pos:
561
+ image = input_ids[i][a + 1 : b - 1].tolist()
562
+ image = image[ : image.index(self.config.visual['image_start_id'] + 2)]
563
+ images.append(bytes(image).decode('utf-8'))
564
+
565
+ images = self.visual.encode(images)
566
+ assert images.shape[0] == len(images)
567
+ fake_images = None
568
+ elif self.training:
569
+ fake_images=torch.zeros(1,3,224,224).to(
570
+ dtype=self.visual.conv1.weight.dtype, device=self.visual.conv1.weight.device)
571
+ images = self.visual(fake_images)
572
+ else:
573
+ fake_images = None
574
+ images = None
575
+
576
+ output_attentions = (
577
+ output_attentions
578
+ if output_attentions is not None
579
+ else self.config.output_attentions
580
+ )
581
+ output_hidden_states = (
582
+ output_hidden_states
583
+ if output_hidden_states is not None
584
+ else self.config.output_hidden_states
585
+ )
586
+ use_cache = use_cache if use_cache is not None else self.config.use_cache
587
+ return_dict = (
588
+ return_dict if return_dict is not None else self.config.use_return_dict
589
+ )
590
+
591
+ if input_ids is not None and inputs_embeds is not None:
592
+ raise ValueError(
593
+ "You cannot specify both input_ids and inputs_embeds at the same time"
594
+ )
595
+ elif input_ids is not None:
596
+ input_shape = input_ids.size()
597
+ input_ids = input_ids.view(-1, input_shape[-1])
598
+ batch_size = input_ids.shape[0]
599
+ elif inputs_embeds is not None:
600
+ input_shape = inputs_embeds.size()[:-1]
601
+ batch_size = inputs_embeds.shape[0]
602
+ else:
603
+ raise ValueError("You have to specify either input_ids or inputs_embeds")
604
+
605
+ device = input_ids.device if input_ids is not None else inputs_embeds.device
606
+
607
+ if token_type_ids is not None:
608
+ token_type_ids = token_type_ids.view(-1, input_shape[-1])
609
+ if position_ids is not None:
610
+ position_ids = position_ids.view(-1, input_shape[-1])
611
+
612
+ if past_key_values is None:
613
+ past_length = 0
614
+ past_key_values = tuple([None] * len(self.h))
615
+ else:
616
+ past_length = past_key_values[0][0].size(-2)
617
+
618
+ if position_ids is None:
619
+ position_ids = torch.arange(
620
+ past_length,
621
+ input_shape[-1] + past_length,
622
+ dtype=torch.long,
623
+ device=device,
624
+ )
625
+ position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
626
+
627
+ encoder_attention_mask = None
628
+ head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
629
+
630
+ if inputs_embeds is None:
631
+ inputs_embeds = self.wte(input_ids)
632
+
633
+ if batch_size <= 0:
634
+ raise ValueError("batch_size has to be defined and > 0")
635
+ attention_mask = self._prepare_decoder_attention_mask(
636
+ attention_mask, input_shape, inputs_embeds, past_length
637
+ )
638
+
639
+ hidden_states = inputs_embeds
640
+
641
+ kv_seq_len = hidden_states.size()[1]
642
+ if past_key_values[0] is not None:
643
+ # past key values[0][0] shape: bs * seq_len * head_num * dim
644
+ kv_seq_len += past_key_values[0][0].shape[1]
645
+ if (
646
+ self.use_dynamic_ntk
647
+ and kv_seq_len == hidden_states.size()[1]
648
+ and not self.training
649
+ ):
650
+ context_value = math.log(kv_seq_len / self.seq_length, 2) + 1
651
+ ntk_alpha = 2 ** math.ceil(context_value) - 1
652
+ ntk_alpha = max(ntk_alpha, 1)
653
+ else:
654
+ ntk_alpha = self.rotary_emb._ntk_alpha_cached
655
+
656
+ rotary_pos_emb = self.rotary_emb(kv_seq_len, ntk_alpha=ntk_alpha)
657
+ for idx in range(len(rotary_pos_emb)):
658
+ rotary_pos_emb[idx] = rotary_pos_emb[idx].to(hidden_states.device)
659
+
660
+ hidden_states = self.drop(hidden_states).clone()
661
+ if fake_images is not None:
662
+ hidden_states = hidden_states + images.mean()*0
663
+ elif images is not None:
664
+ for idx, (i, a, b) in enumerate(img_pos):
665
+ hidden_states[i][a + 1 : b] = images[idx]
666
+ output_shape = input_shape + (hidden_states.size(-1),)
667
+
668
+ if self.gradient_checkpointing and self.training:
669
+ if use_cache:
670
+ logger.warning_once(
671
+ "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
672
+ )
673
+ use_cache = False
674
+
675
+ presents = () if use_cache else None
676
+ all_self_attentions = () if output_attentions else None
677
+ all_hidden_states = () if output_hidden_states else None
678
+ for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
679
+
680
+ if output_hidden_states:
681
+ all_hidden_states = all_hidden_states + (hidden_states,)
682
+
683
+ if self.gradient_checkpointing and self.training:
684
+
685
+ def create_custom_forward(module):
686
+ def custom_forward(*inputs):
687
+ # None for past_key_value
688
+ return module(*inputs, use_cache, output_attentions)
689
+
690
+ return custom_forward
691
+
692
+ outputs = torch.utils.checkpoint.checkpoint(
693
+ create_custom_forward(block),
694
+ hidden_states,
695
+ rotary_pos_emb,
696
+ self.registered_causal_mask,
697
+ None,
698
+ attention_mask,
699
+ head_mask[i],
700
+ encoder_hidden_states,
701
+ encoder_attention_mask,
702
+ )
703
+ else:
704
+ outputs = block(
705
+ hidden_states,
706
+ layer_past=layer_past,
707
+ rotary_pos_emb=rotary_pos_emb,
708
+ registered_causal_mask=self.registered_causal_mask,
709
+ attention_mask=attention_mask,
710
+ head_mask=head_mask[i],
711
+ encoder_hidden_states=encoder_hidden_states,
712
+ encoder_attention_mask=encoder_attention_mask,
713
+ use_cache=use_cache,
714
+ output_attentions=output_attentions,
715
+ )
716
+
717
+ hidden_states = outputs[0]
718
+ if use_cache is True:
719
+ presents = presents + (outputs[1],)
720
+
721
+ if output_attentions:
722
+ all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
723
+
724
+ hidden_states = self.ln_f(hidden_states)
725
+ hidden_states = hidden_states.view(output_shape)
726
+ # Add last hidden state
727
+ if output_hidden_states:
728
+ all_hidden_states = all_hidden_states + (hidden_states,)
729
+
730
+ if not return_dict:
731
+ return tuple(
732
+ v for v in [hidden_states, presents, all_hidden_states] if v is not None
733
+ )
734
+
735
+ return BaseModelOutputWithPast(
736
+ last_hidden_state=hidden_states,
737
+ past_key_values=presents,
738
+ hidden_states=all_hidden_states,
739
+ attentions=all_self_attentions,
740
+ )
741
+
742
+
743
+ class QWenLMHeadModel(QWenPreTrainedModel):
744
+ _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.rotary_emb\.inv_freq"]
745
+ _keys_to_ignore_on_load_unexpected = [r"h\.\d+\.attn\.masked_bias"]
746
+
747
+ def __init__(self, config):
748
+ super().__init__(config)
749
+ assert (
750
+ config.bf16 + config.fp16 + config.fp32 <= 1
751
+ ), "Only one of \"bf16\", \"fp16\", \"fp32\" can be true"
752
+
753
+ autoset_precision = config.bf16 + config.fp16 + config.fp32 == 0
754
+
755
+ if autoset_precision:
756
+ if SUPPORT_BF16:
757
+ logger.warn(
758
+ "The model is automatically converting to bf16 for faster inference. "
759
+ "If you want to disable the automatic precision, please manually add bf16/fp16/fp32=True to \"AutoModelForCausalLM.from_pretrained\"."
760
+ )
761
+ config.bf16 = True
762
+ elif SUPPORT_FP16:
763
+ logger.warn(
764
+ "The model is automatically converting to fp16 for faster inference. "
765
+ "If you want to disable the automatic precision, please manually add bf16/fp16/fp32=True to \"AutoModelForCausalLM.from_pretrained\"."
766
+ )
767
+ config.fp16 = True
768
+ else:
769
+ config.fp32 = True
770
+
771
+ if config.bf16 and SUPPORT_CUDA and not SUPPORT_BF16:
772
+ logger.warn("Your device does NOT seem to support bf16, you can switch to fp16 or fp32 by by passing fp16/fp32=True in \"AutoModelForCausalLM.from_pretrained\".")
773
+ if config.fp16 and SUPPORT_CUDA and not SUPPORT_FP16:
774
+ logger.warn("Your device does NOT support faster inference with fp16, please switch to fp32 which is likely to be faster")
775
+ if config.fp32:
776
+ if SUPPORT_BF16:
777
+ logger.warn("Your device support faster inference by passing bf16=True in \"AutoModelForCausalLM.from_pretrained\".")
778
+ elif SUPPORT_FP16:
779
+ logger.warn("Your device support faster inference by passing fp16=True in \"AutoModelForCausalLM.from_pretrained\".")
780
+
781
+ self.transformer = QWenModel(config)
782
+ self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
783
+
784
+ if config.bf16:
785
+ self.transformer.bfloat16()
786
+ self.lm_head.bfloat16()
787
+ if config.fp16:
788
+ self.transformer.half()
789
+ self.lm_head.half()
790
+ self.post_init()
791
+
792
+ def get_output_embeddings(self):
793
+ return self.lm_head
794
+
795
+ def set_output_embeddings(self, new_embeddings):
796
+ self.lm_head = new_embeddings
797
+
798
+ def prepare_inputs_for_generation(
799
+ self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs
800
+ ):
801
+ token_type_ids = kwargs.get("token_type_ids", None)
802
+ if past_key_values:
803
+ input_ids = input_ids[:, -1].unsqueeze(-1)
804
+ if token_type_ids is not None:
805
+ token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
806
+
807
+ attention_mask = kwargs.get("attention_mask", None)
808
+ position_ids = kwargs.get("position_ids", None)
809
+
810
+ if attention_mask is not None and position_ids is None:
811
+ position_ids = attention_mask.long().cumsum(-1) - 1
812
+ position_ids.masked_fill_(attention_mask == 0, 1)
813
+ if past_key_values:
814
+ position_ids = position_ids[:, -1].unsqueeze(-1)
815
+ else:
816
+ position_ids = None
817
+
818
+ if inputs_embeds is not None and past_key_values is None:
819
+ model_inputs = {"inputs_embeds": inputs_embeds}
820
+ else:
821
+ model_inputs = {"input_ids": input_ids}
822
+
823
+ model_inputs.update(
824
+ {
825
+ "past_key_values": past_key_values,
826
+ "use_cache": kwargs.get("use_cache"),
827
+ "position_ids": position_ids,
828
+ "attention_mask": attention_mask,
829
+ "token_type_ids": token_type_ids,
830
+ }
831
+ )
832
+ return model_inputs
833
+
834
+ def forward(
835
+ self,
836
+ input_ids: Optional[torch.LongTensor] = None,
837
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
838
+ attention_mask: Optional[torch.FloatTensor] = None,
839
+ token_type_ids: Optional[torch.LongTensor] = None,
840
+ position_ids: Optional[torch.LongTensor] = None,
841
+ head_mask: Optional[torch.FloatTensor] = None,
842
+ inputs_embeds: Optional[torch.FloatTensor] = None,
843
+ encoder_hidden_states: Optional[torch.Tensor] = None,
844
+ encoder_attention_mask: Optional[torch.FloatTensor] = None,
845
+ labels: Optional[torch.LongTensor] = None,
846
+ use_cache: Optional[bool] = None,
847
+ output_attentions: Optional[bool] = None,
848
+ output_hidden_states: Optional[bool] = None,
849
+ return_dict: Optional[bool] = None,
850
+ ) -> Union[Tuple, CausalLMOutputWithPast]:
851
+
852
+ return_dict = (
853
+ return_dict if return_dict is not None else self.config.use_return_dict
854
+ )
855
+
856
+ transformer_outputs = self.transformer(
857
+ input_ids,
858
+ past_key_values=past_key_values,
859
+ attention_mask=attention_mask,
860
+ token_type_ids=token_type_ids,
861
+ position_ids=position_ids,
862
+ head_mask=head_mask,
863
+ inputs_embeds=inputs_embeds,
864
+ encoder_hidden_states=encoder_hidden_states,
865
+ encoder_attention_mask=encoder_attention_mask,
866
+ use_cache=use_cache,
867
+ output_attentions=output_attentions,
868
+ output_hidden_states=output_hidden_states,
869
+ return_dict=return_dict,
870
+ )
871
+ hidden_states = transformer_outputs[0]
872
+
873
+ lm_logits = self.lm_head(hidden_states)
874
+
875
+ loss = None
876
+ if labels is not None:
877
+ labels = labels.to(lm_logits.device)
878
+ shift_logits = lm_logits[..., :-1, :].contiguous()
879
+ shift_labels = labels[..., 1:].contiguous()
880
+ loss_fct = CrossEntropyLoss()
881
+ loss = loss_fct(
882
+ shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1)
883
+ )
884
+
885
+ if not return_dict:
886
+ output = (lm_logits,) + transformer_outputs[1:]
887
+ return ((loss,) + output) if loss is not None else output
888
+
889
+ return CausalLMOutputWithPast(
890
+ loss=loss,
891
+ logits=lm_logits,
892
+ past_key_values=transformer_outputs.past_key_values,
893
+ hidden_states=transformer_outputs.hidden_states,
894
+ attentions=transformer_outputs.attentions,
895
+ )
896
+
897
+ @staticmethod
898
+ def _reorder_cache(
899
+ past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
900
+ ) -> Tuple[Tuple[torch.Tensor]]:
901
+
902
+ return tuple(
903
+ tuple(
904
+ past_state.index_select(0, beam_idx.to(past_state.device))
905
+ for past_state in layer_past
906
+ )
907
+ for layer_past in past_key_values
908
+ )
909
+
910
+ def chat(
911
+ self,
912
+ tokenizer: PreTrainedTokenizer,
913
+ query: str,
914
+ history: Optional[HistoryType],
915
+ system: str = "You are a helpful assistant.",
916
+ append_history: bool = True,
917
+ stream: Optional[bool] = _SENTINEL,
918
+ stop_words_ids: Optional[List[List[int]]] = None,
919
+ generation_config: Optional[GenerationConfig] = None,
920
+ **kwargs,
921
+ ) -> Tuple[str, HistoryType]:
922
+ generation_config = generation_config if generation_config is not None else self.generation_config
923
+
924
+ assert stream is _SENTINEL, _ERROR_STREAM_IN_CHAT
925
+ assert generation_config.chat_format == 'chatml', _ERROR_BAD_CHAT_FORMAT
926
+ if history is None:
927
+ history = []
928
+ if stop_words_ids is None:
929
+ stop_words_ids = []
930
+
931
+ max_window_size = kwargs.get('max_window_size', None)
932
+ if max_window_size is None:
933
+ max_window_size = generation_config.max_window_size
934
+ raw_text, context_tokens = make_context(
935
+ tokenizer,
936
+ query,
937
+ history=history,
938
+ system=system,
939
+ max_window_size=max_window_size,
940
+ chat_format=generation_config.chat_format,
941
+ )
942
+
943
+ stop_words_ids.extend(get_stop_words_ids(
944
+ generation_config.chat_format, tokenizer
945
+ ))
946
+ input_ids = torch.tensor([context_tokens]).to(self.device)
947
+ outputs = self.generate(
948
+ input_ids,
949
+ stop_words_ids=stop_words_ids,
950
+ return_dict_in_generate=False,
951
+ generation_config=generation_config,
952
+ **kwargs,
953
+ )
954
+
955
+ response = decode_tokens(
956
+ outputs[0],
957
+ tokenizer,
958
+ raw_text_len=len(raw_text),
959
+ context_length=len(context_tokens),
960
+ chat_format=generation_config.chat_format,
961
+ verbose=False,
962
+ errors='replace'
963
+ )
964
+
965
+ if append_history:
966
+ history.append((query, response))
967
+
968
+ return response, history
969
+
970
+ def chat_stream(
971
+ self,
972
+ tokenizer: PreTrainedTokenizer,
973
+ query: str,
974
+ history: Optional[HistoryType],
975
+ system: str = "You are a helpful assistant.",
976
+ stop_words_ids: Optional[List[List[int]]] = None,
977
+ logits_processor: Optional[LogitsProcessorList] = None,
978
+ generation_config: Optional[GenerationConfig] = None,
979
+ **kwargs,
980
+ ) -> Generator[str, Any, None]:
981
+ generation_config = generation_config if generation_config is not None else self.generation_config
982
+ assert generation_config.chat_format == 'chatml', _ERROR_BAD_CHAT_FORMAT
983
+ if history is None:
984
+ history = []
985
+ if stop_words_ids is None:
986
+ stop_words_ids = []
987
+
988
+ max_window_size = kwargs.get('max_window_size', None)
989
+ if max_window_size is None:
990
+ max_window_size = generation_config.max_window_size
991
+ raw_text, context_tokens = make_context(
992
+ tokenizer,
993
+ query,
994
+ history=history,
995
+ system=system,
996
+ max_window_size=max_window_size,
997
+ chat_format=generation_config.chat_format,
998
+ )
999
+
1000
+ stop_words_ids.extend(get_stop_words_ids(
1001
+ generation_config.chat_format, tokenizer
1002
+ ))
1003
+ if stop_words_ids is not None:
1004
+ stop_words_logits_processor = StopWordsLogitsProcessor(
1005
+ stop_words_ids=stop_words_ids,
1006
+ eos_token_id=generation_config.eos_token_id,
1007
+ )
1008
+ if logits_processor is None:
1009
+ logits_processor = LogitsProcessorList([stop_words_logits_processor])
1010
+ else:
1011
+ logits_processor.append(stop_words_logits_processor)
1012
+ input_ids = torch.tensor([context_tokens]).to(self.device)
1013
+
1014
+ from transformers_stream_generator.main import NewGenerationMixin, StreamGenerationConfig
1015
+ self.__class__.generate_stream = NewGenerationMixin.generate
1016
+ self.__class__.sample_stream = NewGenerationMixin.sample_stream
1017
+ stream_config = StreamGenerationConfig(**generation_config.to_dict(), do_stream=True)
1018
+
1019
+ def stream_generator():
1020
+ outputs = []
1021
+ for token in self.generate_stream(
1022
+ input_ids,
1023
+ return_dict_in_generate=False,
1024
+ generation_config=stream_config,
1025
+ logits_processor=logits_processor,
1026
+ seed=-1,
1027
+ **kwargs):
1028
+ outputs.append(token.item())
1029
+ yield tokenizer.decode(outputs, skip_special_tokens=True, errors='ignore', keep_image_special=True)
1030
+
1031
+ return stream_generator()
1032
+
1033
+ def generate(
1034
+ self,
1035
+ inputs: Optional[torch.Tensor] = None,
1036
+ generation_config: Optional[GenerationConfig] = None,
1037
+ logits_processor: Optional[LogitsProcessorList] = None,
1038
+ stopping_criteria: Optional[StoppingCriteriaList] = None,
1039
+ prefix_allowed_tokens_fn: Optional[
1040
+ Callable[[int, torch.Tensor], List[int]]
1041
+ ] = None,
1042
+ synced_gpus: Optional[bool] = None,
1043
+ assistant_model: Optional["PreTrainedModel"] = None,
1044
+ streamer: Optional["BaseStreamer"] = None,
1045
+ **kwargs,
1046
+ ) -> Union[GenerateOutput, torch.LongTensor]:
1047
+ generation_config = generation_config if generation_config is not None else self.generation_config
1048
+
1049
+ # Process stop_words_ids.
1050
+ stop_words_ids = kwargs.pop("stop_words_ids", None)
1051
+ if stop_words_ids is None and generation_config is not None:
1052
+ stop_words_ids = getattr(generation_config, "stop_words_ids", None)
1053
+ if stop_words_ids is None:
1054
+ stop_words_ids = getattr(generation_config, "stop_words_ids", None)
1055
+
1056
+ if stop_words_ids is not None:
1057
+ stop_words_logits_processor = StopWordsLogitsProcessor(
1058
+ stop_words_ids=stop_words_ids,
1059
+ eos_token_id=generation_config.eos_token_id,
1060
+ )
1061
+ if logits_processor is None:
1062
+ logits_processor = LogitsProcessorList([stop_words_logits_processor])
1063
+ else:
1064
+ logits_processor.append(stop_words_logits_processor)
1065
+
1066
+ return super().generate(
1067
+ inputs,
1068
+ generation_config=generation_config,
1069
+ logits_processor=logits_processor,
1070
+ stopping_criteria=stopping_criteria,
1071
+ prefix_allowed_tokens_fn=prefix_allowed_tokens_fn,
1072
+ synced_gpus=synced_gpus,
1073
+ assistant_model=assistant_model,
1074
+ streamer=streamer,
1075
+ **kwargs,
1076
+ )
1077
+
1078
+
1079
+ class RotaryEmbedding(torch.nn.Module):
1080
+ def __init__(self, dim, base=10000):
1081
+ super().__init__()
1082
+ self.dim = dim
1083
+ self.base = base
1084
+ self.inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
1085
+ if importlib.util.find_spec("einops") is None:
1086
+ raise RuntimeError("einops is required for Rotary Embedding")
1087
+
1088
+ self._rotary_pos_emb_cache = None
1089
+ self._seq_len_cached = 0
1090
+ self._ntk_alpha_cached = 1.0
1091
+
1092
+ def update_rotary_pos_emb_cache(self, max_seq_len, offset=0, ntk_alpha=1.0):
1093
+ seqlen = max_seq_len + offset
1094
+ if seqlen > self._seq_len_cached or ntk_alpha != self._ntk_alpha_cached:
1095
+ base = self.base * ntk_alpha ** (self.dim / (self.dim - 2))
1096
+ self.inv_freq = 1.0 / (
1097
+ base
1098
+ ** (
1099
+ torch.arange(0, self.dim, 2, device=self.inv_freq.device).float()
1100
+ / self.dim
1101
+ )
1102
+ )
1103
+ self._seq_len_cached = max(2 * seqlen, 16)
1104
+ self._ntk_alpha_cached = ntk_alpha
1105
+ seq = torch.arange(self._seq_len_cached, device=self.inv_freq.device)
1106
+ freqs = torch.outer(seq.type_as(self.inv_freq), self.inv_freq)
1107
+
1108
+ emb = torch.cat((freqs, freqs), dim=-1)
1109
+ from einops import rearrange
1110
+
1111
+ emb = rearrange(emb, "n d -> 1 n 1 d")
1112
+
1113
+ cos, sin = emb.cos(), emb.sin()
1114
+ self._rotary_pos_emb_cache = [cos, sin]
1115
+
1116
+ def forward(self, max_seq_len, offset=0, ntk_alpha=1.0):
1117
+ self.update_rotary_pos_emb_cache(max_seq_len, offset, ntk_alpha)
1118
+ cos, sin = self._rotary_pos_emb_cache
1119
+ return [cos[:, offset : offset + max_seq_len], sin[:, offset : offset + max_seq_len]]
1120
+
1121
+
1122
+ def _rotate_half(x):
1123
+ from einops import rearrange
1124
+
1125
+ x = rearrange(x, "... (j d) -> ... j d", j=2)
1126
+ x1, x2 = x.unbind(dim=-2)
1127
+ return torch.cat((-x2, x1), dim=-1)
1128
+
1129
+
1130
+ def apply_rotary_pos_emb(t, freqs):
1131
+ cos, sin = freqs
1132
+ if apply_rotary_emb_func is not None and t.is_cuda:
1133
+ t_ = t.float()
1134
+ cos = cos.squeeze(0).squeeze(1)[:, : cos.shape[-1] // 2]
1135
+ sin = sin.squeeze(0).squeeze(1)[:, : sin.shape[-1] // 2]
1136
+ output = apply_rotary_emb_func(t_, cos, sin).type_as(t)
1137
+ return output
1138
+ else:
1139
+ rot_dim = freqs[0].shape[-1]
1140
+ cos, sin = freqs
1141
+ t_, t_pass_ = t[..., :rot_dim], t[..., rot_dim:]
1142
+ t_ = t_.float()
1143
+ t_pass_ = t_pass_.float()
1144
+ t_ = (t_ * cos) + (_rotate_half(t_) * sin)
1145
+ return torch.cat((t_, t_pass_), dim=-1).type_as(t)
1146
+
1147
+
1148
+ class RMSNorm(torch.nn.Module):
1149
+ def __init__(self, dim: int, eps: float = 1e-6):
1150
+ super().__init__()
1151
+ self.eps = eps
1152
+ self.weight = nn.Parameter(torch.ones(dim))
1153
+
1154
+ def _norm(self, x):
1155
+ return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
1156
+
1157
+ def forward(self, x):
1158
+ if rms_norm is not None and x.is_cuda:
1159
+ return rms_norm(x, self.weight, self.eps)
1160
+ else:
1161
+ output = self._norm(x.float()).type_as(x)
1162
+ return output * self.weight
qwen.tiktoken ADDED
The diff for this file is too large to render. See raw diff
 
qwen_generation_utils.py ADDED
@@ -0,0 +1,420 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) Alibaba Cloud.
2
+ #
3
+ # This source code is licensed under the license found in the
4
+ # LICENSE file in the root directory of this source tree.
5
+
6
+ """Generation support."""
7
+
8
+ from typing import Tuple, List, Union, Iterable
9
+
10
+ import numpy as np
11
+ import torch
12
+ import torch.nn.functional as F
13
+ from transformers import PreTrainedTokenizer
14
+ from transformers import logging
15
+ from transformers.generation import LogitsProcessor
16
+
17
+ logger = logging.get_logger(__name__)
18
+
19
+ # Types.
20
+ HistoryType = List[Tuple[str, str]]
21
+ TokensType = List[int]
22
+ BatchTokensType = List[List[int]]
23
+
24
+
25
+ def pad_batch(batch: BatchTokensType, pad_id: int, seq_length: int) -> BatchTokensType:
26
+ for tokens in batch:
27
+ context_length = len(tokens)
28
+ if context_length < seq_length:
29
+ tokens.extend([pad_id] * (seq_length - context_length))
30
+ return batch
31
+
32
+
33
+ def get_ltor_masks_and_position_ids(
34
+ data,
35
+ eod_token,
36
+ reset_position_ids,
37
+ reset_attention_mask,
38
+ eod_mask_loss,
39
+ ):
40
+ """Build masks and position id for left to right model."""
41
+
42
+ # Extract batch size and sequence length.
43
+ micro_batch_size, seq_length = data.size()
44
+
45
+ # Attention mask (lower triangular).
46
+ if reset_attention_mask:
47
+ att_mask_batch = micro_batch_size
48
+ else:
49
+ att_mask_batch = 1
50
+ attention_mask = torch.tril(
51
+ torch.ones((att_mask_batch, seq_length, seq_length), device=data.device)
52
+ ).view(att_mask_batch, 1, seq_length, seq_length)
53
+
54
+ # Loss mask.
55
+ loss_mask = torch.ones(data.size(), dtype=torch.float, device=data.device)
56
+ if eod_mask_loss:
57
+ loss_mask[data == eod_token] = 0.0
58
+
59
+ # Position ids.
60
+ position_ids = torch.arange(seq_length, dtype=torch.long, device=data.device)
61
+ position_ids = position_ids.unsqueeze(0).expand_as(data)
62
+ # We need to clone as the ids will be modifed based on batch index.
63
+ if reset_position_ids:
64
+ position_ids = position_ids.clone()
65
+
66
+ if reset_position_ids or reset_attention_mask:
67
+ # Loop through the batches:
68
+ for b in range(micro_batch_size):
69
+
70
+ # Find indecies where EOD token is.
71
+ eod_index = position_ids[b, data[b] == eod_token]
72
+ # Detach indecies from positions if going to modify positions.
73
+ if reset_position_ids:
74
+ eod_index = eod_index.clone()
75
+
76
+ # Loop through EOD indecies:
77
+ prev_index = 0
78
+ for j in range(eod_index.size()[0]):
79
+ i = eod_index[j]
80
+ # Mask attention loss.
81
+ if reset_attention_mask:
82
+ attention_mask[b, 0, (i + 1) :, : (i + 1)] = 0
83
+ # Reset positions.
84
+ if reset_position_ids:
85
+ position_ids[b, (i + 1) :] -= i + 1 - prev_index
86
+ prev_index = i + 1
87
+
88
+ # Convert attention mask to binary:
89
+ attention_mask = attention_mask < 0.5
90
+
91
+ return attention_mask, loss_mask, position_ids
92
+
93
+
94
+ def get_batch(context_tokens: torch.LongTensor, eod_id: int):
95
+ """Generate batch from context tokens."""
96
+ # Move to GPU.
97
+ tokens = context_tokens.contiguous().to(context_tokens.device)
98
+ # Get the attention mask and postition ids.
99
+ attention_mask, _, position_ids = get_ltor_masks_and_position_ids(
100
+ tokens,
101
+ eod_id,
102
+ reset_position_ids=False,
103
+ reset_attention_mask=False,
104
+ eod_mask_loss=False,
105
+ )
106
+ return tokens, attention_mask, position_ids
107
+
108
+
109
+ def get_stop_words_ids(chat_format, tokenizer):
110
+ if chat_format == "raw":
111
+ stop_words_ids = [tokenizer.encode("Human:"), [tokenizer.eod_id]]
112
+ elif chat_format == "chatml":
113
+ stop_words_ids = [[tokenizer.im_end_id], [tokenizer.im_start_id]]
114
+ else:
115
+ raise NotImplementedError(f"Unknown chat format {chat_format!r}")
116
+ return stop_words_ids
117
+
118
+
119
+ def make_context(
120
+ tokenizer: PreTrainedTokenizer,
121
+ query: str,
122
+ history: List[Tuple[str, str]] = None,
123
+ system: str = "",
124
+ max_window_size: int = 6144,
125
+ chat_format: str = "chatml",
126
+ ):
127
+ if history is None:
128
+ history = []
129
+
130
+ if chat_format == "chatml":
131
+ im_start, im_end = "<|im_start|>", "<|im_end|>"
132
+ im_start_tokens = [tokenizer.im_start_id]
133
+ im_end_tokens = [tokenizer.im_end_id]
134
+ nl_tokens = tokenizer.encode("\n")
135
+
136
+ def _tokenize_str(role, content):
137
+ return f"{role}\n{content}", tokenizer.encode(
138
+ role, allowed_special=set(tokenizer.IMAGE_ST)
139
+ ) + nl_tokens + tokenizer.encode(content, allowed_special=set(tokenizer.IMAGE_ST))
140
+
141
+ system_text, system_tokens_part = _tokenize_str("system", system)
142
+ system_tokens = im_start_tokens + system_tokens_part + im_end_tokens
143
+
144
+ raw_text = ""
145
+ context_tokens = []
146
+
147
+ for turn_query, turn_response in reversed(history):
148
+ query_text, query_tokens_part = _tokenize_str("user", turn_query)
149
+ query_tokens = im_start_tokens + query_tokens_part + im_end_tokens
150
+ if turn_response is not None:
151
+ response_text, response_tokens_part = _tokenize_str(
152
+ "assistant", turn_response
153
+ )
154
+ response_tokens = im_start_tokens + response_tokens_part + im_end_tokens
155
+
156
+ next_context_tokens = nl_tokens + query_tokens + nl_tokens + response_tokens
157
+ prev_chat = (
158
+ f"\n{im_start}{query_text}{im_end}\n{im_start}{response_text}{im_end}"
159
+ )
160
+ else:
161
+ next_context_tokens = nl_tokens + query_tokens + nl_tokens
162
+ prev_chat = f"\n{im_start}{query_text}{im_end}\n"
163
+
164
+ current_context_size = (
165
+ len(system_tokens) + len(next_context_tokens) + len(context_tokens)
166
+ )
167
+ if current_context_size < max_window_size:
168
+ context_tokens = next_context_tokens + context_tokens
169
+ raw_text = prev_chat + raw_text
170
+ else:
171
+ break
172
+
173
+ context_tokens = system_tokens + context_tokens
174
+ raw_text = f"{im_start}{system_text}{im_end}" + raw_text
175
+ context_tokens += (
176
+ nl_tokens
177
+ + im_start_tokens
178
+ + _tokenize_str("user", query)[1]
179
+ + im_end_tokens
180
+ + nl_tokens
181
+ + im_start_tokens
182
+ + tokenizer.encode("assistant")
183
+ + nl_tokens
184
+ )
185
+ raw_text += f"\n{im_start}user\n{query}{im_end}\n{im_start}assistant\n"
186
+
187
+ elif chat_format == "raw":
188
+ raw_text = query
189
+ context_tokens = tokenizer.encode(raw_text)
190
+ else:
191
+ raise NotImplementedError(f"Unknown chat format {chat_format!r}")
192
+
193
+ return raw_text, context_tokens
194
+
195
+
196
+ def _decode_default(
197
+ tokens: List[int],
198
+ *,
199
+ stop_words: List[str],
200
+ eod_words: List[str],
201
+ tokenizer: PreTrainedTokenizer,
202
+ raw_text_len: int,
203
+ verbose: bool = False,
204
+ return_end_reason: bool = False,
205
+ errors: str='replace',
206
+ ):
207
+ trim_decode_tokens = tokenizer.decode(tokens, errors=errors)[raw_text_len:]
208
+ if verbose:
209
+ print("\nRaw Generate: ", trim_decode_tokens)
210
+
211
+ end_reason = f"Gen length {len(tokens)}"
212
+ for stop_word in stop_words:
213
+ trim_decode_tokens = trim_decode_tokens.replace(stop_word, "").strip()
214
+ for eod_word in eod_words:
215
+ if eod_word in trim_decode_tokens:
216
+ end_reason = f"Gen {eod_word!r}"
217
+ trim_decode_tokens = trim_decode_tokens.split(eod_word)[0]
218
+ trim_decode_tokens = trim_decode_tokens.strip()
219
+ if verbose:
220
+ print("\nEnd Reason:", end_reason)
221
+ print("\nGenerate: ", trim_decode_tokens)
222
+
223
+ if return_end_reason:
224
+ return trim_decode_tokens, end_reason
225
+ else:
226
+ return trim_decode_tokens
227
+
228
+
229
+ def _decode_chatml(
230
+ tokens: List[int],
231
+ *,
232
+ stop_words: List[str],
233
+ eod_token_ids: List[int],
234
+ tokenizer: PreTrainedTokenizer,
235
+ raw_text_len: int,
236
+ context_length: int,
237
+ verbose: bool = False,
238
+ return_end_reason: bool = False,
239
+ errors: str='replace'
240
+ ):
241
+ end_reason = f"Gen length {len(tokens)}"
242
+ eod_token_idx = context_length
243
+ for eod_token_idx in range(context_length, len(tokens)):
244
+ if tokens[eod_token_idx] in eod_token_ids:
245
+ end_reason = f"Gen {tokenizer.decode([tokens[eod_token_idx]])!r}"
246
+ break
247
+
248
+ trim_decode_tokens = tokenizer.decode(tokens[:eod_token_idx], errors=errors)[raw_text_len:]
249
+ if verbose:
250
+ print("\nRaw Generate w/o EOD:", tokenizer.decode(tokens, errors=errors)[raw_text_len:])
251
+ print("\nRaw Generate:", trim_decode_tokens)
252
+ print("\nEnd Reason:", end_reason)
253
+ for stop_word in stop_words:
254
+ trim_decode_tokens = trim_decode_tokens.replace(stop_word, "").strip()
255
+ trim_decode_tokens = trim_decode_tokens.strip()
256
+ if verbose:
257
+ print("\nGenerate:", trim_decode_tokens)
258
+
259
+ if return_end_reason:
260
+ return trim_decode_tokens, end_reason
261
+ else:
262
+ return trim_decode_tokens
263
+
264
+
265
+ def decode_tokens(
266
+ tokens: Union[torch.LongTensor, TokensType],
267
+ tokenizer: PreTrainedTokenizer,
268
+ raw_text_len: int,
269
+ context_length: int,
270
+ chat_format: str,
271
+ verbose: bool = False,
272
+ return_end_reason: bool = False,
273
+ errors: str="replace",
274
+ ) -> str:
275
+ if torch.is_tensor(tokens):
276
+ tokens = tokens.cpu().numpy().tolist()
277
+
278
+ if chat_format == "chatml":
279
+ return _decode_chatml(
280
+ tokens,
281
+ stop_words=[],
282
+ eod_token_ids=[tokenizer.im_start_id, tokenizer.im_end_id],
283
+ tokenizer=tokenizer,
284
+ raw_text_len=raw_text_len,
285
+ context_length=context_length,
286
+ verbose=verbose,
287
+ return_end_reason=return_end_reason,
288
+ errors=errors,
289
+ )
290
+ elif chat_format == "raw":
291
+ return _decode_default(
292
+ tokens,
293
+ stop_words=["<|endoftext|>"],
294
+ eod_words=["<|endoftext|>"],
295
+ tokenizer=tokenizer,
296
+ raw_text_len=raw_text_len,
297
+ verbose=verbose,
298
+ return_end_reason=return_end_reason,
299
+ errors=errors,
300
+ )
301
+ else:
302
+ raise NotImplementedError(f"Unknown chat format {chat_format!r}")
303
+
304
+
305
+ class StopWordsLogitsProcessor(LogitsProcessor):
306
+ """
307
+ :class:`transformers.LogitsProcessor` that enforces that when specified sequences appear, stop geration.
308
+
309
+ Args:
310
+ stop_words_ids (:obj:`List[List[int]]`):
311
+ List of list of token ids of stop ids. In order to get the tokens of the words
312
+ that should not appear in the generated text, use :obj:`tokenizer(bad_word,
313
+ add_prefix_space=True).input_ids`.
314
+ eos_token_id (:obj:`int`):
315
+ The id of the `end-of-sequence` token.
316
+ """
317
+
318
+ def __init__(self, stop_words_ids: Iterable[Iterable[int]], eos_token_id: int):
319
+
320
+ if not isinstance(stop_words_ids, List) or len(stop_words_ids) == 0:
321
+ raise ValueError(
322
+ f"`stop_words_ids` has to be a non-emtpy list, but is {stop_words_ids}."
323
+ )
324
+ if any(not isinstance(bad_word_ids, list) for bad_word_ids in stop_words_ids):
325
+ raise ValueError(
326
+ f"`stop_words_ids` has to be a list of lists, but is {stop_words_ids}."
327
+ )
328
+ if any(
329
+ any(
330
+ (not isinstance(token_id, (int, np.integer)) or token_id < 0)
331
+ for token_id in stop_word_ids
332
+ )
333
+ for stop_word_ids in stop_words_ids
334
+ ):
335
+ raise ValueError(
336
+ f"Each list in `stop_words_ids` has to be a list of positive integers, but is {stop_words_ids}."
337
+ )
338
+
339
+ self.stop_words_ids = list(
340
+ filter(
341
+ lambda bad_token_seq: bad_token_seq != [eos_token_id], stop_words_ids
342
+ )
343
+ )
344
+ self.eos_token_id = eos_token_id
345
+ for stop_token_seq in self.stop_words_ids:
346
+ assert (
347
+ len(stop_token_seq) > 0
348
+ ), "Stop words token sequences {} cannot have an empty list".format(
349
+ stop_words_ids
350
+ )
351
+
352
+ def __call__(
353
+ self, input_ids: torch.LongTensor, scores: torch.FloatTensor
354
+ ) -> torch.FloatTensor:
355
+ stopped_samples = self._calc_stopped_samples(input_ids)
356
+ for i, should_stop in enumerate(stopped_samples):
357
+ if should_stop:
358
+ scores[i, self.eos_token_id] = float(2**15)
359
+ return scores
360
+
361
+ def _tokens_match(self, prev_tokens: torch.LongTensor, tokens: List[int]) -> bool:
362
+ if len(tokens) == 0:
363
+ # if bad word tokens is just one token always ban it
364
+ return True
365
+ elif len(tokens) > len(prev_tokens):
366
+ # if bad word tokens are longer then prev input_ids they can't be equal
367
+ return False
368
+ elif prev_tokens[-len(tokens) :].tolist() == tokens:
369
+ # if tokens match
370
+ return True
371
+ else:
372
+ return False
373
+
374
+ def _calc_stopped_samples(self, prev_input_ids: Iterable[int]) -> Iterable[int]:
375
+ stopped_samples = []
376
+ for prev_input_ids_slice in prev_input_ids:
377
+ match = False
378
+ for stop_token_seq in self.stop_words_ids:
379
+ if self._tokens_match(prev_input_ids_slice, stop_token_seq):
380
+ # if tokens do not match continue
381
+ match = True
382
+ break
383
+ stopped_samples.append(match)
384
+
385
+ return stopped_samples
386
+
387
+
388
+ def top_k_logits(logits, top_k=0, top_p=0.0, filter_value=-float("Inf")):
389
+ """This function has been mostly taken from huggingface conversational
390
+ ai code at
391
+ https://medium.com/huggingface/how-to-build-a-state-of-the-art-
392
+ conversational-ai-with-transfer-learning-2d818ac26313"""
393
+
394
+ if top_k > 0:
395
+ # Remove all tokens with a probability less than the
396
+ # last token of the top-k
397
+ indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
398
+ logits[indices_to_remove] = filter_value
399
+
400
+ if top_p > 0.0:
401
+ # Cconvert to 1D
402
+ sorted_logits, sorted_indices = torch.sort(logits, descending=True, dim=-1)
403
+ cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
404
+
405
+ # Remove tokens with cumulative probability above the threshold
406
+ sorted_indices_to_remove = cumulative_probs > top_p
407
+ # Shift the indices to the right to keep also the first token
408
+ # above the threshold
409
+ sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
410
+ sorted_indices_to_remove[..., 0] = 0
411
+ for i in range(sorted_indices.size(0)):
412
+ indices_to_remove = sorted_indices[i][sorted_indices_to_remove[i]]
413
+ logits[i][indices_to_remove] = filter_value
414
+
415
+ return logits
416
+
417
+
418
+ def switch(val1, val2, boolean):
419
+ boolean = boolean.type_as(val1)
420
+ return (1 - boolean) * val1 + boolean * val2
tokenization_qwen.py ADDED
@@ -0,0 +1,598 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) Alibaba Cloud.
2
+ #
3
+ # This source code is licensed under the license found in the
4
+ # LICENSE file in the root directory of this source tree.
5
+
6
+ """Tokenization classes for QWen."""
7
+
8
+ import base64
9
+ import logging
10
+ import os
11
+ import requests
12
+ import unicodedata
13
+ from typing import Collection, Dict, List, Set, Tuple, Union, Any, Callable, Optional
14
+
15
+ import tiktoken
16
+ import numpy as np
17
+ from PIL import Image
18
+ from PIL import ImageFont
19
+ from PIL import ImageDraw
20
+ from transformers import PreTrainedTokenizer, AddedToken
21
+ from transformers.utils import try_to_load_from_cache
22
+
23
+ import matplotlib.colors as mcolors
24
+ from matplotlib.font_manager import FontProperties
25
+
26
+ logger = logging.getLogger(__name__)
27
+
28
+
29
+ VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
30
+ FONT_PATH = try_to_load_from_cache("Qwen/Qwen-VL-Chat", "SimSun.ttf")
31
+ if FONT_PATH is None:
32
+ if not os.path.exists("SimSun.ttf"):
33
+ ttf = requests.get("https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/SimSun.ttf")
34
+ open("SimSun.ttf", "wb").write(ttf.content)
35
+ FONT_PATH = "SimSun.ttf"
36
+
37
+ PAT_STR = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""
38
+ ENDOFTEXT = "<|endoftext|>"
39
+ IMSTART = "<|im_start|>"
40
+ IMEND = "<|im_end|>"
41
+ # as the default behavior is changed to allow special tokens in
42
+ # regular texts, the surface forms of special tokens need to be
43
+ # as different as possible to minimize the impact
44
+ EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
45
+ SPECIAL_TOKENS = (
46
+ ENDOFTEXT,
47
+ IMSTART,
48
+ IMEND,
49
+ ) + EXTRAS
50
+ IMG_TOKEN_SPAN = 256
51
+
52
+
53
+ def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
54
+ with open(tiktoken_bpe_file, "rb") as f:
55
+ contents = f.read()
56
+ return {
57
+ base64.b64decode(token): int(rank)
58
+ for token, rank in (line.split() for line in contents.splitlines() if line)
59
+ }
60
+
61
+ def _list_find(
62
+ input_list: List[Any],
63
+ candidates: Tuple[Any],
64
+ start: int = 0,
65
+ ):
66
+ for i in range(start, len(input_list)):
67
+ if input_list[i] in candidates:
68
+ return i
69
+ return -1
70
+
71
+ def _replace_closed_tag(
72
+ input_tokens: List[Any],
73
+ start_tags: Union[Any, Tuple[Any]],
74
+ end_tags: Union[Any, Tuple[Any]],
75
+ inclusive_replace_func: Callable,
76
+ exclusive_replace_func: Callable = lambda x: x,
77
+ ):
78
+ if isinstance(start_tags, (str, int)):
79
+ start_tags = (start_tags,)
80
+ if isinstance(end_tags, (str, int)):
81
+ end_tags = (end_tags,)
82
+ assert len(start_tags) == len(end_tags)
83
+
84
+ output_tokens = []
85
+ end = 0
86
+ while True:
87
+ start = _list_find(input_tokens, start_tags, end)
88
+ if start == -1:
89
+ break
90
+ output_tokens.extend(exclusive_replace_func(input_tokens[end : start]))
91
+ tag_idx = start_tags.index(input_tokens[start])
92
+ end = _list_find(input_tokens, (end_tags[tag_idx],), start)
93
+ if end == -1:
94
+ raise ValueError("Unclosed image token")
95
+ output_tokens.extend(inclusive_replace_func(input_tokens[start : end + 1]))
96
+ end += 1
97
+ output_tokens.extend(exclusive_replace_func(input_tokens[end : ]))
98
+ return output_tokens
99
+
100
+ class QWenTokenizer(PreTrainedTokenizer):
101
+ """QWen tokenizer."""
102
+
103
+ vocab_files_names = VOCAB_FILES_NAMES
104
+
105
+ def __init__(
106
+ self,
107
+ vocab_file,
108
+ errors="replace",
109
+ image_start_tag='<img>',
110
+ image_end_tag='</img>',
111
+ image_pad_tag='<imgpad>',
112
+ ref_start_tag='<ref>',
113
+ ref_end_tag='</ref>',
114
+ box_start_tag='<box>',
115
+ box_end_tag='</box>',
116
+ quad_start_tag='<quad>',
117
+ quad_end_tag='</quad>',
118
+ **kwargs,
119
+ ):
120
+ super().__init__(**kwargs)
121
+ self.image_start_tag = image_start_tag
122
+ self.image_end_tag = image_end_tag
123
+ self.image_pad_tag = image_pad_tag
124
+ self.ref_start_tag = ref_start_tag
125
+ self.ref_end_tag = ref_end_tag
126
+ self.box_start_tag = box_start_tag
127
+ self.box_end_tag = box_end_tag
128
+ self.quad_start_tag = quad_start_tag
129
+ self.quad_end_tag = quad_end_tag
130
+ self.IMAGE_ST = (
131
+ ref_start_tag, ref_end_tag,
132
+ box_start_tag, box_end_tag,
133
+ quad_start_tag, quad_end_tag,
134
+ image_start_tag, image_end_tag,
135
+ image_pad_tag
136
+ )
137
+
138
+ self.errors = errors # how to handle errors in decoding
139
+
140
+ self.mergeable_ranks = _load_tiktoken_bpe(vocab_file) # type: dict[bytes, int]
141
+ self.special_tokens = {
142
+ token: index
143
+ for index, token in enumerate(
144
+ SPECIAL_TOKENS + self.IMAGE_ST, start=len(self.mergeable_ranks)
145
+ )
146
+ }
147
+ self.img_start_id = self.special_tokens[self.image_start_tag]
148
+ self.img_end_id = self.special_tokens[self.image_end_tag]
149
+ self.img_pad_id = self.special_tokens[self.image_pad_tag]
150
+ self.ref_start_id = self.special_tokens[self.ref_start_tag]
151
+ self.ref_end_id = self.special_tokens[self.ref_end_tag]
152
+ self.box_start_id = self.special_tokens[self.box_start_tag]
153
+ self.box_end_id = self.special_tokens[self.box_end_tag]
154
+ self.quad_start_id = self.special_tokens[self.quad_start_tag]
155
+ self.quad_end_id = self.special_tokens[self.quad_end_tag]
156
+ self.image_special_tokens = set([
157
+ self.ref_start_id, self.ref_end_id, self.box_start_id, self.box_end_id,
158
+ self.quad_start_id, self.quad_end_id,
159
+ ])
160
+
161
+ enc = tiktoken.Encoding(
162
+ "Qwen",
163
+ pat_str=PAT_STR,
164
+ mergeable_ranks=self.mergeable_ranks,
165
+ special_tokens=self.special_tokens,
166
+ )
167
+ assert (
168
+ len(self.mergeable_ranks) + len(self.special_tokens) == enc.n_vocab
169
+ ), f"{len(self.mergeable_ranks) + len(self.special_tokens)} != {enc.n_vocab} in encoding"
170
+
171
+ self.decoder = {
172
+ v: k for k, v in self.mergeable_ranks.items()
173
+ } # type: dict[int, bytes|str]
174
+ self.decoder.update({v: k for k, v in self.special_tokens.items()})
175
+
176
+ self.tokenizer = enc # type: tiktoken.Encoding
177
+
178
+ self.eod_id = self.tokenizer.eot_token
179
+ self.im_start_id = self.special_tokens[IMSTART]
180
+ self.im_end_id = self.special_tokens[IMEND]
181
+
182
+ def __getstate__(self):
183
+ # for pickle lovers
184
+ state = self.__dict__.copy()
185
+ del state['tokenizer']
186
+ return state
187
+
188
+ def __setstate__(self, state):
189
+ # tokenizer is not python native; don't pass it; rebuild it
190
+ self.__dict__.update(state)
191
+ enc = tiktoken.Encoding(
192
+ "Qwen",
193
+ pat_str=PAT_STR,
194
+ mergeable_ranks=self.mergeable_ranks,
195
+ special_tokens=self.special_tokens,
196
+ )
197
+ self.tokenizer = enc
198
+
199
+
200
+ def __len__(self) -> int:
201
+ return self.tokenizer.n_vocab
202
+
203
+ def get_vocab(self) -> Dict[bytes, int]:
204
+ return self.mergeable_ranks
205
+
206
+ def convert_tokens_to_ids(
207
+ self, tokens: Union[bytes, str, List[Union[bytes, str]]]
208
+ ) -> List[int]:
209
+ ids = []
210
+ if isinstance(tokens, (str, bytes)):
211
+ if tokens in self.special_tokens:
212
+ return self.special_tokens[tokens]
213
+ else:
214
+ return self.mergeable_ranks.get(tokens)
215
+ for token in tokens:
216
+ if token in self.special_tokens:
217
+ ids.append(self.special_tokens[token])
218
+ else:
219
+ ids.append(self.mergeable_ranks.get(token))
220
+ return ids
221
+
222
+ def _add_tokens(self, new_tokens: Union[List[str], List[AddedToken]], special_tokens: bool = False) -> int:
223
+ if not special_tokens and new_tokens:
224
+ raise ValueError('Adding regular tokens is not supported')
225
+ for token in new_tokens:
226
+ surface_form = token.content if isinstance(token, AddedToken) else token
227
+ if surface_form not in SPECIAL_TOKENS + self.IMAGE_ST:
228
+ raise ValueError('Adding unknown special tokens is not supported')
229
+ return 0
230
+
231
+ def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
232
+ """
233
+ Save only the vocabulary of the tokenizer (vocabulary).
234
+
235
+ Returns:
236
+ `Tuple(str)`: Paths to the files saved.
237
+ """
238
+ file_path = os.path.join(save_directory, "qwen.tiktoken")
239
+ with open(file_path, "w", encoding="utf8") as w:
240
+ for k, v in self.mergeable_ranks.items():
241
+ line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
242
+ w.write(line)
243
+ return (file_path,)
244
+
245
+ def tokenize(
246
+ self,
247
+ text: str,
248
+ allowed_special: Union[Set, str] = "all",
249
+ disallowed_special: Union[Collection, str] = (),
250
+ **kwargs,
251
+ ) -> List[Union[bytes, str]]:
252
+ """
253
+ Converts a string in a sequence of tokens.
254
+
255
+ Args:
256
+ text (`str`):
257
+ The sequence to be encoded.
258
+ allowed_special (`Literal["all"]` or `set`):
259
+ The surface forms of the tokens to be encoded as special tokens in regular texts.
260
+ Default to "all".
261
+ disallowed_special (`Literal["all"]` or `Collection`):
262
+ The surface forms of the tokens that should not be in regular texts and trigger errors.
263
+ Default to an empty tuple.
264
+
265
+ kwargs (additional keyword arguments, *optional*):
266
+ Will be passed to the underlying model specific encode method.
267
+
268
+ Returns:
269
+ `List[bytes|str]`: The list of tokens.
270
+ """
271
+ tokens = []
272
+ text = unicodedata.normalize("NFC", text)
273
+
274
+ # this implementation takes a detour: text -> token id -> token surface forms
275
+ for t in self.tokenizer.encode(
276
+ text, allowed_special=allowed_special, disallowed_special=disallowed_special
277
+ ):
278
+ tokens.append(self.decoder[t])
279
+
280
+ def _encode_imgurl(img_tokens):
281
+ assert img_tokens[0] == self.image_start_tag and img_tokens[-1] == self.image_end_tag
282
+ img_tokens = img_tokens[1:-1]
283
+ img_url = b''.join(img_tokens)
284
+ out_img_tokens = list(map(self.decoder.get, img_url))
285
+ if len(out_img_tokens) > IMG_TOKEN_SPAN:
286
+ raise ValueError("The content in {}..{} is too long".format(
287
+ self.image_start_tag, self.image_end_tag))
288
+ out_img_tokens.extend([self.image_pad_tag] * (IMG_TOKEN_SPAN - len(out_img_tokens)))
289
+ out_img_tokens = [self.image_start_tag] + out_img_tokens + [self.image_end_tag]
290
+ return out_img_tokens
291
+
292
+ return _replace_closed_tag(tokens, self.image_start_tag, self.image_end_tag, _encode_imgurl)
293
+
294
+ def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
295
+ """
296
+ Converts a sequence of tokens in a single string.
297
+ """
298
+ text = ""
299
+ temp = b""
300
+ for t in tokens:
301
+ if isinstance(t, str):
302
+ if temp:
303
+ text += temp.decode("utf-8", errors=self.errors)
304
+ temp = b""
305
+ text += t
306
+ elif isinstance(t, bytes):
307
+ temp += t
308
+ else:
309
+ raise TypeError("token should only be of type types or str")
310
+ if temp:
311
+ text += temp.decode("utf-8", errors=self.errors)
312
+ return text
313
+
314
+ @property
315
+ def vocab_size(self):
316
+ return self.tokenizer.n_vocab
317
+
318
+ def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
319
+ """Converts an id to a token, special tokens included"""
320
+ if index in self.decoder:
321
+ return self.decoder[index]
322
+ raise ValueError("unknown ids")
323
+
324
+ def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
325
+ """Converts a token to an id using the vocab, special tokens included"""
326
+ if token in self.special_tokens:
327
+ return self.special_tokens[token]
328
+ if token in self.mergeable_ranks:
329
+ return self.mergeable_ranks[token]
330
+ raise ValueError("unknown token")
331
+
332
+ def _tokenize(self, text: str, **kwargs):
333
+ """
334
+ Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
335
+ vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
336
+
337
+ Do NOT take care of added tokens.
338
+ """
339
+ raise NotImplementedError
340
+
341
+ def _decode(
342
+ self,
343
+ token_ids: Union[int, List[int]],
344
+ skip_special_tokens: bool = False,
345
+ errors: str = None,
346
+ **kwargs,
347
+ ) -> str:
348
+ if isinstance(token_ids, int):
349
+ token_ids = [token_ids]
350
+
351
+ def _decode_imgurl(img_token_ids):
352
+ assert img_token_ids[0] == self.img_start_id and img_token_ids[-1] == self.img_end_id
353
+ img_token_ids = img_token_ids[1:-1]
354
+ img_token_ids = img_token_ids[ : img_token_ids.index(self.img_pad_id)]
355
+ img_url = bytes(img_token_ids).decode('utf-8')
356
+ return [self.img_start_id] + self.tokenizer.encode(img_url) + [self.img_end_id]
357
+
358
+ token_ids = _replace_closed_tag(token_ids, self.img_start_id, self.img_end_id, _decode_imgurl)
359
+
360
+ if skip_special_tokens:
361
+ if kwargs.get('keep_image_special', False):
362
+ token_ids = [i for i in token_ids if i < self.eod_id
363
+ or i in self.image_special_tokens]
364
+ else:
365
+ token_ids = [i for i in token_ids if i < self.eod_id]
366
+ return self.tokenizer.decode(token_ids, errors=errors or self.errors)
367
+
368
+ def to_list_format(self, text: str):
369
+ text = unicodedata.normalize("NFC", text)
370
+ token_ids = self.tokenizer.encode(
371
+ text, allowed_special=set(self.IMAGE_ST + (ENDOFTEXT,)))
372
+
373
+ def _encode_vl_info(tokens):
374
+ if len(tokens) == 0:
375
+ return []
376
+ if tokens[0] == self.img_start_id and tokens[-1] == self.img_end_id:
377
+ key = 'image'
378
+ elif tokens[0] == self.ref_start_id and tokens[-1] == self.ref_end_id:
379
+ key = 'ref'
380
+ elif tokens[0] == self.box_start_id and tokens[-1] == self.box_end_id:
381
+ key = 'box'
382
+ elif tokens[0] == self.quad_start_id and tokens[-1] == self.quad_end_id:
383
+ key = 'quad'
384
+ else:
385
+ _tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
386
+ return [{'text': b''.join(map(_tobytes, map(self.decoder.get, tokens))).decode('utf-8')}]
387
+ _tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
388
+ val = b''.join(map(_tobytes, map(self.decoder.get, tokens[1:-1]))).decode('utf-8')
389
+ return [{key: val}]
390
+
391
+ return _replace_closed_tag(
392
+ token_ids,
393
+ (self.img_start_id, self.ref_start_id, self.box_start_id, self.quad_start_id),
394
+ (self.img_end_id, self.ref_end_id, self.box_end_id, self.quad_end_id),
395
+ _encode_vl_info,
396
+ _encode_vl_info,
397
+ )
398
+
399
+ def from_list_format(self, list_format: List[Dict]):
400
+ text = ''
401
+ num_images = 0
402
+ for ele in list_format:
403
+ if 'image' in ele:
404
+ num_images += 1
405
+ text += f'Picture {num_images}: '
406
+ text += self.image_start_tag + ele['image'] + self.image_end_tag
407
+ text += '\n'
408
+ elif 'text' in ele:
409
+ text += ele['text']
410
+ elif 'box' in ele:
411
+ if 'ref' in ele:
412
+ text += self.ref_start_tag + ele['ref'] + self.ref_end_tag
413
+ for box in ele['box']:
414
+ text += self.box_start_tag + '(%d,%d),(%d,%d)' % (box[0], box[1], box[2], box[3]) + self.box_end_tag
415
+ else:
416
+ raise ValueError("Unsupport element: " + str(ele))
417
+ return text
418
+
419
+ def _fetch_latest_picture(self, response, history):
420
+ if history is None:
421
+ history = []
422
+ _history = history + [(response, None)]
423
+ for q, r in _history[::-1]:
424
+ for ele in self.to_list_format(q)[::-1]:
425
+ if 'image' in ele:
426
+ return ele['image']
427
+ return None
428
+
429
+ def _fetch_all_box_with_ref(self, text):
430
+ list_format = self.to_list_format(text)
431
+ output = []
432
+ for i, ele in enumerate(list_format):
433
+ if 'box' in ele:
434
+ bbox = tuple(map(int, ele['box'].replace('(', '').replace(')', '').split(',')))
435
+ assert len(bbox) == 4
436
+ output.append({'box': bbox})
437
+ if i > 0 and 'ref' in list_format[i-1]:
438
+ output[-1]['ref'] = list_format[i-1]['ref'].strip()
439
+ return output
440
+
441
+ def draw_bbox_on_latest_picture(
442
+ self,
443
+ response,
444
+ history=None,
445
+ ) -> Optional[Image.Image]:
446
+ image = self._fetch_latest_picture(response, history)
447
+ if image is None:
448
+ return None
449
+ if image.startswith("http://") or image.startswith("https://"):
450
+ image = Image.open(requests.get(image, stream=True).raw).convert("RGB")
451
+ h, w = image.height, image.width
452
+ else:
453
+ image = np.asarray(Image.open(image).convert("RGB"))
454
+ h, w = image.shape[0], image.shape[1]
455
+ visualizer = Visualizer(image)
456
+
457
+ boxes = self._fetch_all_box_with_ref(response)
458
+ if not boxes:
459
+ return None
460
+ color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()]) # init color
461
+ for box in boxes:
462
+ if 'ref' in box: # random new color for new refexps
463
+ color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()])
464
+ x1, y1, x2, y2 = box['box']
465
+ x1, y1, x2, y2 = (int(x1 / 1000 * w), int(y1 / 1000 * h), int(x2 / 1000 * w), int(y2 / 1000 * h))
466
+ visualizer.draw_box((x1, y1, x2, y2), alpha=1, edge_color=color)
467
+ if 'ref' in box:
468
+ visualizer.draw_text(box['ref'], (x1, y1), color=color, horizontal_alignment="left")
469
+ return visualizer.output
470
+
471
+
472
+ import colorsys
473
+ import logging
474
+ import math
475
+ import numpy as np
476
+ import matplotlib as mpl
477
+ import matplotlib.colors as mplc
478
+ import matplotlib.figure as mplfigure
479
+ import torch
480
+ from matplotlib.backends.backend_agg import FigureCanvasAgg
481
+ from PIL import Image
482
+ import random
483
+
484
+ logger = logging.getLogger(__name__)
485
+
486
+
487
+ class VisImage:
488
+ def __init__(self, img, scale=1.0):
489
+ self.img = img
490
+ self.scale = scale
491
+ self.width, self.height = img.shape[1], img.shape[0]
492
+ self._setup_figure(img)
493
+
494
+ def _setup_figure(self, img):
495
+ fig = mplfigure.Figure(frameon=False)
496
+ self.dpi = fig.get_dpi()
497
+ # add a small 1e-2 to avoid precision lost due to matplotlib's truncation
498
+ # (https://github.com/matplotlib/matplotlib/issues/15363)
499
+ fig.set_size_inches(
500
+ (self.width * self.scale + 1e-2) / self.dpi,
501
+ (self.height * self.scale + 1e-2) / self.dpi,
502
+ )
503
+ self.canvas = FigureCanvasAgg(fig)
504
+ # self.canvas = mpl.backends.backend_cairo.FigureCanvasCairo(fig)
505
+ ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
506
+ ax.axis("off")
507
+ self.fig = fig
508
+ self.ax = ax
509
+ self.reset_image(img)
510
+
511
+ def reset_image(self, img):
512
+ img = img.astype("uint8")
513
+ self.ax.imshow(img, extent=(0, self.width, self.height, 0), interpolation="nearest")
514
+
515
+ def save(self, filepath):
516
+ self.fig.savefig(filepath)
517
+
518
+ def get_image(self):
519
+ canvas = self.canvas
520
+ s, (width, height) = canvas.print_to_buffer()
521
+
522
+ buffer = np.frombuffer(s, dtype="uint8")
523
+
524
+ img_rgba = buffer.reshape(height, width, 4)
525
+ rgb, alpha = np.split(img_rgba, [3], axis=2)
526
+ return rgb.astype("uint8")
527
+
528
+
529
+ class Visualizer:
530
+ def __init__(self, img_rgb, metadata=None, scale=1.0):
531
+ self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
532
+ self.font_path = FONT_PATH
533
+ self.output = VisImage(self.img, scale=scale)
534
+ self.cpu_device = torch.device("cpu")
535
+
536
+ # too small texts are useless, therefore clamp to 14
537
+ self._default_font_size = max(
538
+ np.sqrt(self.output.height * self.output.width) // 30, 15 // scale
539
+ )
540
+
541
+ def draw_text(
542
+ self,
543
+ text,
544
+ position,
545
+ *,
546
+ font_size=None,
547
+ color="g",
548
+ horizontal_alignment="center",
549
+ rotation=0,
550
+ ):
551
+ if not font_size:
552
+ font_size = self._default_font_size
553
+
554
+ # since the text background is dark, we don't want the text to be dark
555
+ color = np.maximum(list(mplc.to_rgb(color)), 0.2)
556
+ color[np.argmax(color)] = max(0.8, np.max(color))
557
+
558
+ x, y = position
559
+ self.output.ax.text(
560
+ x,
561
+ y,
562
+ text,
563
+ size=font_size * self.output.scale,
564
+ fontproperties=FontProperties(fname=self.font_path),
565
+ bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
566
+ verticalalignment="top",
567
+ horizontalalignment=horizontal_alignment,
568
+ color=color,
569
+ zorder=10,
570
+ rotation=rotation,
571
+ )
572
+ return self.output
573
+
574
+ def draw_box(self, box_coord, alpha=0.5, edge_color="g", line_style="-"):
575
+
576
+ x0, y0, x1, y1 = box_coord
577
+ width = x1 - x0
578
+ height = y1 - y0
579
+
580
+ linewidth = max(self._default_font_size / 4, 1)
581
+
582
+ self.output.ax.add_patch(
583
+ mpl.patches.Rectangle(
584
+ (x0, y0),
585
+ width,
586
+ height,
587
+ fill=False,
588
+ edgecolor=edge_color,
589
+ linewidth=linewidth * self.output.scale,
590
+ alpha=alpha,
591
+ linestyle=line_style,
592
+ )
593
+ )
594
+ return self.output
595
+
596
+ def get_output(self):
597
+
598
+ return self.output
tokenizer_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_max_length": 8192,
3
+ "tokenizer_class": "QWenTokenizer",
4
+ "auto_map": {
5
+ "AutoTokenizer": [
6
+ "tokenization_qwen.QWenTokenizer",
7
+ null
8
+ ]
9
+ }
10
+ }
visual.py ADDED
@@ -0,0 +1,426 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) Alibaba Cloud.
2
+ #
3
+ # This source code is licensed under the license found in the
4
+ # LICENSE file in the root directory of this source tree.
5
+
6
+ from collections import OrderedDict
7
+ import math
8
+ import requests
9
+ from io import BytesIO
10
+ from functools import partial
11
+ from PIL import Image
12
+ from typing import Callable, Optional, Sequence, Tuple, List
13
+ import numpy as np
14
+
15
+ import torch
16
+ from torch import nn
17
+ from torch.nn import functional as F
18
+ from torch.nn.init import trunc_normal_
19
+ from torchvision import transforms
20
+ from torchvision.transforms import InterpolationMode
21
+
22
+
23
+ def get_abs_pos(abs_pos, tgt_size):
24
+ # abs_pos: L, C
25
+ # tgt_size: M
26
+ # return: M, C
27
+ src_size = int(math.sqrt(abs_pos.size(0)))
28
+ tgt_size = int(math.sqrt(tgt_size))
29
+ dtype = abs_pos.dtype
30
+
31
+ if src_size != tgt_size:
32
+ return F.interpolate(
33
+ abs_pos.float().reshape(1, src_size, src_size, -1).permute(0, 3, 1, 2),
34
+ size=(tgt_size, tgt_size),
35
+ mode="bicubic",
36
+ align_corners=False,
37
+ ).permute(0, 2, 3, 1).flatten(0, 2).to(dtype=dtype)
38
+ else:
39
+ return abs_pos
40
+
41
+ # https://github.com/facebookresearch/mae/blob/efb2a8062c206524e35e47d04501ed4f544c0ae8/util/pos_embed.py#L20
42
+ def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
43
+ """
44
+ grid_size: int of the grid height and width
45
+ return:
46
+ pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
47
+ """
48
+ grid_h = np.arange(grid_size, dtype=np.float32)
49
+ grid_w = np.arange(grid_size, dtype=np.float32)
50
+ grid = np.meshgrid(grid_w, grid_h) # here w goes first
51
+ grid = np.stack(grid, axis=0)
52
+
53
+ grid = grid.reshape([2, 1, grid_size, grid_size])
54
+ pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
55
+ if cls_token:
56
+ pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
57
+ return pos_embed
58
+
59
+
60
+ def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
61
+ assert embed_dim % 2 == 0
62
+
63
+ # use half of dimensions to encode grid_h
64
+ emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0]) # (H*W, D/2)
65
+ emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1]) # (H*W, D/2)
66
+
67
+ emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
68
+ return emb
69
+
70
+
71
+ def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
72
+ """
73
+ embed_dim: output dimension for each position
74
+ pos: a list of positions to be encoded: size (M,)
75
+ out: (M, D)
76
+ """
77
+ assert embed_dim % 2 == 0
78
+ omega = np.arange(embed_dim // 2, dtype=np.float32)
79
+ omega /= embed_dim / 2.
80
+ omega = 1. / 10000**omega # (D/2,)
81
+
82
+ pos = pos.reshape(-1) # (M,)
83
+ out = np.einsum('m,d->md', pos, omega) # (M, D/2), outer product
84
+
85
+ emb_sin = np.sin(out) # (M, D/2)
86
+ emb_cos = np.cos(out) # (M, D/2)
87
+
88
+ emb = np.concatenate([emb_sin, emb_cos], axis=1) # (M, D)
89
+ return emb
90
+
91
+
92
+ class Resampler(nn.Module):
93
+ """
94
+ A 2D perceiver-resampler network with one cross attention layers by
95
+ (grid_size**2) learnable queries and 2d sincos pos_emb
96
+ Outputs:
97
+ A tensor with the shape of (grid_size**2, embed_dim)
98
+ """
99
+ def __init__(
100
+ self,
101
+ grid_size,
102
+ embed_dim,
103
+ num_heads,
104
+ kv_dim=None,
105
+ norm_layer=nn.LayerNorm
106
+ ):
107
+ super().__init__()
108
+ self.num_queries = grid_size ** 2
109
+ self.embed_dim = embed_dim
110
+ self.num_heads = num_heads
111
+
112
+ self.pos_embed = nn.Parameter(
113
+ torch.from_numpy(get_2d_sincos_pos_embed(embed_dim, grid_size)).float()
114
+ ).requires_grad_(False)
115
+
116
+ self.query = nn.Parameter(torch.zeros(self.num_queries, embed_dim))
117
+ trunc_normal_(self.query, std=.02)
118
+
119
+ if kv_dim is not None and kv_dim != embed_dim:
120
+ self.kv_proj = nn.Linear(kv_dim, embed_dim, bias=False)
121
+ else:
122
+ self.kv_proj = nn.Identity()
123
+
124
+ self.attn = nn.MultiheadAttention(embed_dim, num_heads)
125
+ self.ln_q = norm_layer(embed_dim)
126
+ self.ln_kv = norm_layer(embed_dim)
127
+
128
+ # self.apply(self._init_weights)
129
+
130
+ def _init_weights(self, m):
131
+ if isinstance(m, nn.Linear):
132
+ trunc_normal_(m.weight, std=.02)
133
+ if isinstance(m, nn.Linear) and m.bias is not None:
134
+ nn.init.constant_(m.bias, 0)
135
+ elif isinstance(m, nn.LayerNorm):
136
+ nn.init.constant_(m.bias, 0)
137
+ nn.init.constant_(m.weight, 1.0)
138
+
139
+ def forward(self, x, attn_mask=None):
140
+
141
+ pos_embed = get_abs_pos(self.pos_embed, x.size(1))
142
+
143
+ x = self.kv_proj(x)
144
+ x = self.ln_kv(x).permute(1, 0, 2)
145
+
146
+ N = x.shape[1]
147
+ q = self.ln_q(self.query)
148
+ out = self.attn(
149
+ self._repeat(q, N) + self.pos_embed.unsqueeze(1),
150
+ x + pos_embed.unsqueeze(1),
151
+ x,
152
+ attn_mask=attn_mask)[0]
153
+ return out.permute(1, 0, 2)
154
+
155
+ def _repeat(self, query, N: int):
156
+ return query.unsqueeze(1).repeat(1, N, 1)
157
+
158
+
159
+ class VisualAttention(nn.Module):
160
+ """self-attention layer class.
161
+
162
+ Self-attention layer takes input with size [s, b, h]
163
+ and returns output of the same size.
164
+ """
165
+
166
+ def __init__(self, embed_dim, num_heads,
167
+ bias=True, kdim=None, vdim=None):
168
+ super(VisualAttention, self).__init__()
169
+ self.embed_dim = embed_dim
170
+ self.kdim = kdim if kdim is not None else embed_dim
171
+ self.vdim = vdim if vdim is not None else embed_dim
172
+ self._qkv_same_embed_dim = self.kdim == embed_dim and self.vdim == embed_dim
173
+
174
+ self.num_heads = num_heads
175
+
176
+ # Per attention head and per partition values.
177
+ assert embed_dim % num_heads == 0
178
+ self.hidden_size_per_attention_head = embed_dim // num_heads
179
+ self.num_attention_heads_per_partition = num_heads
180
+ self.hidden_size_per_partition = embed_dim
181
+
182
+ # Strided linear layer.
183
+ assert self._qkv_same_embed_dim, 'Only Support SelfAttention Currently'
184
+ self.in_proj = nn.Linear(embed_dim, 3 * embed_dim)
185
+ self.out_proj = nn.Linear(embed_dim, embed_dim)
186
+ self.norm_factor = math.sqrt(self.hidden_size_per_attention_head)
187
+
188
+ def forward(self, query, key, value, attn_mask = None):
189
+ # query/key/value: [sq, b, h]
190
+ sq, b, _ = query.size()
191
+
192
+ assert torch.allclose(query, key), 'Only Support Self-Attention Currently'
193
+ sk = sq
194
+ mixed_x_layer = self.in_proj(query)
195
+
196
+ # [sq, b, (np * 3 * hn)] --> [sq, b, np, 3 * hn]
197
+ new_tensor_shape = mixed_x_layer.size()[:-1] + \
198
+ (self.num_attention_heads_per_partition,
199
+ 3 * self.hidden_size_per_attention_head)
200
+ mixed_x_layer = mixed_x_layer.view(*new_tensor_shape)
201
+
202
+ # [sq, b, np, 3 * hn] --> 3 [sq, b, np, hn]
203
+ query_layer, key_layer, value_layer = mixed_x_layer.split(
204
+ self.hidden_size_per_attention_head, dim=-1)
205
+
206
+ # [sq, b, np, hn] -> [sq, b * np, hn]
207
+ query_layer = query_layer.view(sq,
208
+ b * self.num_attention_heads_per_partition,
209
+ self.hidden_size_per_attention_head).transpose(0, 1)
210
+ # [sk, b, np, hn] -> [sk, b * np, hn]
211
+ key_layer = key_layer.view(sk,
212
+ b * self.num_attention_heads_per_partition,
213
+ self.hidden_size_per_attention_head).transpose(0, 1)
214
+
215
+ q_scaled = query_layer / self.norm_factor
216
+ if attn_mask is not None:
217
+ attention_probs = torch.baddbmm(attn_mask, q_scaled, key_layer.transpose(-2, -1))
218
+ else:
219
+ attention_probs = torch.bmm(q_scaled, key_layer.transpose(-2, -1))
220
+ attention_probs = attention_probs.softmax(dim=-1)
221
+
222
+ value_layer = value_layer.view(sk,
223
+ b * self.num_attention_heads_per_partition,
224
+ self.hidden_size_per_attention_head).transpose(0, 1)
225
+
226
+ # matmul: [b * np, sq, hn]
227
+ context_layer = torch.bmm(attention_probs, value_layer)
228
+
229
+ # change view [b, np, sq, hn]
230
+ context_layer = context_layer.view(b,
231
+ self.num_attention_heads_per_partition,
232
+ sq, self.hidden_size_per_attention_head)
233
+
234
+ # [b, np, sq, hn] --> [sq, b, np, hn]
235
+ context_layer = context_layer.permute(2, 0, 1, 3).contiguous()
236
+
237
+ # [sq, b, np, hn] --> [sq, b, hp]
238
+ new_context_layer_shape = context_layer.size()[:-2] + \
239
+ (self.hidden_size_per_partition,)
240
+ context_layer = context_layer.view(*new_context_layer_shape)
241
+
242
+ output = self.out_proj(context_layer)
243
+
244
+ return output
245
+
246
+
247
+ class VisualAttentionBlock(nn.Module):
248
+ def __init__(
249
+ self,
250
+ d_model: int,
251
+ n_head: int,
252
+ mlp_ratio: float = 4.0,
253
+ act_layer: Callable = nn.GELU,
254
+ norm_layer: Callable = nn.LayerNorm,
255
+ is_cross_attention: bool = False,
256
+ ):
257
+ super().__init__()
258
+
259
+ self.ln_1 = norm_layer(d_model)
260
+ if is_cross_attention:
261
+ self.ln_1_kv = norm_layer(d_model)
262
+
263
+ self.ln_2 = norm_layer(d_model)
264
+ mlp_width = int(d_model * mlp_ratio)
265
+ self.attn = VisualAttention(d_model, n_head)
266
+ self.mlp = nn.Sequential(OrderedDict([
267
+ ("c_fc", nn.Linear(d_model, mlp_width)),
268
+ ("gelu", act_layer()),
269
+ ("c_proj", nn.Linear(mlp_width, d_model))
270
+ ]))
271
+
272
+ def attention(
273
+ self,
274
+ q_x: torch.Tensor,
275
+ k_x: Optional[torch.Tensor] = None,
276
+ v_x: Optional[torch.Tensor] = None,
277
+ attn_mask: Optional[torch.Tensor] = None,
278
+ ):
279
+ k_x = k_x if k_x is not None else q_x
280
+ v_x = v_x if v_x is not None else q_x
281
+
282
+ attn_mask = attn_mask.to(q_x.dtype) if attn_mask is not None else None
283
+ return self.attn(q_x, k_x, v_x, attn_mask=attn_mask)
284
+
285
+ def forward(
286
+ self,
287
+ q_x: torch.Tensor,
288
+ k_x: Optional[torch.Tensor] = None,
289
+ v_x: Optional[torch.Tensor] = None,
290
+ attn_mask: Optional[torch.Tensor] = None,
291
+ ):
292
+ k_x = self.ln_1_kv(k_x) if hasattr(self, "ln_1_kv") and k_x is not None else None
293
+ v_x = self.ln_1_kv(v_x) if hasattr(self, "ln_1_kv") and v_x is not None else None
294
+
295
+ x = q_x + self.attention(q_x=self.ln_1(q_x), k_x=k_x, v_x=v_x, attn_mask=attn_mask)
296
+ x = x + self.mlp(self.ln_2(x))
297
+ return x
298
+
299
+
300
+ class TransformerBlock(nn.Module):
301
+ def __init__(
302
+ self,
303
+ width: int,
304
+ layers: int,
305
+ heads: int,
306
+ mlp_ratio: float = 4.0,
307
+ act_layer: Callable = nn.GELU,
308
+ norm_layer: Callable = nn.LayerNorm,
309
+ ):
310
+ super().__init__()
311
+ self.width = width
312
+ self.layers = layers
313
+
314
+ self.resblocks = nn.ModuleList([
315
+ VisualAttentionBlock(
316
+ width, heads, mlp_ratio, act_layer=act_layer, norm_layer=norm_layer)
317
+ for _ in range(layers)
318
+ ])
319
+
320
+ def get_cast_dtype(self) -> torch.dtype:
321
+ return self.resblocks[0].mlp.c_fc.weight.dtype
322
+
323
+ def get_cast_device(self) -> torch.device:
324
+ return self.resblocks[0].mlp.c_fc.weight.device
325
+
326
+ def forward(self, x: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
327
+ for r in self.resblocks:
328
+ x = r(x, attn_mask=attn_mask)
329
+ return x
330
+
331
+
332
+ class VisionTransformer(nn.Module):
333
+
334
+ def __init__(
335
+ self,
336
+ image_size: int,
337
+ patch_size: int,
338
+ width: int,
339
+ layers: int,
340
+ heads: int,
341
+ mlp_ratio: float,
342
+ n_queries: int = 256,
343
+ output_dim: int = 512,
344
+ **kwargs
345
+ ):
346
+ super().__init__()
347
+ image_height, image_width = self.image_size = (image_size, image_size)
348
+ patch_height, patch_width = self.patch_size = (patch_size, patch_size)
349
+ self.grid_size = (image_height // patch_height, image_width // patch_width)
350
+ self.output_dim = output_dim
351
+
352
+ mean = (0.48145466, 0.4578275, 0.40821073)
353
+ std = (0.26862954, 0.26130258, 0.27577711)
354
+ self.image_transform = transforms.Compose([
355
+ transforms.Resize(
356
+ (image_size, image_size),
357
+ interpolation=InterpolationMode.BICUBIC
358
+ ),
359
+ transforms.ToTensor(),
360
+ transforms.Normalize(mean=mean, std=std),
361
+ ])
362
+
363
+ self.conv1 = nn.Conv2d(in_channels=3, out_channels=width, kernel_size=patch_size, stride=patch_size, bias=False)
364
+
365
+ # class embeddings and positional embeddings
366
+ scale = width ** -0.5
367
+ self.positional_embedding = nn.Parameter(scale * torch.randn(256, width))
368
+
369
+ norm_layer = partial(nn.LayerNorm, eps=1e-6)
370
+ act_layer = nn.GELU
371
+
372
+ self.ln_pre = norm_layer(width)
373
+ self.transformer = TransformerBlock(
374
+ width,
375
+ layers,
376
+ heads,
377
+ mlp_ratio,
378
+ act_layer=act_layer,
379
+ norm_layer=norm_layer,
380
+ )
381
+
382
+ self.attn_pool = Resampler(
383
+ grid_size=int(math.sqrt(n_queries)),
384
+ embed_dim=output_dim,
385
+ num_heads=output_dim // 128,
386
+ kv_dim=width,
387
+ norm_layer=norm_layer,
388
+ )
389
+ self.ln_post = norm_layer(output_dim)
390
+ self.proj = nn.Parameter((output_dim** -0.5) * torch.randn(output_dim, output_dim))
391
+
392
+ def forward(self, x: torch.Tensor):
393
+ x = x.to(
394
+ dtype=self.transformer.get_cast_dtype(),
395
+ device=self.transformer.get_cast_device(),
396
+ )
397
+ # to patches
398
+ x = self.conv1(x) # shape = [*, width, grid, grid]
399
+ x = x.reshape(x.shape[0], x.shape[1], -1) # shape = [*, width, grid ** 2]
400
+ x = x.permute(0, 2, 1) # shape = [*, grid ** 2, width]
401
+
402
+ x = x + get_abs_pos(self.positional_embedding, x.size(1))
403
+
404
+ x = self.ln_pre(x)
405
+
406
+ x = x.permute(1, 0, 2) # NLD -> LND
407
+ x = self.transformer(x)
408
+ x = x.permute(1, 0, 2) # LND -> NLD
409
+
410
+ x = self.attn_pool(x)
411
+ x = self.ln_post(x)
412
+ x = x @ self.proj
413
+
414
+ return x
415
+
416
+ def encode(self, image_paths: List[str]):
417
+ images = []
418
+ for image_path in image_paths:
419
+ if image_path.startswith("http://") or image_path.startswith("https://"):
420
+ image = Image.open(requests.get(image_path, stream=True).raw)
421
+ else:
422
+ image = Image.open(image_path)
423
+ image = image.convert("RGB")
424
+ images.append(self.image_transform(image))
425
+ images = torch.stack(images, dim=0)
426
+ return self(images)