y8lc2wpi
commited on
Commit
•
1ec36f8
1
Parent(s):
e6b3ddd
Upload folder using huggingface_hub
Browse files- .ipynb_checkpoints/config-checkpoint.json +26 -0
- .ipynb_checkpoints/generation_config-checkpoint.json +20 -0
- .ipynb_checkpoints/special_tokens_map-checkpoint.json +5 -0
- .ipynb_checkpoints/tokenization_qwen-checkpoint.py +566 -0
- .ipynb_checkpoints/tokenizer_config-checkpoint.json +10 -0
- config.json +26 -0
- generation_config.json +20 -0
- merges.txt +0 -0
- pytorch_model-00001-of-00008.bin +3 -0
- pytorch_model-00002-of-00008.bin +3 -0
- pytorch_model-00003-of-00008.bin +3 -0
- pytorch_model-00004-of-00008.bin +3 -0
- pytorch_model-00005-of-00008.bin +3 -0
- pytorch_model-00006-of-00008.bin +3 -0
- pytorch_model-00007-of-00008.bin +3 -0
- pytorch_model-00008-of-00008.bin +3 -0
- pytorch_model.bin.index.json +330 -0
- qwen.tiktoken +0 -0
- special_tokens_map.json +5 -0
- tokenization_qwen.py +566 -0
- tokenizer_config.json +10 -0
- vision.bin +3 -0
- visual.py +428 -0
- vocab.json +0 -0
.ipynb_checkpoints/config-checkpoint.json
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "/notebooks/qwenv",
|
3 |
+
"architectures": [
|
4 |
+
"LlamaForCausalLM"
|
5 |
+
],
|
6 |
+
"bos_token_id": 151643,
|
7 |
+
"eos_token_id": 151643,
|
8 |
+
"hidden_act": "silu",
|
9 |
+
"hidden_size": 4096,
|
10 |
+
"initializer_range": 0.02,
|
11 |
+
"intermediate_size": 11008,
|
12 |
+
"max_position_embeddings": 6144,
|
13 |
+
"model_type": "llama",
|
14 |
+
"num_attention_heads": 32,
|
15 |
+
"num_hidden_layers": 32,
|
16 |
+
"num_key_value_heads": 32,
|
17 |
+
"pad_token_id": 0,
|
18 |
+
"pretraining_tp": 1,
|
19 |
+
"rms_norm_eps": 1e-05,
|
20 |
+
"rope_scaling": null,
|
21 |
+
"tie_word_embeddings": false,
|
22 |
+
"torch_dtype": "float16",
|
23 |
+
"transformers_version": "4.32.0.dev0",
|
24 |
+
"use_cache": false,
|
25 |
+
"vocab_size": 151936
|
26 |
+
}
|
.ipynb_checkpoints/generation_config-checkpoint.json
ADDED
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"chat_format": "chatml",
|
3 |
+
"decay_bound": 0.0,
|
4 |
+
"decay_factor": 1.0,
|
5 |
+
"do_sample": true,
|
6 |
+
"eos_token_id": 151643,
|
7 |
+
"factual_nucleus_sampling": false,
|
8 |
+
"max_context_size": 1024,
|
9 |
+
"max_generate_size": 512,
|
10 |
+
"max_new_tokens": 512,
|
11 |
+
"pad_token_id": 151643,
|
12 |
+
"stop_words_ids": [
|
13 |
+
[
|
14 |
+
151643
|
15 |
+
]
|
16 |
+
],
|
17 |
+
"top_k": 0,
|
18 |
+
"top_p": 0.8,
|
19 |
+
"transformers_version": "4.32.0.dev0"
|
20 |
+
}
|
.ipynb_checkpoints/special_tokens_map-checkpoint.json
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "<|endoftext|>",
|
3 |
+
"eos_token": "<|endoftext|>",
|
4 |
+
"unk_token": "<|endoftext|>"
|
5 |
+
}
|
.ipynb_checkpoints/tokenization_qwen-checkpoint.py
ADDED
@@ -0,0 +1,566 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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.pyplot as plt
|
24 |
+
import matplotlib.colors as mcolors
|
25 |
+
from matplotlib.font_manager import FontProperties
|
26 |
+
|
27 |
+
logger = logging.getLogger(__name__)
|
28 |
+
|
29 |
+
|
30 |
+
VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
|
31 |
+
|
32 |
+
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+"""
|
33 |
+
ENDOFTEXT = "<|endoftext|>"
|
34 |
+
IMSTART = "<|im_start|>"
|
35 |
+
IMEND = "<|im_end|>"
|
36 |
+
# as the default behavior is changed to allow special tokens in
|
37 |
+
# regular texts, the surface forms of special tokens need to be
|
38 |
+
# as different as possible to minimize the impact
|
39 |
+
EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
|
40 |
+
SPECIAL_TOKENS = (
|
41 |
+
ENDOFTEXT,
|
42 |
+
IMSTART,
|
43 |
+
IMEND,
|
44 |
+
) + EXTRAS
|
45 |
+
IMG_TOKEN_SPAN = 256
|
46 |
+
|
47 |
+
|
48 |
+
def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
|
49 |
+
with open(tiktoken_bpe_file, "rb") as f:
|
50 |
+
contents = f.read()
|
51 |
+
return {
|
52 |
+
base64.b64decode(token): int(rank)
|
53 |
+
for token, rank in (line.split() for line in contents.splitlines() if line)
|
54 |
+
}
|
55 |
+
|
56 |
+
def _list_find(
|
57 |
+
input_list: List[Any],
|
58 |
+
candidates: Tuple[Any],
|
59 |
+
start: int = 0,
|
60 |
+
):
|
61 |
+
for i in range(start, len(input_list)):
|
62 |
+
if input_list[i] in candidates:
|
63 |
+
return i
|
64 |
+
return -1
|
65 |
+
|
66 |
+
def _replace_closed_tag(
|
67 |
+
input_tokens: List[Any],
|
68 |
+
start_tags: Union[Any, Tuple[Any]],
|
69 |
+
end_tags: Union[Any, Tuple[Any]],
|
70 |
+
inclusive_replace_func: Callable,
|
71 |
+
exclusive_replace_func: Callable = lambda x: x,
|
72 |
+
):
|
73 |
+
if isinstance(start_tags, (str, int)):
|
74 |
+
start_tags = (start_tags,)
|
75 |
+
if isinstance(end_tags, (str, int)):
|
76 |
+
end_tags = (end_tags,)
|
77 |
+
assert len(start_tags) == len(end_tags)
|
78 |
+
|
79 |
+
output_tokens = []
|
80 |
+
end = 0
|
81 |
+
while True:
|
82 |
+
start = _list_find(input_tokens, start_tags, end)
|
83 |
+
if start == -1:
|
84 |
+
break
|
85 |
+
output_tokens.extend(exclusive_replace_func(input_tokens[end : start]))
|
86 |
+
tag_idx = start_tags.index(input_tokens[start])
|
87 |
+
end = _list_find(input_tokens, (end_tags[tag_idx],), start)
|
88 |
+
if end == -1:
|
89 |
+
raise ValueError("Unclosed image token")
|
90 |
+
output_tokens.extend(inclusive_replace_func(input_tokens[start : end + 1]))
|
91 |
+
end += 1
|
92 |
+
output_tokens.extend(exclusive_replace_func(input_tokens[end : ]))
|
93 |
+
return output_tokens
|
94 |
+
|
95 |
+
class QWenTokenizer(PreTrainedTokenizer):
|
96 |
+
"""QWen tokenizer."""
|
97 |
+
|
98 |
+
vocab_files_names = VOCAB_FILES_NAMES
|
99 |
+
|
100 |
+
def __init__(
|
101 |
+
self,
|
102 |
+
vocab_file,
|
103 |
+
errors="replace",
|
104 |
+
image_start_tag='<img>',
|
105 |
+
image_end_tag='</img>',
|
106 |
+
image_pad_tag='<imgpad>',
|
107 |
+
ref_start_tag='<ref>',
|
108 |
+
ref_end_tag='</ref>',
|
109 |
+
box_start_tag='<box>',
|
110 |
+
box_end_tag='</box>',
|
111 |
+
quad_start_tag='<quad>',
|
112 |
+
quad_end_tag='</quad>',
|
113 |
+
**kwargs,
|
114 |
+
):
|
115 |
+
super().__init__(**kwargs)
|
116 |
+
self.image_start_tag = image_start_tag
|
117 |
+
self.image_end_tag = image_end_tag
|
118 |
+
self.image_pad_tag = image_pad_tag
|
119 |
+
self.ref_start_tag = ref_start_tag
|
120 |
+
self.ref_end_tag = ref_end_tag
|
121 |
+
self.box_start_tag = box_start_tag
|
122 |
+
self.box_end_tag = box_end_tag
|
123 |
+
self.quad_start_tag = quad_start_tag
|
124 |
+
self.quad_end_tag = quad_end_tag
|
125 |
+
self.IMAGE_ST = (
|
126 |
+
ref_start_tag, ref_end_tag,
|
127 |
+
box_start_tag, box_end_tag,
|
128 |
+
quad_start_tag, quad_end_tag,
|
129 |
+
image_start_tag, image_end_tag,
|
130 |
+
image_pad_tag
|
131 |
+
)
|
132 |
+
|
133 |
+
self.errors = errors # how to handle errors in decoding
|
134 |
+
|
135 |
+
self.mergeable_ranks = _load_tiktoken_bpe(vocab_file) # type: dict[bytes, int]
|
136 |
+
self.special_tokens = {
|
137 |
+
token: index
|
138 |
+
for index, token in enumerate(
|
139 |
+
SPECIAL_TOKENS + self.IMAGE_ST, start=len(self.mergeable_ranks)
|
140 |
+
)
|
141 |
+
}
|
142 |
+
self.img_start_id = self.special_tokens[self.image_start_tag]
|
143 |
+
self.img_end_id = self.special_tokens[self.image_end_tag]
|
144 |
+
self.img_pad_id = self.special_tokens[self.image_pad_tag]
|
145 |
+
self.ref_start_id = self.special_tokens[self.ref_start_tag]
|
146 |
+
self.ref_end_id = self.special_tokens[self.ref_end_tag]
|
147 |
+
self.box_start_id = self.special_tokens[self.box_start_tag]
|
148 |
+
self.box_end_id = self.special_tokens[self.box_end_tag]
|
149 |
+
self.quad_start_id = self.special_tokens[self.quad_start_tag]
|
150 |
+
self.quad_end_id = self.special_tokens[self.quad_end_tag]
|
151 |
+
|
152 |
+
enc = tiktoken.Encoding(
|
153 |
+
"Qwen",
|
154 |
+
pat_str=PAT_STR,
|
155 |
+
mergeable_ranks=self.mergeable_ranks,
|
156 |
+
special_tokens=self.special_tokens,
|
157 |
+
)
|
158 |
+
assert (
|
159 |
+
len(self.mergeable_ranks) + len(self.special_tokens) == enc.n_vocab
|
160 |
+
), f"{len(self.mergeable_ranks) + len(self.special_tokens)} != {enc.n_vocab} in encoding"
|
161 |
+
|
162 |
+
self.decoder = {
|
163 |
+
v: k for k, v in self.mergeable_ranks.items()
|
164 |
+
} # type: dict[int, bytes|str]
|
165 |
+
self.decoder.update({v: k for k, v in self.special_tokens.items()})
|
166 |
+
|
167 |
+
self.tokenizer = enc # type: tiktoken.Encoding
|
168 |
+
|
169 |
+
self.eod_id = self.tokenizer.eot_token
|
170 |
+
self.im_start_id = self.special_tokens[IMSTART]
|
171 |
+
self.im_end_id = self.special_tokens[IMEND]
|
172 |
+
|
173 |
+
def __len__(self) -> int:
|
174 |
+
return self.tokenizer.n_vocab
|
175 |
+
|
176 |
+
def get_vocab(self) -> Dict[bytes, int]:
|
177 |
+
return self.mergeable_ranks
|
178 |
+
|
179 |
+
def convert_tokens_to_ids(
|
180 |
+
self, tokens: Union[bytes, str, List[Union[bytes, str]]]
|
181 |
+
) -> List[int]:
|
182 |
+
ids = []
|
183 |
+
if isinstance(tokens, (str, bytes)):
|
184 |
+
if tokens in self.special_tokens:
|
185 |
+
return self.special_tokens[tokens]
|
186 |
+
else:
|
187 |
+
return self.mergeable_ranks.get(tokens)
|
188 |
+
for token in tokens:
|
189 |
+
if token in self.special_tokens:
|
190 |
+
ids.append(self.special_tokens[token])
|
191 |
+
else:
|
192 |
+
ids.append(self.mergeable_ranks.get(token))
|
193 |
+
return ids
|
194 |
+
|
195 |
+
def _add_tokens(self, new_tokens: Union[List[str], List[AddedToken]], special_tokens: bool = False) -> int:
|
196 |
+
if not special_tokens and new_tokens:
|
197 |
+
raise ValueError('Adding regular tokens is not supported')
|
198 |
+
for token in new_tokens:
|
199 |
+
surface_form = token.content if isinstance(token, AddedToken) else token
|
200 |
+
if surface_form not in SPECIAL_TOKENS + self.IMAGE_ST:
|
201 |
+
raise ValueError('Adding unknown special tokens is not supported')
|
202 |
+
return 0
|
203 |
+
|
204 |
+
def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
|
205 |
+
"""
|
206 |
+
Save only the vocabulary of the tokenizer (vocabulary).
|
207 |
+
|
208 |
+
Returns:
|
209 |
+
`Tuple(str)`: Paths to the files saved.
|
210 |
+
"""
|
211 |
+
file_path = os.path.join(save_directory, "qwen.tiktoken")
|
212 |
+
with open(file_path, "w", encoding="utf8") as w:
|
213 |
+
for k, v in self.mergeable_ranks.items():
|
214 |
+
line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
|
215 |
+
w.write(line)
|
216 |
+
return (file_path,)
|
217 |
+
|
218 |
+
def tokenize(
|
219 |
+
self,
|
220 |
+
text: str,
|
221 |
+
allowed_special: Union[Set, str] = "all",
|
222 |
+
disallowed_special: Union[Collection, str] = (),
|
223 |
+
**kwargs,
|
224 |
+
) -> List[Union[bytes, str]]:
|
225 |
+
"""
|
226 |
+
Converts a string in a sequence of tokens.
|
227 |
+
|
228 |
+
Args:
|
229 |
+
text (`str`):
|
230 |
+
The sequence to be encoded.
|
231 |
+
allowed_special (`Literal["all"]` or `set`):
|
232 |
+
The surface forms of the tokens to be encoded as special tokens in regular texts.
|
233 |
+
Default to "all".
|
234 |
+
disallowed_special (`Literal["all"]` or `Collection`):
|
235 |
+
The surface forms of the tokens that should not be in regular texts and trigger errors.
|
236 |
+
Default to an empty tuple.
|
237 |
+
|
238 |
+
kwargs (additional keyword arguments, *optional*):
|
239 |
+
Will be passed to the underlying model specific encode method.
|
240 |
+
|
241 |
+
Returns:
|
242 |
+
`List[bytes|str]`: The list of tokens.
|
243 |
+
"""
|
244 |
+
tokens = []
|
245 |
+
text = unicodedata.normalize("NFC", text)
|
246 |
+
|
247 |
+
# this implementation takes a detour: text -> token id -> token surface forms
|
248 |
+
for t in self.tokenizer.encode(
|
249 |
+
text, allowed_special=allowed_special, disallowed_special=disallowed_special
|
250 |
+
):
|
251 |
+
tokens.append(self.decoder[t])
|
252 |
+
|
253 |
+
def _encode_imgurl(img_tokens):
|
254 |
+
assert img_tokens[0] == self.image_start_tag and img_tokens[-1] == self.image_end_tag
|
255 |
+
img_tokens = img_tokens[1:-1]
|
256 |
+
img_url = b''.join(img_tokens)
|
257 |
+
out_img_tokens = list(map(self.decoder.get, img_url))
|
258 |
+
if len(out_img_tokens) > IMG_TOKEN_SPAN:
|
259 |
+
raise ValueError("The content in {}..{} is too long".format(
|
260 |
+
self.image_start_tag, self.image_end_tag))
|
261 |
+
out_img_tokens.extend([self.image_pad_tag] * (IMG_TOKEN_SPAN - len(out_img_tokens)))
|
262 |
+
out_img_tokens = [self.image_start_tag] + out_img_tokens + [self.image_end_tag]
|
263 |
+
return out_img_tokens
|
264 |
+
|
265 |
+
return _replace_closed_tag(tokens, self.image_start_tag, self.image_end_tag, _encode_imgurl)
|
266 |
+
|
267 |
+
def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
|
268 |
+
"""
|
269 |
+
Converts a sequence of tokens in a single string.
|
270 |
+
"""
|
271 |
+
text = ""
|
272 |
+
temp = b""
|
273 |
+
for t in tokens:
|
274 |
+
if isinstance(t, str):
|
275 |
+
if temp:
|
276 |
+
text += temp.decode("utf-8", errors=self.errors)
|
277 |
+
temp = b""
|
278 |
+
text += t
|
279 |
+
elif isinstance(t, bytes):
|
280 |
+
temp += t
|
281 |
+
else:
|
282 |
+
raise TypeError("token should only be of type types or str")
|
283 |
+
if temp:
|
284 |
+
text += temp.decode("utf-8", errors=self.errors)
|
285 |
+
return text
|
286 |
+
|
287 |
+
@property
|
288 |
+
def vocab_size(self):
|
289 |
+
return self.tokenizer.n_vocab
|
290 |
+
|
291 |
+
def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
|
292 |
+
"""Converts an id to a token, special tokens included"""
|
293 |
+
if index in self.decoder:
|
294 |
+
return self.decoder[index]
|
295 |
+
raise ValueError("unknown ids")
|
296 |
+
|
297 |
+
def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
|
298 |
+
"""Converts a token to an id using the vocab, special tokens included"""
|
299 |
+
if token in self.special_tokens:
|
300 |
+
return self.special_tokens[token]
|
301 |
+
if token in self.mergeable_ranks:
|
302 |
+
return self.mergeable_ranks[token]
|
303 |
+
raise ValueError("unknown token")
|
304 |
+
|
305 |
+
def _tokenize(self, text: str, **kwargs):
|
306 |
+
"""
|
307 |
+
Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
|
308 |
+
vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
|
309 |
+
|
310 |
+
Do NOT take care of added tokens.
|
311 |
+
"""
|
312 |
+
raise NotImplementedError
|
313 |
+
|
314 |
+
def _decode(
|
315 |
+
self,
|
316 |
+
token_ids: Union[int, List[int]],
|
317 |
+
skip_special_tokens: bool = False,
|
318 |
+
errors: str = None,
|
319 |
+
**kwargs,
|
320 |
+
) -> str:
|
321 |
+
if isinstance(token_ids, int):
|
322 |
+
token_ids = [token_ids]
|
323 |
+
|
324 |
+
def _decode_imgurl(img_token_ids):
|
325 |
+
assert img_token_ids[0] == self.img_start_id and img_token_ids[-1] == self.img_end_id
|
326 |
+
img_token_ids = img_token_ids[1:-1]
|
327 |
+
img_token_ids = img_token_ids[ : img_token_ids.index(self.img_pad_id)]
|
328 |
+
img_url = bytes(img_token_ids).decode('utf-8')
|
329 |
+
return [self.img_start_id] + self.tokenizer.encode(img_url) + [self.img_end_id]
|
330 |
+
|
331 |
+
token_ids = _replace_closed_tag(token_ids, self.img_start_id, self.img_end_id, _decode_imgurl)
|
332 |
+
|
333 |
+
if skip_special_tokens:
|
334 |
+
token_ids = [i for i in token_ids if i < self.eod_id]
|
335 |
+
return self.tokenizer.decode(token_ids, errors=errors or self.errors)
|
336 |
+
|
337 |
+
def to_list_format(self, text: str):
|
338 |
+
text = unicodedata.normalize("NFC", text)
|
339 |
+
token_ids = self.tokenizer.encode(
|
340 |
+
text, allowed_special=set(self.IMAGE_ST + (ENDOFTEXT,)))
|
341 |
+
|
342 |
+
def _encode_vl_info(tokens):
|
343 |
+
if len(tokens) == 0:
|
344 |
+
return []
|
345 |
+
if tokens[0] == self.img_start_id and tokens[-1] == self.img_end_id:
|
346 |
+
key = 'image'
|
347 |
+
elif tokens[0] == self.ref_start_id and tokens[-1] == self.ref_end_id:
|
348 |
+
key = 'ref'
|
349 |
+
elif tokens[0] == self.box_start_id and tokens[-1] == self.box_end_id:
|
350 |
+
key = 'box'
|
351 |
+
elif tokens[0] == self.quad_start_id and tokens[-1] == self.quad_end_id:
|
352 |
+
key = 'quad'
|
353 |
+
else:
|
354 |
+
_tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
|
355 |
+
return [{'text': b''.join(map(_tobytes, map(self.decoder.get, tokens))).decode('utf-8')}]
|
356 |
+
val = b''.join(map(self.decoder.get, tokens[1:-1])).decode('utf-8')
|
357 |
+
return [{key: val}]
|
358 |
+
|
359 |
+
return _replace_closed_tag(
|
360 |
+
token_ids,
|
361 |
+
(self.img_start_id, self.ref_start_id, self.box_start_id, self.quad_start_id),
|
362 |
+
(self.img_end_id, self.ref_end_id, self.box_end_id, self.quad_end_id),
|
363 |
+
_encode_vl_info,
|
364 |
+
_encode_vl_info,
|
365 |
+
)
|
366 |
+
|
367 |
+
def from_list_format(self, list_format: List[Dict]):
|
368 |
+
text = ''
|
369 |
+
num_images = 0
|
370 |
+
for ele in list_format:
|
371 |
+
if 'image' in ele:
|
372 |
+
num_images += 1
|
373 |
+
text += f'Picture {num_images}:'
|
374 |
+
text += self.image_start_tag + ele['image'] + self.image_end_tag
|
375 |
+
text += '\n'
|
376 |
+
elif 'text' in ele:
|
377 |
+
text += ele['text']
|
378 |
+
elif 'box' in ele:
|
379 |
+
if 'ref' in ele:
|
380 |
+
text += self.ref_start_tag + ele['ref'] + self.ref_end_tag
|
381 |
+
for box in ele['box']:
|
382 |
+
text += self.box_start_tag + '(%d,%d),(%d,%d)' % (box[0], box[1], box[2], box[3]) + self.box_end_tag
|
383 |
+
else:
|
384 |
+
raise ValueError("Unsupport element: " + str(ele))
|
385 |
+
return text
|
386 |
+
|
387 |
+
def _fetch_latest_picture(self, response, history):
|
388 |
+
if history is None:
|
389 |
+
history = []
|
390 |
+
_history = history + [(response, None)]
|
391 |
+
for q, r in _history[::-1]:
|
392 |
+
for ele in self.to_list_format(q)[::-1]:
|
393 |
+
if 'image' in ele:
|
394 |
+
return ele['image']
|
395 |
+
return None
|
396 |
+
|
397 |
+
def _fetch_all_box_with_ref(self, text):
|
398 |
+
list_format = self.to_list_format(text)
|
399 |
+
output = []
|
400 |
+
for i, ele in enumerate(list_format):
|
401 |
+
if 'box' in ele:
|
402 |
+
bbox = tuple(map(int, ele['box'].replace('(', '').replace(')', '').split(',')))
|
403 |
+
assert len(bbox) == 4
|
404 |
+
output.append({'box': bbox})
|
405 |
+
if i > 0 and 'ref' in list_format[i-1]:
|
406 |
+
output[-1]['ref'] = list_format[i-1]['ref'].strip()
|
407 |
+
return output
|
408 |
+
|
409 |
+
def draw_bbox_on_latest_picture(
|
410 |
+
self,
|
411 |
+
response,
|
412 |
+
history=None,
|
413 |
+
) -> Optional[Image.Image]:
|
414 |
+
image = self._fetch_latest_picture(response, history)
|
415 |
+
if image is None:
|
416 |
+
return None
|
417 |
+
if image.startswith("http://") or image.startswith("https://"):
|
418 |
+
image = Image.open(requests.get(image, stream=True).raw).convert("RGB")
|
419 |
+
h, w = image.height, image.width
|
420 |
+
else:
|
421 |
+
image = plt.imread(image)
|
422 |
+
h, w = image.shape[0], image.shape[1]
|
423 |
+
visualizer = Visualizer(image)
|
424 |
+
|
425 |
+
boxes = self._fetch_all_box_with_ref(response)
|
426 |
+
if not boxes:
|
427 |
+
return None
|
428 |
+
color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()]) # init color
|
429 |
+
for box in boxes:
|
430 |
+
if 'ref' in box: # random new color for new refexps
|
431 |
+
color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()])
|
432 |
+
x1, y1, x2, y2 = box['box']
|
433 |
+
x1, y1, x2, y2 = (int(x1 / 1000 * w), int(y1 / 1000 * h), int(x2 / 1000 * w), int(y2 / 1000 * h))
|
434 |
+
visualizer.draw_box((x1, y1, x2, y2), alpha=1, edge_color=color)
|
435 |
+
if 'ref' in box:
|
436 |
+
visualizer.draw_text(box['ref'], (x1, y1), color=color, horizontal_alignment="left")
|
437 |
+
return visualizer.output
|
438 |
+
|
439 |
+
|
440 |
+
import colorsys
|
441 |
+
import logging
|
442 |
+
import math
|
443 |
+
import numpy as np
|
444 |
+
import matplotlib as mpl
|
445 |
+
import matplotlib.colors as mplc
|
446 |
+
import matplotlib.figure as mplfigure
|
447 |
+
import torch
|
448 |
+
from matplotlib.backends.backend_agg import FigureCanvasAgg
|
449 |
+
from PIL import Image
|
450 |
+
import random
|
451 |
+
|
452 |
+
logger = logging.getLogger(__name__)
|
453 |
+
|
454 |
+
|
455 |
+
class VisImage:
|
456 |
+
def __init__(self, img, scale=1.0):
|
457 |
+
self.img = img
|
458 |
+
self.scale = scale
|
459 |
+
self.width, self.height = img.shape[1], img.shape[0]
|
460 |
+
self._setup_figure(img)
|
461 |
+
|
462 |
+
def _setup_figure(self, img):
|
463 |
+
fig = mplfigure.Figure(frameon=False)
|
464 |
+
self.dpi = fig.get_dpi()
|
465 |
+
# add a small 1e-2 to avoid precision lost due to matplotlib's truncation
|
466 |
+
# (https://github.com/matplotlib/matplotlib/issues/15363)
|
467 |
+
fig.set_size_inches(
|
468 |
+
(self.width * self.scale + 1e-2) / self.dpi,
|
469 |
+
(self.height * self.scale + 1e-2) / self.dpi,
|
470 |
+
)
|
471 |
+
self.canvas = FigureCanvasAgg(fig)
|
472 |
+
# self.canvas = mpl.backends.backend_cairo.FigureCanvasCairo(fig)
|
473 |
+
ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
|
474 |
+
ax.axis("off")
|
475 |
+
self.fig = fig
|
476 |
+
self.ax = ax
|
477 |
+
self.reset_image(img)
|
478 |
+
|
479 |
+
def reset_image(self, img):
|
480 |
+
img = img.astype("uint8")
|
481 |
+
self.ax.imshow(img, extent=(0, self.width, self.height, 0), interpolation="nearest")
|
482 |
+
|
483 |
+
def save(self, filepath):
|
484 |
+
self.fig.savefig(filepath)
|
485 |
+
|
486 |
+
def get_image(self):
|
487 |
+
canvas = self.canvas
|
488 |
+
s, (width, height) = canvas.print_to_buffer()
|
489 |
+
|
490 |
+
buffer = np.frombuffer(s, dtype="uint8")
|
491 |
+
|
492 |
+
img_rgba = buffer.reshape(height, width, 4)
|
493 |
+
rgb, alpha = np.split(img_rgba, [3], axis=2)
|
494 |
+
return rgb.astype("uint8")
|
495 |
+
|
496 |
+
|
497 |
+
class Visualizer:
|
498 |
+
def __init__(self, img_rgb, metadata=None, scale=1.0):
|
499 |
+
self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
|
500 |
+
self.font_path = try_to_load_from_cache("Qwen/Qwen-VL-Chat", "SimSun.ttf")
|
501 |
+
self.output = VisImage(self.img, scale=scale)
|
502 |
+
self.cpu_device = torch.device("cpu")
|
503 |
+
|
504 |
+
# too small texts are useless, therefore clamp to 14
|
505 |
+
self._default_font_size = max(
|
506 |
+
np.sqrt(self.output.height * self.output.width) // 30, 15 // scale
|
507 |
+
)
|
508 |
+
|
509 |
+
def draw_text(
|
510 |
+
self,
|
511 |
+
text,
|
512 |
+
position,
|
513 |
+
*,
|
514 |
+
font_size=None,
|
515 |
+
color="g",
|
516 |
+
horizontal_alignment="center",
|
517 |
+
rotation=0,
|
518 |
+
):
|
519 |
+
if not font_size:
|
520 |
+
font_size = self._default_font_size
|
521 |
+
|
522 |
+
# since the text background is dark, we don't want the text to be dark
|
523 |
+
color = np.maximum(list(mplc.to_rgb(color)), 0.2)
|
524 |
+
color[np.argmax(color)] = max(0.8, np.max(color))
|
525 |
+
|
526 |
+
x, y = position
|
527 |
+
self.output.ax.text(
|
528 |
+
x,
|
529 |
+
y,
|
530 |
+
text,
|
531 |
+
size=font_size * self.output.scale,
|
532 |
+
fontproperties=FontProperties(fname=self.font_path),
|
533 |
+
bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
|
534 |
+
verticalalignment="top",
|
535 |
+
horizontalalignment=horizontal_alignment,
|
536 |
+
color=color,
|
537 |
+
zorder=10,
|
538 |
+
rotation=rotation,
|
539 |
+
)
|
540 |
+
return self.output
|
541 |
+
|
542 |
+
def draw_box(self, box_coord, alpha=0.5, edge_color="g", line_style="-"):
|
543 |
+
|
544 |
+
x0, y0, x1, y1 = box_coord
|
545 |
+
width = x1 - x0
|
546 |
+
height = y1 - y0
|
547 |
+
|
548 |
+
linewidth = max(self._default_font_size / 4, 1)
|
549 |
+
|
550 |
+
self.output.ax.add_patch(
|
551 |
+
mpl.patches.Rectangle(
|
552 |
+
(x0, y0),
|
553 |
+
width,
|
554 |
+
height,
|
555 |
+
fill=False,
|
556 |
+
edgecolor=edge_color,
|
557 |
+
linewidth=linewidth * self.output.scale,
|
558 |
+
alpha=alpha,
|
559 |
+
linestyle=line_style,
|
560 |
+
)
|
561 |
+
)
|
562 |
+
return self.output
|
563 |
+
|
564 |
+
def get_output(self):
|
565 |
+
|
566 |
+
return self.output
|
.ipynb_checkpoints/tokenizer_config-checkpoint.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_max_length": 999999999999999999,
|
3 |
+
"tokenizer_class": "QWenTokenizer",
|
4 |
+
"auto_map": {
|
5 |
+
"AutoTokenizer": [
|
6 |
+
"tokenization_qwen.QWenTokenizer",
|
7 |
+
null
|
8 |
+
]
|
9 |
+
}
|
10 |
+
}
|
config.json
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "/notebooks/pub1",
|
3 |
+
"architectures": [
|
4 |
+
"LlamaForCausalLM"
|
5 |
+
],
|
6 |
+
"bos_token_id": 151643,
|
7 |
+
"eos_token_id": 151643,
|
8 |
+
"hidden_act": "silu",
|
9 |
+
"hidden_size": 4096,
|
10 |
+
"initializer_range": 0.02,
|
11 |
+
"intermediate_size": 11008,
|
12 |
+
"max_position_embeddings": 6144,
|
13 |
+
"model_type": "llama",
|
14 |
+
"num_attention_heads": 32,
|
15 |
+
"num_hidden_layers": 32,
|
16 |
+
"num_key_value_heads": 32,
|
17 |
+
"pad_token_id": 0,
|
18 |
+
"pretraining_tp": 1,
|
19 |
+
"rms_norm_eps": 1e-05,
|
20 |
+
"rope_scaling": null,
|
21 |
+
"tie_word_embeddings": false,
|
22 |
+
"torch_dtype": "float16",
|
23 |
+
"transformers_version": "4.32.0.dev0",
|
24 |
+
"use_cache": false,
|
25 |
+
"vocab_size": 151936
|
26 |
+
}
|
generation_config.json
ADDED
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"chat_format": "chatml",
|
3 |
+
"decay_bound": 0.0,
|
4 |
+
"decay_factor": 1.0,
|
5 |
+
"do_sample": true,
|
6 |
+
"eos_token_id": 151643,
|
7 |
+
"factual_nucleus_sampling": false,
|
8 |
+
"max_context_size": 1024,
|
9 |
+
"max_generate_size": 512,
|
10 |
+
"max_new_tokens": 512,
|
11 |
+
"pad_token_id": 151643,
|
12 |
+
"stop_words_ids": [
|
13 |
+
[
|
14 |
+
151643
|
15 |
+
]
|
16 |
+
],
|
17 |
+
"top_k": 0,
|
18 |
+
"top_p": 0.8,
|
19 |
+
"transformers_version": "4.32.0.dev0"
|
20 |
+
}
|
merges.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
pytorch_model-00001-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:530544f45e358bdef6e89de90b1080ae00539242bee59bd5b620c867842a068f
|
3 |
+
size 1964005971
|
pytorch_model-00002-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b0788b211ea06307fc12af22c67f40b4da7413dfe3d3b3bfba67f1562e3697b7
|
3 |
+
size 1933673729
|
pytorch_model-00003-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c11031bababa6d8628fba295ffbe78206e25b86ba746460e0fb3eda9fe98060d
|
3 |
+
size 1933673729
|
pytorch_model-00004-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3a7f9275ddb70230b630f2fdd5ea902cc651a117737bbb960c3d57c201bb9f08
|
3 |
+
size 1990296303
|
pytorch_model-00005-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:648cca7986994ccdb372dd88208f6670f12dfc6d596f2011e5b1560e48481654
|
3 |
+
size 1990296897
|
pytorch_model-00006-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cd7e821703872b76a9e4055362a449e030e6b5c1301be5464df9b37cebfc2657
|
3 |
+
size 1990296897
|
pytorch_model-00007-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2876ed1f592aabb3c02dde1441e3cf486029d02287280146c69f60dad81c46d9
|
3 |
+
size 1990296897
|
pytorch_model-00008-of-00008.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4d4cd3230172a536631e7e29e72508a7fb517cef05d94f3bac8f59fa67da6ca3
|
3 |
+
size 1649439207
|
pytorch_model.bin.index.json
ADDED
@@ -0,0 +1,330 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"metadata": {
|
3 |
+
"total_size": 15441870848
|
4 |
+
},
|
5 |
+
"weight_map": {
|
6 |
+
"lm_head.weight": "pytorch_model-00008-of-00008.bin",
|
7 |
+
"model.embed_tokens.weight": "pytorch_model-00001-of-00008.bin",
|
8 |
+
"model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
|
9 |
+
"model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00008.bin",
|
10 |
+
"model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
|
11 |
+
"model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
|
12 |
+
"model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
|
13 |
+
"model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
|
14 |
+
"model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
|
15 |
+
"model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
|
16 |
+
"model.layers.0.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
|
17 |
+
"model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
|
18 |
+
"model.layers.1.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
19 |
+
"model.layers.1.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
|
20 |
+
"model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
|
21 |
+
"model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
|
22 |
+
"model.layers.1.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
23 |
+
"model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
|
24 |
+
"model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
|
25 |
+
"model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
|
26 |
+
"model.layers.1.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
|
27 |
+
"model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
|
28 |
+
"model.layers.10.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
29 |
+
"model.layers.10.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
|
30 |
+
"model.layers.10.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
|
31 |
+
"model.layers.10.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
|
32 |
+
"model.layers.10.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
33 |
+
"model.layers.10.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
|
34 |
+
"model.layers.10.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
|
35 |
+
"model.layers.10.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
|
36 |
+
"model.layers.10.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
|
37 |
+
"model.layers.10.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
|
38 |
+
"model.layers.11.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
39 |
+
"model.layers.11.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
|
40 |
+
"model.layers.11.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
|
41 |
+
"model.layers.11.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
|
42 |
+
"model.layers.11.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
43 |
+
"model.layers.11.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
|
44 |
+
"model.layers.11.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
|
45 |
+
"model.layers.11.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
|
46 |
+
"model.layers.11.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
|
47 |
+
"model.layers.11.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
|
48 |
+
"model.layers.12.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
49 |
+
"model.layers.12.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
|
50 |
+
"model.layers.12.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
|
51 |
+
"model.layers.12.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
|
52 |
+
"model.layers.12.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
53 |
+
"model.layers.12.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
|
54 |
+
"model.layers.12.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
|
55 |
+
"model.layers.12.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
|
56 |
+
"model.layers.12.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
|
57 |
+
"model.layers.12.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
|
58 |
+
"model.layers.13.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
59 |
+
"model.layers.13.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
|
60 |
+
"model.layers.13.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
|
61 |
+
"model.layers.13.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
|
62 |
+
"model.layers.13.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
63 |
+
"model.layers.13.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
|
64 |
+
"model.layers.13.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
|
65 |
+
"model.layers.13.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
|
66 |
+
"model.layers.13.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
|
67 |
+
"model.layers.13.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
|
68 |
+
"model.layers.14.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
69 |
+
"model.layers.14.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
|
70 |
+
"model.layers.14.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
|
71 |
+
"model.layers.14.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
|
72 |
+
"model.layers.14.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
73 |
+
"model.layers.14.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
|
74 |
+
"model.layers.14.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
|
75 |
+
"model.layers.14.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
|
76 |
+
"model.layers.14.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
|
77 |
+
"model.layers.14.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
|
78 |
+
"model.layers.15.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
79 |
+
"model.layers.15.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
|
80 |
+
"model.layers.15.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
|
81 |
+
"model.layers.15.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
|
82 |
+
"model.layers.15.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
|
83 |
+
"model.layers.15.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
|
84 |
+
"model.layers.15.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
|
85 |
+
"model.layers.15.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
|
86 |
+
"model.layers.15.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
|
87 |
+
"model.layers.15.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
|
88 |
+
"model.layers.16.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
89 |
+
"model.layers.16.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
|
90 |
+
"model.layers.16.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
|
91 |
+
"model.layers.16.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
|
92 |
+
"model.layers.16.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
93 |
+
"model.layers.16.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
|
94 |
+
"model.layers.16.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
|
95 |
+
"model.layers.16.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
|
96 |
+
"model.layers.16.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
|
97 |
+
"model.layers.16.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
|
98 |
+
"model.layers.17.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
99 |
+
"model.layers.17.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
|
100 |
+
"model.layers.17.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
|
101 |
+
"model.layers.17.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
|
102 |
+
"model.layers.17.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
103 |
+
"model.layers.17.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
|
104 |
+
"model.layers.17.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
|
105 |
+
"model.layers.17.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
|
106 |
+
"model.layers.17.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
|
107 |
+
"model.layers.17.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
|
108 |
+
"model.layers.18.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
109 |
+
"model.layers.18.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
|
110 |
+
"model.layers.18.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
|
111 |
+
"model.layers.18.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
|
112 |
+
"model.layers.18.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
113 |
+
"model.layers.18.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
|
114 |
+
"model.layers.18.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
|
115 |
+
"model.layers.18.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
|
116 |
+
"model.layers.18.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
|
117 |
+
"model.layers.18.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
|
118 |
+
"model.layers.19.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
119 |
+
"model.layers.19.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
|
120 |
+
"model.layers.19.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
|
121 |
+
"model.layers.19.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
|
122 |
+
"model.layers.19.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
123 |
+
"model.layers.19.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
|
124 |
+
"model.layers.19.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
|
125 |
+
"model.layers.19.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
|
126 |
+
"model.layers.19.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
|
127 |
+
"model.layers.19.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
|
128 |
+
"model.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
129 |
+
"model.layers.2.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
|
130 |
+
"model.layers.2.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
|
131 |
+
"model.layers.2.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
|
132 |
+
"model.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
133 |
+
"model.layers.2.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
|
134 |
+
"model.layers.2.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
|
135 |
+
"model.layers.2.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
|
136 |
+
"model.layers.2.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
|
137 |
+
"model.layers.2.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
|
138 |
+
"model.layers.20.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
139 |
+
"model.layers.20.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
|
140 |
+
"model.layers.20.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
|
141 |
+
"model.layers.20.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
|
142 |
+
"model.layers.20.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
|
143 |
+
"model.layers.20.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
|
144 |
+
"model.layers.20.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
|
145 |
+
"model.layers.20.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
|
146 |
+
"model.layers.20.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
|
147 |
+
"model.layers.20.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
|
148 |
+
"model.layers.21.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
149 |
+
"model.layers.21.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
|
150 |
+
"model.layers.21.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
|
151 |
+
"model.layers.21.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
|
152 |
+
"model.layers.21.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
153 |
+
"model.layers.21.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
|
154 |
+
"model.layers.21.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
|
155 |
+
"model.layers.21.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
|
156 |
+
"model.layers.21.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
|
157 |
+
"model.layers.21.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
|
158 |
+
"model.layers.22.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
159 |
+
"model.layers.22.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
|
160 |
+
"model.layers.22.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
|
161 |
+
"model.layers.22.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
|
162 |
+
"model.layers.22.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
163 |
+
"model.layers.22.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
|
164 |
+
"model.layers.22.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
|
165 |
+
"model.layers.22.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
|
166 |
+
"model.layers.22.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
|
167 |
+
"model.layers.22.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
|
168 |
+
"model.layers.23.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
169 |
+
"model.layers.23.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
|
170 |
+
"model.layers.23.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
|
171 |
+
"model.layers.23.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
|
172 |
+
"model.layers.23.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
173 |
+
"model.layers.23.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
|
174 |
+
"model.layers.23.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
|
175 |
+
"model.layers.23.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
|
176 |
+
"model.layers.23.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
|
177 |
+
"model.layers.23.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
|
178 |
+
"model.layers.24.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
179 |
+
"model.layers.24.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
|
180 |
+
"model.layers.24.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
|
181 |
+
"model.layers.24.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
|
182 |
+
"model.layers.24.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
183 |
+
"model.layers.24.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
|
184 |
+
"model.layers.24.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
|
185 |
+
"model.layers.24.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
|
186 |
+
"model.layers.24.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
|
187 |
+
"model.layers.24.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
|
188 |
+
"model.layers.25.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
189 |
+
"model.layers.25.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
|
190 |
+
"model.layers.25.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
|
191 |
+
"model.layers.25.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
|
192 |
+
"model.layers.25.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
|
193 |
+
"model.layers.25.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
|
194 |
+
"model.layers.25.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
|
195 |
+
"model.layers.25.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
|
196 |
+
"model.layers.25.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
|
197 |
+
"model.layers.25.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
|
198 |
+
"model.layers.26.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
199 |
+
"model.layers.26.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
|
200 |
+
"model.layers.26.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
|
201 |
+
"model.layers.26.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
|
202 |
+
"model.layers.26.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
203 |
+
"model.layers.26.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
|
204 |
+
"model.layers.26.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
|
205 |
+
"model.layers.26.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
|
206 |
+
"model.layers.26.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
|
207 |
+
"model.layers.26.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
|
208 |
+
"model.layers.27.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
209 |
+
"model.layers.27.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
|
210 |
+
"model.layers.27.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
|
211 |
+
"model.layers.27.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
|
212 |
+
"model.layers.27.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
213 |
+
"model.layers.27.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
|
214 |
+
"model.layers.27.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
|
215 |
+
"model.layers.27.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
|
216 |
+
"model.layers.27.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
|
217 |
+
"model.layers.27.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
|
218 |
+
"model.layers.28.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
219 |
+
"model.layers.28.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
|
220 |
+
"model.layers.28.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
|
221 |
+
"model.layers.28.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
|
222 |
+
"model.layers.28.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
223 |
+
"model.layers.28.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
|
224 |
+
"model.layers.28.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
|
225 |
+
"model.layers.28.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
|
226 |
+
"model.layers.28.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
|
227 |
+
"model.layers.28.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
|
228 |
+
"model.layers.29.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
229 |
+
"model.layers.29.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
|
230 |
+
"model.layers.29.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
|
231 |
+
"model.layers.29.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
|
232 |
+
"model.layers.29.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
233 |
+
"model.layers.29.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
|
234 |
+
"model.layers.29.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
|
235 |
+
"model.layers.29.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
|
236 |
+
"model.layers.29.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
|
237 |
+
"model.layers.29.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
|
238 |
+
"model.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
239 |
+
"model.layers.3.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
|
240 |
+
"model.layers.3.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
|
241 |
+
"model.layers.3.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
|
242 |
+
"model.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
243 |
+
"model.layers.3.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
|
244 |
+
"model.layers.3.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
|
245 |
+
"model.layers.3.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
|
246 |
+
"model.layers.3.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
|
247 |
+
"model.layers.3.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
|
248 |
+
"model.layers.30.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
249 |
+
"model.layers.30.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
|
250 |
+
"model.layers.30.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
|
251 |
+
"model.layers.30.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
|
252 |
+
"model.layers.30.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
|
253 |
+
"model.layers.30.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
|
254 |
+
"model.layers.30.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
|
255 |
+
"model.layers.30.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
|
256 |
+
"model.layers.30.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
|
257 |
+
"model.layers.30.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
|
258 |
+
"model.layers.31.input_layernorm.weight": "pytorch_model-00008-of-00008.bin",
|
259 |
+
"model.layers.31.mlp.down_proj.weight": "pytorch_model-00008-of-00008.bin",
|
260 |
+
"model.layers.31.mlp.gate_proj.weight": "pytorch_model-00008-of-00008.bin",
|
261 |
+
"model.layers.31.mlp.up_proj.weight": "pytorch_model-00008-of-00008.bin",
|
262 |
+
"model.layers.31.post_attention_layernorm.weight": "pytorch_model-00008-of-00008.bin",
|
263 |
+
"model.layers.31.self_attn.k_proj.weight": "pytorch_model-00008-of-00008.bin",
|
264 |
+
"model.layers.31.self_attn.o_proj.weight": "pytorch_model-00008-of-00008.bin",
|
265 |
+
"model.layers.31.self_attn.q_proj.weight": "pytorch_model-00008-of-00008.bin",
|
266 |
+
"model.layers.31.self_attn.rotary_emb.inv_freq": "pytorch_model-00008-of-00008.bin",
|
267 |
+
"model.layers.31.self_attn.v_proj.weight": "pytorch_model-00008-of-00008.bin",
|
268 |
+
"model.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
269 |
+
"model.layers.4.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
|
270 |
+
"model.layers.4.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
|
271 |
+
"model.layers.4.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
|
272 |
+
"model.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
273 |
+
"model.layers.4.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
|
274 |
+
"model.layers.4.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
|
275 |
+
"model.layers.4.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
|
276 |
+
"model.layers.4.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
|
277 |
+
"model.layers.4.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
|
278 |
+
"model.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
279 |
+
"model.layers.5.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
|
280 |
+
"model.layers.5.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
|
281 |
+
"model.layers.5.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
|
282 |
+
"model.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
|
283 |
+
"model.layers.5.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
|
284 |
+
"model.layers.5.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
|
285 |
+
"model.layers.5.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
|
286 |
+
"model.layers.5.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
|
287 |
+
"model.layers.5.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
|
288 |
+
"model.layers.6.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
289 |
+
"model.layers.6.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
|
290 |
+
"model.layers.6.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
|
291 |
+
"model.layers.6.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
|
292 |
+
"model.layers.6.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
293 |
+
"model.layers.6.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
|
294 |
+
"model.layers.6.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
|
295 |
+
"model.layers.6.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
|
296 |
+
"model.layers.6.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
|
297 |
+
"model.layers.6.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
|
298 |
+
"model.layers.7.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
299 |
+
"model.layers.7.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
|
300 |
+
"model.layers.7.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
|
301 |
+
"model.layers.7.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
|
302 |
+
"model.layers.7.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
303 |
+
"model.layers.7.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
|
304 |
+
"model.layers.7.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
|
305 |
+
"model.layers.7.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
|
306 |
+
"model.layers.7.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
|
307 |
+
"model.layers.7.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
|
308 |
+
"model.layers.8.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
309 |
+
"model.layers.8.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
|
310 |
+
"model.layers.8.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
|
311 |
+
"model.layers.8.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
|
312 |
+
"model.layers.8.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
313 |
+
"model.layers.8.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
|
314 |
+
"model.layers.8.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
|
315 |
+
"model.layers.8.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
|
316 |
+
"model.layers.8.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
|
317 |
+
"model.layers.8.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
|
318 |
+
"model.layers.9.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
319 |
+
"model.layers.9.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
|
320 |
+
"model.layers.9.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
|
321 |
+
"model.layers.9.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
|
322 |
+
"model.layers.9.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
|
323 |
+
"model.layers.9.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
|
324 |
+
"model.layers.9.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
|
325 |
+
"model.layers.9.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
|
326 |
+
"model.layers.9.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
|
327 |
+
"model.layers.9.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
|
328 |
+
"model.norm.weight": "pytorch_model-00008-of-00008.bin"
|
329 |
+
}
|
330 |
+
}
|
qwen.tiktoken
ADDED
The diff for this file is too large to render.
See raw diff
|
|
special_tokens_map.json
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "<|endoftext|>",
|
3 |
+
"eos_token": "<|endoftext|>",
|
4 |
+
"unk_token": "<|endoftext|>"
|
5 |
+
}
|
tokenization_qwen.py
ADDED
@@ -0,0 +1,566 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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.pyplot as plt
|
24 |
+
import matplotlib.colors as mcolors
|
25 |
+
from matplotlib.font_manager import FontProperties
|
26 |
+
|
27 |
+
logger = logging.getLogger(__name__)
|
28 |
+
|
29 |
+
|
30 |
+
VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
|
31 |
+
|
32 |
+
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+"""
|
33 |
+
ENDOFTEXT = "<|endoftext|>"
|
34 |
+
IMSTART = "<|im_start|>"
|
35 |
+
IMEND = "<|im_end|>"
|
36 |
+
# as the default behavior is changed to allow special tokens in
|
37 |
+
# regular texts, the surface forms of special tokens need to be
|
38 |
+
# as different as possible to minimize the impact
|
39 |
+
EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
|
40 |
+
SPECIAL_TOKENS = (
|
41 |
+
ENDOFTEXT,
|
42 |
+
IMSTART,
|
43 |
+
IMEND,
|
44 |
+
) + EXTRAS
|
45 |
+
IMG_TOKEN_SPAN = 256
|
46 |
+
|
47 |
+
|
48 |
+
def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
|
49 |
+
with open(tiktoken_bpe_file, "rb") as f:
|
50 |
+
contents = f.read()
|
51 |
+
return {
|
52 |
+
base64.b64decode(token): int(rank)
|
53 |
+
for token, rank in (line.split() for line in contents.splitlines() if line)
|
54 |
+
}
|
55 |
+
|
56 |
+
def _list_find(
|
57 |
+
input_list: List[Any],
|
58 |
+
candidates: Tuple[Any],
|
59 |
+
start: int = 0,
|
60 |
+
):
|
61 |
+
for i in range(start, len(input_list)):
|
62 |
+
if input_list[i] in candidates:
|
63 |
+
return i
|
64 |
+
return -1
|
65 |
+
|
66 |
+
def _replace_closed_tag(
|
67 |
+
input_tokens: List[Any],
|
68 |
+
start_tags: Union[Any, Tuple[Any]],
|
69 |
+
end_tags: Union[Any, Tuple[Any]],
|
70 |
+
inclusive_replace_func: Callable,
|
71 |
+
exclusive_replace_func: Callable = lambda x: x,
|
72 |
+
):
|
73 |
+
if isinstance(start_tags, (str, int)):
|
74 |
+
start_tags = (start_tags,)
|
75 |
+
if isinstance(end_tags, (str, int)):
|
76 |
+
end_tags = (end_tags,)
|
77 |
+
assert len(start_tags) == len(end_tags)
|
78 |
+
|
79 |
+
output_tokens = []
|
80 |
+
end = 0
|
81 |
+
while True:
|
82 |
+
start = _list_find(input_tokens, start_tags, end)
|
83 |
+
if start == -1:
|
84 |
+
break
|
85 |
+
output_tokens.extend(exclusive_replace_func(input_tokens[end : start]))
|
86 |
+
tag_idx = start_tags.index(input_tokens[start])
|
87 |
+
end = _list_find(input_tokens, (end_tags[tag_idx],), start)
|
88 |
+
if end == -1:
|
89 |
+
raise ValueError("Unclosed image token")
|
90 |
+
output_tokens.extend(inclusive_replace_func(input_tokens[start : end + 1]))
|
91 |
+
end += 1
|
92 |
+
output_tokens.extend(exclusive_replace_func(input_tokens[end : ]))
|
93 |
+
return output_tokens
|
94 |
+
|
95 |
+
class QWenTokenizer(PreTrainedTokenizer):
|
96 |
+
"""QWen tokenizer."""
|
97 |
+
|
98 |
+
vocab_files_names = VOCAB_FILES_NAMES
|
99 |
+
|
100 |
+
def __init__(
|
101 |
+
self,
|
102 |
+
vocab_file,
|
103 |
+
errors="replace",
|
104 |
+
image_start_tag='<img>',
|
105 |
+
image_end_tag='</img>',
|
106 |
+
image_pad_tag='<imgpad>',
|
107 |
+
ref_start_tag='<ref>',
|
108 |
+
ref_end_tag='</ref>',
|
109 |
+
box_start_tag='<box>',
|
110 |
+
box_end_tag='</box>',
|
111 |
+
quad_start_tag='<quad>',
|
112 |
+
quad_end_tag='</quad>',
|
113 |
+
**kwargs,
|
114 |
+
):
|
115 |
+
super().__init__(**kwargs)
|
116 |
+
self.image_start_tag = image_start_tag
|
117 |
+
self.image_end_tag = image_end_tag
|
118 |
+
self.image_pad_tag = image_pad_tag
|
119 |
+
self.ref_start_tag = ref_start_tag
|
120 |
+
self.ref_end_tag = ref_end_tag
|
121 |
+
self.box_start_tag = box_start_tag
|
122 |
+
self.box_end_tag = box_end_tag
|
123 |
+
self.quad_start_tag = quad_start_tag
|
124 |
+
self.quad_end_tag = quad_end_tag
|
125 |
+
self.IMAGE_ST = (
|
126 |
+
ref_start_tag, ref_end_tag,
|
127 |
+
box_start_tag, box_end_tag,
|
128 |
+
quad_start_tag, quad_end_tag,
|
129 |
+
image_start_tag, image_end_tag,
|
130 |
+
image_pad_tag
|
131 |
+
)
|
132 |
+
|
133 |
+
self.errors = errors # how to handle errors in decoding
|
134 |
+
|
135 |
+
self.mergeable_ranks = _load_tiktoken_bpe(vocab_file) # type: dict[bytes, int]
|
136 |
+
self.special_tokens = {
|
137 |
+
token: index
|
138 |
+
for index, token in enumerate(
|
139 |
+
SPECIAL_TOKENS + self.IMAGE_ST, start=len(self.mergeable_ranks)
|
140 |
+
)
|
141 |
+
}
|
142 |
+
self.img_start_id = self.special_tokens[self.image_start_tag]
|
143 |
+
self.img_end_id = self.special_tokens[self.image_end_tag]
|
144 |
+
self.img_pad_id = self.special_tokens[self.image_pad_tag]
|
145 |
+
self.ref_start_id = self.special_tokens[self.ref_start_tag]
|
146 |
+
self.ref_end_id = self.special_tokens[self.ref_end_tag]
|
147 |
+
self.box_start_id = self.special_tokens[self.box_start_tag]
|
148 |
+
self.box_end_id = self.special_tokens[self.box_end_tag]
|
149 |
+
self.quad_start_id = self.special_tokens[self.quad_start_tag]
|
150 |
+
self.quad_end_id = self.special_tokens[self.quad_end_tag]
|
151 |
+
|
152 |
+
enc = tiktoken.Encoding(
|
153 |
+
"Qwen",
|
154 |
+
pat_str=PAT_STR,
|
155 |
+
mergeable_ranks=self.mergeable_ranks,
|
156 |
+
special_tokens=self.special_tokens,
|
157 |
+
)
|
158 |
+
assert (
|
159 |
+
len(self.mergeable_ranks) + len(self.special_tokens) == enc.n_vocab
|
160 |
+
), f"{len(self.mergeable_ranks) + len(self.special_tokens)} != {enc.n_vocab} in encoding"
|
161 |
+
|
162 |
+
self.decoder = {
|
163 |
+
v: k for k, v in self.mergeable_ranks.items()
|
164 |
+
} # type: dict[int, bytes|str]
|
165 |
+
self.decoder.update({v: k for k, v in self.special_tokens.items()})
|
166 |
+
|
167 |
+
self.tokenizer = enc # type: tiktoken.Encoding
|
168 |
+
|
169 |
+
self.eod_id = self.tokenizer.eot_token
|
170 |
+
self.im_start_id = self.special_tokens[IMSTART]
|
171 |
+
self.im_end_id = self.special_tokens[IMEND]
|
172 |
+
|
173 |
+
def __len__(self) -> int:
|
174 |
+
return self.tokenizer.n_vocab
|
175 |
+
|
176 |
+
def get_vocab(self) -> Dict[bytes, int]:
|
177 |
+
return self.mergeable_ranks
|
178 |
+
|
179 |
+
def convert_tokens_to_ids(
|
180 |
+
self, tokens: Union[bytes, str, List[Union[bytes, str]]]
|
181 |
+
) -> List[int]:
|
182 |
+
ids = []
|
183 |
+
if isinstance(tokens, (str, bytes)):
|
184 |
+
if tokens in self.special_tokens:
|
185 |
+
return self.special_tokens[tokens]
|
186 |
+
else:
|
187 |
+
return self.mergeable_ranks.get(tokens)
|
188 |
+
for token in tokens:
|
189 |
+
if token in self.special_tokens:
|
190 |
+
ids.append(self.special_tokens[token])
|
191 |
+
else:
|
192 |
+
ids.append(self.mergeable_ranks.get(token))
|
193 |
+
return ids
|
194 |
+
|
195 |
+
def _add_tokens(self, new_tokens: Union[List[str], List[AddedToken]], special_tokens: bool = False) -> int:
|
196 |
+
if not special_tokens and new_tokens:
|
197 |
+
raise ValueError('Adding regular tokens is not supported')
|
198 |
+
for token in new_tokens:
|
199 |
+
surface_form = token.content if isinstance(token, AddedToken) else token
|
200 |
+
if surface_form not in SPECIAL_TOKENS + self.IMAGE_ST:
|
201 |
+
raise ValueError('Adding unknown special tokens is not supported')
|
202 |
+
return 0
|
203 |
+
|
204 |
+
def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
|
205 |
+
"""
|
206 |
+
Save only the vocabulary of the tokenizer (vocabulary).
|
207 |
+
|
208 |
+
Returns:
|
209 |
+
`Tuple(str)`: Paths to the files saved.
|
210 |
+
"""
|
211 |
+
file_path = os.path.join(save_directory, "qwen.tiktoken")
|
212 |
+
with open(file_path, "w", encoding="utf8") as w:
|
213 |
+
for k, v in self.mergeable_ranks.items():
|
214 |
+
line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
|
215 |
+
w.write(line)
|
216 |
+
return (file_path,)
|
217 |
+
|
218 |
+
def tokenize(
|
219 |
+
self,
|
220 |
+
text: str,
|
221 |
+
allowed_special: Union[Set, str] = "all",
|
222 |
+
disallowed_special: Union[Collection, str] = (),
|
223 |
+
**kwargs,
|
224 |
+
) -> List[Union[bytes, str]]:
|
225 |
+
"""
|
226 |
+
Converts a string in a sequence of tokens.
|
227 |
+
|
228 |
+
Args:
|
229 |
+
text (`str`):
|
230 |
+
The sequence to be encoded.
|
231 |
+
allowed_special (`Literal["all"]` or `set`):
|
232 |
+
The surface forms of the tokens to be encoded as special tokens in regular texts.
|
233 |
+
Default to "all".
|
234 |
+
disallowed_special (`Literal["all"]` or `Collection`):
|
235 |
+
The surface forms of the tokens that should not be in regular texts and trigger errors.
|
236 |
+
Default to an empty tuple.
|
237 |
+
|
238 |
+
kwargs (additional keyword arguments, *optional*):
|
239 |
+
Will be passed to the underlying model specific encode method.
|
240 |
+
|
241 |
+
Returns:
|
242 |
+
`List[bytes|str]`: The list of tokens.
|
243 |
+
"""
|
244 |
+
tokens = []
|
245 |
+
text = unicodedata.normalize("NFC", text)
|
246 |
+
|
247 |
+
# this implementation takes a detour: text -> token id -> token surface forms
|
248 |
+
for t in self.tokenizer.encode(
|
249 |
+
text, allowed_special=allowed_special, disallowed_special=disallowed_special
|
250 |
+
):
|
251 |
+
tokens.append(self.decoder[t])
|
252 |
+
|
253 |
+
def _encode_imgurl(img_tokens):
|
254 |
+
assert img_tokens[0] == self.image_start_tag and img_tokens[-1] == self.image_end_tag
|
255 |
+
img_tokens = img_tokens[1:-1]
|
256 |
+
img_url = b''.join(img_tokens)
|
257 |
+
out_img_tokens = list(map(self.decoder.get, img_url))
|
258 |
+
if len(out_img_tokens) > IMG_TOKEN_SPAN:
|
259 |
+
raise ValueError("The content in {}..{} is too long".format(
|
260 |
+
self.image_start_tag, self.image_end_tag))
|
261 |
+
out_img_tokens.extend([self.image_pad_tag] * (IMG_TOKEN_SPAN - len(out_img_tokens)))
|
262 |
+
out_img_tokens = [self.image_start_tag] + out_img_tokens + [self.image_end_tag]
|
263 |
+
return out_img_tokens
|
264 |
+
|
265 |
+
return _replace_closed_tag(tokens, self.image_start_tag, self.image_end_tag, _encode_imgurl)
|
266 |
+
|
267 |
+
def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
|
268 |
+
"""
|
269 |
+
Converts a sequence of tokens in a single string.
|
270 |
+
"""
|
271 |
+
text = ""
|
272 |
+
temp = b""
|
273 |
+
for t in tokens:
|
274 |
+
if isinstance(t, str):
|
275 |
+
if temp:
|
276 |
+
text += temp.decode("utf-8", errors=self.errors)
|
277 |
+
temp = b""
|
278 |
+
text += t
|
279 |
+
elif isinstance(t, bytes):
|
280 |
+
temp += t
|
281 |
+
else:
|
282 |
+
raise TypeError("token should only be of type types or str")
|
283 |
+
if temp:
|
284 |
+
text += temp.decode("utf-8", errors=self.errors)
|
285 |
+
return text
|
286 |
+
|
287 |
+
@property
|
288 |
+
def vocab_size(self):
|
289 |
+
return self.tokenizer.n_vocab
|
290 |
+
|
291 |
+
def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
|
292 |
+
"""Converts an id to a token, special tokens included"""
|
293 |
+
if index in self.decoder:
|
294 |
+
return self.decoder[index]
|
295 |
+
raise ValueError("unknown ids")
|
296 |
+
|
297 |
+
def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
|
298 |
+
"""Converts a token to an id using the vocab, special tokens included"""
|
299 |
+
if token in self.special_tokens:
|
300 |
+
return self.special_tokens[token]
|
301 |
+
if token in self.mergeable_ranks:
|
302 |
+
return self.mergeable_ranks[token]
|
303 |
+
raise ValueError("unknown token")
|
304 |
+
|
305 |
+
def _tokenize(self, text: str, **kwargs):
|
306 |
+
"""
|
307 |
+
Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
|
308 |
+
vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
|
309 |
+
|
310 |
+
Do NOT take care of added tokens.
|
311 |
+
"""
|
312 |
+
raise NotImplementedError
|
313 |
+
|
314 |
+
def _decode(
|
315 |
+
self,
|
316 |
+
token_ids: Union[int, List[int]],
|
317 |
+
skip_special_tokens: bool = False,
|
318 |
+
errors: str = None,
|
319 |
+
**kwargs,
|
320 |
+
) -> str:
|
321 |
+
if isinstance(token_ids, int):
|
322 |
+
token_ids = [token_ids]
|
323 |
+
|
324 |
+
def _decode_imgurl(img_token_ids):
|
325 |
+
assert img_token_ids[0] == self.img_start_id and img_token_ids[-1] == self.img_end_id
|
326 |
+
img_token_ids = img_token_ids[1:-1]
|
327 |
+
img_token_ids = img_token_ids[ : img_token_ids.index(self.img_pad_id)]
|
328 |
+
img_url = bytes(img_token_ids).decode('utf-8')
|
329 |
+
return [self.img_start_id] + self.tokenizer.encode(img_url) + [self.img_end_id]
|
330 |
+
|
331 |
+
token_ids = _replace_closed_tag(token_ids, self.img_start_id, self.img_end_id, _decode_imgurl)
|
332 |
+
|
333 |
+
if skip_special_tokens:
|
334 |
+
token_ids = [i for i in token_ids if i < self.eod_id]
|
335 |
+
return self.tokenizer.decode(token_ids, errors=errors or self.errors)
|
336 |
+
|
337 |
+
def to_list_format(self, text: str):
|
338 |
+
text = unicodedata.normalize("NFC", text)
|
339 |
+
token_ids = self.tokenizer.encode(
|
340 |
+
text, allowed_special=set(self.IMAGE_ST + (ENDOFTEXT,)))
|
341 |
+
|
342 |
+
def _encode_vl_info(tokens):
|
343 |
+
if len(tokens) == 0:
|
344 |
+
return []
|
345 |
+
if tokens[0] == self.img_start_id and tokens[-1] == self.img_end_id:
|
346 |
+
key = 'image'
|
347 |
+
elif tokens[0] == self.ref_start_id and tokens[-1] == self.ref_end_id:
|
348 |
+
key = 'ref'
|
349 |
+
elif tokens[0] == self.box_start_id and tokens[-1] == self.box_end_id:
|
350 |
+
key = 'box'
|
351 |
+
elif tokens[0] == self.quad_start_id and tokens[-1] == self.quad_end_id:
|
352 |
+
key = 'quad'
|
353 |
+
else:
|
354 |
+
_tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
|
355 |
+
return [{'text': b''.join(map(_tobytes, map(self.decoder.get, tokens))).decode('utf-8')}]
|
356 |
+
val = b''.join(map(self.decoder.get, tokens[1:-1])).decode('utf-8')
|
357 |
+
return [{key: val}]
|
358 |
+
|
359 |
+
return _replace_closed_tag(
|
360 |
+
token_ids,
|
361 |
+
(self.img_start_id, self.ref_start_id, self.box_start_id, self.quad_start_id),
|
362 |
+
(self.img_end_id, self.ref_end_id, self.box_end_id, self.quad_end_id),
|
363 |
+
_encode_vl_info,
|
364 |
+
_encode_vl_info,
|
365 |
+
)
|
366 |
+
|
367 |
+
def from_list_format(self, list_format: List[Dict]):
|
368 |
+
text = ''
|
369 |
+
num_images = 0
|
370 |
+
for ele in list_format:
|
371 |
+
if 'image' in ele:
|
372 |
+
num_images += 1
|
373 |
+
text += f'Picture {num_images}:'
|
374 |
+
text += self.image_start_tag + ele['image'] + self.image_end_tag
|
375 |
+
text += '\n'
|
376 |
+
elif 'text' in ele:
|
377 |
+
text += ele['text']
|
378 |
+
elif 'box' in ele:
|
379 |
+
if 'ref' in ele:
|
380 |
+
text += self.ref_start_tag + ele['ref'] + self.ref_end_tag
|
381 |
+
for box in ele['box']:
|
382 |
+
text += self.box_start_tag + '(%d,%d),(%d,%d)' % (box[0], box[1], box[2], box[3]) + self.box_end_tag
|
383 |
+
else:
|
384 |
+
raise ValueError("Unsupport element: " + str(ele))
|
385 |
+
return text
|
386 |
+
|
387 |
+
def _fetch_latest_picture(self, response, history):
|
388 |
+
if history is None:
|
389 |
+
history = []
|
390 |
+
_history = history + [(response, None)]
|
391 |
+
for q, r in _history[::-1]:
|
392 |
+
for ele in self.to_list_format(q)[::-1]:
|
393 |
+
if 'image' in ele:
|
394 |
+
return ele['image']
|
395 |
+
return None
|
396 |
+
|
397 |
+
def _fetch_all_box_with_ref(self, text):
|
398 |
+
list_format = self.to_list_format(text)
|
399 |
+
output = []
|
400 |
+
for i, ele in enumerate(list_format):
|
401 |
+
if 'box' in ele:
|
402 |
+
bbox = tuple(map(int, ele['box'].replace('(', '').replace(')', '').split(',')))
|
403 |
+
assert len(bbox) == 4
|
404 |
+
output.append({'box': bbox})
|
405 |
+
if i > 0 and 'ref' in list_format[i-1]:
|
406 |
+
output[-1]['ref'] = list_format[i-1]['ref'].strip()
|
407 |
+
return output
|
408 |
+
|
409 |
+
def draw_bbox_on_latest_picture(
|
410 |
+
self,
|
411 |
+
response,
|
412 |
+
history=None,
|
413 |
+
) -> Optional[Image.Image]:
|
414 |
+
image = self._fetch_latest_picture(response, history)
|
415 |
+
if image is None:
|
416 |
+
return None
|
417 |
+
if image.startswith("http://") or image.startswith("https://"):
|
418 |
+
image = Image.open(requests.get(image, stream=True).raw).convert("RGB")
|
419 |
+
h, w = image.height, image.width
|
420 |
+
else:
|
421 |
+
image = plt.imread(image)
|
422 |
+
h, w = image.shape[0], image.shape[1]
|
423 |
+
visualizer = Visualizer(image)
|
424 |
+
|
425 |
+
boxes = self._fetch_all_box_with_ref(response)
|
426 |
+
if not boxes:
|
427 |
+
return None
|
428 |
+
color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()]) # init color
|
429 |
+
for box in boxes:
|
430 |
+
if 'ref' in box: # random new color for new refexps
|
431 |
+
color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()])
|
432 |
+
x1, y1, x2, y2 = box['box']
|
433 |
+
x1, y1, x2, y2 = (int(x1 / 1000 * w), int(y1 / 1000 * h), int(x2 / 1000 * w), int(y2 / 1000 * h))
|
434 |
+
visualizer.draw_box((x1, y1, x2, y2), alpha=1, edge_color=color)
|
435 |
+
if 'ref' in box:
|
436 |
+
visualizer.draw_text(box['ref'], (x1, y1), color=color, horizontal_alignment="left")
|
437 |
+
return visualizer.output
|
438 |
+
|
439 |
+
|
440 |
+
import colorsys
|
441 |
+
import logging
|
442 |
+
import math
|
443 |
+
import numpy as np
|
444 |
+
import matplotlib as mpl
|
445 |
+
import matplotlib.colors as mplc
|
446 |
+
import matplotlib.figure as mplfigure
|
447 |
+
import torch
|
448 |
+
from matplotlib.backends.backend_agg import FigureCanvasAgg
|
449 |
+
from PIL import Image
|
450 |
+
import random
|
451 |
+
|
452 |
+
logger = logging.getLogger(__name__)
|
453 |
+
|
454 |
+
|
455 |
+
class VisImage:
|
456 |
+
def __init__(self, img, scale=1.0):
|
457 |
+
self.img = img
|
458 |
+
self.scale = scale
|
459 |
+
self.width, self.height = img.shape[1], img.shape[0]
|
460 |
+
self._setup_figure(img)
|
461 |
+
|
462 |
+
def _setup_figure(self, img):
|
463 |
+
fig = mplfigure.Figure(frameon=False)
|
464 |
+
self.dpi = fig.get_dpi()
|
465 |
+
# add a small 1e-2 to avoid precision lost due to matplotlib's truncation
|
466 |
+
# (https://github.com/matplotlib/matplotlib/issues/15363)
|
467 |
+
fig.set_size_inches(
|
468 |
+
(self.width * self.scale + 1e-2) / self.dpi,
|
469 |
+
(self.height * self.scale + 1e-2) / self.dpi,
|
470 |
+
)
|
471 |
+
self.canvas = FigureCanvasAgg(fig)
|
472 |
+
# self.canvas = mpl.backends.backend_cairo.FigureCanvasCairo(fig)
|
473 |
+
ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
|
474 |
+
ax.axis("off")
|
475 |
+
self.fig = fig
|
476 |
+
self.ax = ax
|
477 |
+
self.reset_image(img)
|
478 |
+
|
479 |
+
def reset_image(self, img):
|
480 |
+
img = img.astype("uint8")
|
481 |
+
self.ax.imshow(img, extent=(0, self.width, self.height, 0), interpolation="nearest")
|
482 |
+
|
483 |
+
def save(self, filepath):
|
484 |
+
self.fig.savefig(filepath)
|
485 |
+
|
486 |
+
def get_image(self):
|
487 |
+
canvas = self.canvas
|
488 |
+
s, (width, height) = canvas.print_to_buffer()
|
489 |
+
|
490 |
+
buffer = np.frombuffer(s, dtype="uint8")
|
491 |
+
|
492 |
+
img_rgba = buffer.reshape(height, width, 4)
|
493 |
+
rgb, alpha = np.split(img_rgba, [3], axis=2)
|
494 |
+
return rgb.astype("uint8")
|
495 |
+
|
496 |
+
|
497 |
+
class Visualizer:
|
498 |
+
def __init__(self, img_rgb, metadata=None, scale=1.0):
|
499 |
+
self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
|
500 |
+
self.font_path = try_to_load_from_cache("Qwen/Qwen-VL-Chat", "SimSun.ttf")
|
501 |
+
self.output = VisImage(self.img, scale=scale)
|
502 |
+
self.cpu_device = torch.device("cpu")
|
503 |
+
|
504 |
+
# too small texts are useless, therefore clamp to 14
|
505 |
+
self._default_font_size = max(
|
506 |
+
np.sqrt(self.output.height * self.output.width) // 30, 15 // scale
|
507 |
+
)
|
508 |
+
|
509 |
+
def draw_text(
|
510 |
+
self,
|
511 |
+
text,
|
512 |
+
position,
|
513 |
+
*,
|
514 |
+
font_size=None,
|
515 |
+
color="g",
|
516 |
+
horizontal_alignment="center",
|
517 |
+
rotation=0,
|
518 |
+
):
|
519 |
+
if not font_size:
|
520 |
+
font_size = self._default_font_size
|
521 |
+
|
522 |
+
# since the text background is dark, we don't want the text to be dark
|
523 |
+
color = np.maximum(list(mplc.to_rgb(color)), 0.2)
|
524 |
+
color[np.argmax(color)] = max(0.8, np.max(color))
|
525 |
+
|
526 |
+
x, y = position
|
527 |
+
self.output.ax.text(
|
528 |
+
x,
|
529 |
+
y,
|
530 |
+
text,
|
531 |
+
size=font_size * self.output.scale,
|
532 |
+
fontproperties=FontProperties(fname=self.font_path),
|
533 |
+
bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
|
534 |
+
verticalalignment="top",
|
535 |
+
horizontalalignment=horizontal_alignment,
|
536 |
+
color=color,
|
537 |
+
zorder=10,
|
538 |
+
rotation=rotation,
|
539 |
+
)
|
540 |
+
return self.output
|
541 |
+
|
542 |
+
def draw_box(self, box_coord, alpha=0.5, edge_color="g", line_style="-"):
|
543 |
+
|
544 |
+
x0, y0, x1, y1 = box_coord
|
545 |
+
width = x1 - x0
|
546 |
+
height = y1 - y0
|
547 |
+
|
548 |
+
linewidth = max(self._default_font_size / 4, 1)
|
549 |
+
|
550 |
+
self.output.ax.add_patch(
|
551 |
+
mpl.patches.Rectangle(
|
552 |
+
(x0, y0),
|
553 |
+
width,
|
554 |
+
height,
|
555 |
+
fill=False,
|
556 |
+
edgecolor=edge_color,
|
557 |
+
linewidth=linewidth * self.output.scale,
|
558 |
+
alpha=alpha,
|
559 |
+
linestyle=line_style,
|
560 |
+
)
|
561 |
+
)
|
562 |
+
return self.output
|
563 |
+
|
564 |
+
def get_output(self):
|
565 |
+
|
566 |
+
return self.output
|
tokenizer_config.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_max_length": 999999999999999999,
|
3 |
+
"tokenizer_class": "QWenTokenizer",
|
4 |
+
"auto_map": {
|
5 |
+
"AutoTokenizer": [
|
6 |
+
"tokenization_qwen.QWenTokenizer",
|
7 |
+
null
|
8 |
+
]
|
9 |
+
}
|
10 |
+
}
|
vision.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f293b74085184879a81aa6b67301fd7156ffff8eaee4fa4ea65a1f75a4b44f5d
|
3 |
+
size 3871409417
|
visual.py
ADDED
@@ -0,0 +1,428 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 query is 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):
|
417 |
+
images = []
|
418 |
+
for image_path in image_paths:
|
419 |
+
if isinstance(image_path, Image.Image):
|
420 |
+
image = image_path
|
421 |
+
elif image_path.startswith("http://") or image_path.startswith("https://"):
|
422 |
+
image = Image.open(requests.get(image_path, stream=True).raw)
|
423 |
+
else:
|
424 |
+
image = Image.open(image_path)
|
425 |
+
image = image.convert("RGB")
|
426 |
+
images.append(self.image_transform(image))
|
427 |
+
images = torch.stack(images, dim=0)
|
428 |
+
return self(images)
|
vocab.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|