xinference 1.3.1.post1__py3-none-any.whl → 1.4.1__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of xinference might be problematic. Click here for more details.
- xinference/_compat.py +1 -0
- xinference/_version.py +3 -3
- xinference/api/restful_api.py +4 -0
- xinference/core/chat_interface.py +1 -1
- xinference/core/model.py +23 -3
- xinference/core/supervisor.py +6 -0
- xinference/core/worker.py +54 -11
- xinference/model/llm/__init__.py +7 -2
- xinference/model/llm/core.py +1 -0
- xinference/model/llm/llama_cpp/core.py +50 -15
- xinference/model/llm/llm_family.json +388 -13
- xinference/model/llm/llm_family_modelscope.json +373 -14
- xinference/model/llm/mlx/core.py +15 -11
- xinference/model/llm/reasoning_parser.py +17 -9
- xinference/model/llm/sglang/core.py +112 -12
- xinference/model/llm/transformers/core.py +4 -2
- xinference/model/llm/transformers/deepseek_vl.py +1 -1
- xinference/model/llm/transformers/deepseek_vl2.py +287 -0
- xinference/model/llm/transformers/gemma3.py +185 -0
- xinference/model/llm/transformers/intern_vl.py +0 -2
- xinference/model/llm/utils.py +62 -42
- xinference/model/llm/vllm/core.py +157 -11
- xinference/model/llm/vllm/distributed_executor.py +314 -0
- xinference/model/rerank/core.py +16 -11
- xinference/thirdparty/deepseek_vl2/__init__.py +31 -0
- xinference/thirdparty/deepseek_vl2/models/__init__.py +26 -0
- xinference/thirdparty/deepseek_vl2/models/configuration_deepseek.py +210 -0
- xinference/thirdparty/deepseek_vl2/models/conversation.py +310 -0
- xinference/thirdparty/deepseek_vl2/models/modeling_deepseek.py +1975 -0
- xinference/thirdparty/deepseek_vl2/models/modeling_deepseek_vl_v2.py +697 -0
- xinference/thirdparty/deepseek_vl2/models/processing_deepseek_vl_v2.py +675 -0
- xinference/thirdparty/deepseek_vl2/models/siglip_vit.py +661 -0
- xinference/thirdparty/deepseek_vl2/serve/__init__.py +0 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/__init__.py +0 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/gradio_utils.py +83 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/overwrites.py +81 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/presets.py +115 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/utils.py +333 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/Kelpy-Codos.js +100 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/avatar.png +0 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/custom.css +355 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/custom.js +22 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/favicon.ico +0 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/simsun.ttc +0 -0
- xinference/thirdparty/deepseek_vl2/serve/inference.py +197 -0
- xinference/thirdparty/deepseek_vl2/utils/__init__.py +18 -0
- xinference/thirdparty/deepseek_vl2/utils/io.py +80 -0
- xinference/types.py +2 -2
- xinference/web/ui/build/asset-manifest.json +6 -6
- xinference/web/ui/build/index.html +1 -1
- xinference/web/ui/build/static/css/main.b494ae7e.css +2 -0
- xinference/web/ui/build/static/css/main.b494ae7e.css.map +1 -0
- xinference/web/ui/build/static/js/main.5ca4eea1.js +3 -0
- xinference/web/ui/build/static/js/main.5ca4eea1.js.map +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/0f0967acaec5df1d45b80010949c258d64297ebbb0f44b8bb3afcbd45c6f0ec4.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/68249645124f37d01eef83b1d897e751f895bea919b6fb466f907c1f87cebc84.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/cc97b49285d7717c63374766c789141a4329a04582ab32756d7e0e614d4c5c7f.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/f199e8173f6409a5802ed44acb95f218388131136504b2e9132129e150c92f9a.json +1 -0
- xinference/web/ui/src/locales/en.json +2 -2
- xinference/web/ui/src/locales/zh.json +1 -1
- {xinference-1.3.1.post1.dist-info → xinference-1.4.1.dist-info}/METADATA +4 -4
- {xinference-1.3.1.post1.dist-info → xinference-1.4.1.dist-info}/RECORD +67 -41
- xinference/web/ui/build/static/css/main.f8177338.css +0 -2
- xinference/web/ui/build/static/css/main.f8177338.css.map +0 -1
- xinference/web/ui/build/static/js/main.55b70cb7.js +0 -3
- xinference/web/ui/build/static/js/main.55b70cb7.js.map +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/2deac8d5636974533e3714f34e94fc754f9153a07c6ee11e72846cb8eae47e4b.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/db16a983bc08a05f0439cc61ca0840e49e1d8400eef678909f16c032a418a3d6.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/e23d476fcbf6fd69c8986bf82133d257d28aa8fc9a5cab231d81c1c75c58cd99.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/e7a8c37fda8725cab69c7ef8c627060bd7fc806adc67e00fe628ba148cb86d7f.json +0 -1
- /xinference/web/ui/build/static/js/{main.55b70cb7.js.LICENSE.txt → main.5ca4eea1.js.LICENSE.txt} +0 -0
- {xinference-1.3.1.post1.dist-info → xinference-1.4.1.dist-info}/LICENSE +0 -0
- {xinference-1.3.1.post1.dist-info → xinference-1.4.1.dist-info}/WHEEL +0 -0
- {xinference-1.3.1.post1.dist-info → xinference-1.4.1.dist-info}/entry_points.txt +0 -0
- {xinference-1.3.1.post1.dist-info → xinference-1.4.1.dist-info}/top_level.txt +0 -0
|
@@ -0,0 +1,310 @@
|
|
|
1
|
+
"""
|
|
2
|
+
From https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
|
|
3
|
+
"""
|
|
4
|
+
|
|
5
|
+
import dataclasses
|
|
6
|
+
from enum import IntEnum, auto
|
|
7
|
+
from typing import Any, Dict, List
|
|
8
|
+
|
|
9
|
+
|
|
10
|
+
class SeparatorStyle(IntEnum):
|
|
11
|
+
"""Separator styles."""
|
|
12
|
+
|
|
13
|
+
DeepSeek = auto()
|
|
14
|
+
DeepSeekV2 = auto()
|
|
15
|
+
PLAIN = auto()
|
|
16
|
+
ALIGNMENT = auto()
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
@dataclasses.dataclass
|
|
20
|
+
class Conversation:
|
|
21
|
+
"""A class that manages prompt templates and keeps all conversation history."""
|
|
22
|
+
|
|
23
|
+
# The name of this template
|
|
24
|
+
name: str
|
|
25
|
+
# The template of the system prompt
|
|
26
|
+
system_template: str = "{system_message}"
|
|
27
|
+
# The system message
|
|
28
|
+
system_message: str = ""
|
|
29
|
+
# The names of two roles
|
|
30
|
+
roles: List[str] = (("USER", "ASSISTANT"),)
|
|
31
|
+
# All messages. Each item is (role, message).
|
|
32
|
+
messages: List[List[str]] = ()
|
|
33
|
+
# The number of few shot examples
|
|
34
|
+
offset: int = 0
|
|
35
|
+
# The separator style and configurations
|
|
36
|
+
sep_style: SeparatorStyle = SeparatorStyle.DeepSeek
|
|
37
|
+
sep: str = "\n"
|
|
38
|
+
sep2: str = None
|
|
39
|
+
# Stop criteria (the default one is EOS token)
|
|
40
|
+
stop_str: str = None
|
|
41
|
+
# Stops generation if meeting any token in this list
|
|
42
|
+
stop_token_ids: List[int] = None
|
|
43
|
+
|
|
44
|
+
def get_prompt(self) -> str:
|
|
45
|
+
"""Get the prompt for generation."""
|
|
46
|
+
system_prompt = self.system_template.format(system_message=self.system_message)
|
|
47
|
+
if self.sep_style == SeparatorStyle.DeepSeek:
|
|
48
|
+
seps = [self.sep, self.sep2]
|
|
49
|
+
if system_prompt == "" or system_prompt is None:
|
|
50
|
+
ret = ""
|
|
51
|
+
else:
|
|
52
|
+
ret = system_prompt + seps[0]
|
|
53
|
+
for i, (role, message) in enumerate(self.messages):
|
|
54
|
+
if message:
|
|
55
|
+
ret += role + ": " + message + seps[i % 2]
|
|
56
|
+
else:
|
|
57
|
+
ret += role + ":"
|
|
58
|
+
return ret
|
|
59
|
+
elif self.sep_style == SeparatorStyle.DeepSeekV2:
|
|
60
|
+
seps = [self.sep, self.sep2]
|
|
61
|
+
if system_prompt == "" or system_prompt is None:
|
|
62
|
+
ret = ""
|
|
63
|
+
else:
|
|
64
|
+
ret = system_prompt + seps[0]
|
|
65
|
+
for i, (role, message) in enumerate(self.messages):
|
|
66
|
+
if message:
|
|
67
|
+
if role == "User":
|
|
68
|
+
ret += "<|sft▁begin|>\n" + message + self.sep #<|sft▁begin|>User Input<|sft▁end|>\nResponse<|end▁of▁sentence|>
|
|
69
|
+
else:
|
|
70
|
+
ret += message + self.sep2
|
|
71
|
+
else:
|
|
72
|
+
ret = ret
|
|
73
|
+
return ret
|
|
74
|
+
|
|
75
|
+
elif self.sep_style == SeparatorStyle.PLAIN:
|
|
76
|
+
seps = [self.sep, self.sep2]
|
|
77
|
+
ret = ""
|
|
78
|
+
for i, (role, message) in enumerate(self.messages):
|
|
79
|
+
if message:
|
|
80
|
+
if type(message) is tuple:
|
|
81
|
+
message, _, _ = message
|
|
82
|
+
if i % 2 == 0:
|
|
83
|
+
ret += message + seps[i % 2]
|
|
84
|
+
else:
|
|
85
|
+
ret += message + seps[i % 2]
|
|
86
|
+
else:
|
|
87
|
+
ret += ""
|
|
88
|
+
return ret
|
|
89
|
+
elif self.sep_style == SeparatorStyle.ALIGNMENT:
|
|
90
|
+
seps = [self.sep, self.sep2]
|
|
91
|
+
ret = ""
|
|
92
|
+
for i, (role, message) in enumerate(self.messages):
|
|
93
|
+
if message:
|
|
94
|
+
if type(message) is tuple:
|
|
95
|
+
message, _, _ = message
|
|
96
|
+
if i % 2 == 0:
|
|
97
|
+
ret += '<image>\n' + seps[i % 2]
|
|
98
|
+
else:
|
|
99
|
+
ret += message + seps[i % 2]
|
|
100
|
+
else:
|
|
101
|
+
ret += ""
|
|
102
|
+
return ret
|
|
103
|
+
else:
|
|
104
|
+
raise ValueError(f"Invalid style: {self.sep_style}")
|
|
105
|
+
|
|
106
|
+
def set_system_message(self, system_message: str):
|
|
107
|
+
"""Set the system message."""
|
|
108
|
+
self.system_message = system_message
|
|
109
|
+
|
|
110
|
+
def append_message(self, role: str, message: str):
|
|
111
|
+
"""Append a new message."""
|
|
112
|
+
self.messages.append([role, message])
|
|
113
|
+
|
|
114
|
+
def update_last_message(self, message: str):
|
|
115
|
+
"""Update the last output.
|
|
116
|
+
|
|
117
|
+
The last message is typically set to be None when constructing the prompt,
|
|
118
|
+
so we need to update it in-place after getting the response from a model.
|
|
119
|
+
"""
|
|
120
|
+
self.messages[-1][1] = message
|
|
121
|
+
|
|
122
|
+
def reset_message(self):
|
|
123
|
+
"""Reset a new message."""
|
|
124
|
+
self.messages = []
|
|
125
|
+
|
|
126
|
+
def to_gradio_chatbot(self):
|
|
127
|
+
"""Convert the conversation to gradio chatbot format."""
|
|
128
|
+
ret = []
|
|
129
|
+
for i, (role, msg) in enumerate(self.messages[self.offset :]):
|
|
130
|
+
if i % 2 == 0:
|
|
131
|
+
ret.append([msg, None])
|
|
132
|
+
else:
|
|
133
|
+
ret[-1][-1] = msg
|
|
134
|
+
return ret
|
|
135
|
+
|
|
136
|
+
def to_openai_api_messages(self):
|
|
137
|
+
"""Convert the conversation to OpenAI chat completion format."""
|
|
138
|
+
system_prompt = self.system_template.format(system_message=self.system_message)
|
|
139
|
+
ret = [{"role": "system", "content": system_prompt}]
|
|
140
|
+
|
|
141
|
+
for i, (_, msg) in enumerate(self.messages[self.offset :]):
|
|
142
|
+
if i % 2 == 0:
|
|
143
|
+
ret.append({"role": "user", "content": msg})
|
|
144
|
+
else:
|
|
145
|
+
if msg is not None:
|
|
146
|
+
ret.append({"role": "assistant", "content": msg})
|
|
147
|
+
return ret
|
|
148
|
+
|
|
149
|
+
def copy(self):
|
|
150
|
+
return Conversation(
|
|
151
|
+
name=self.name,
|
|
152
|
+
system_template=self.system_template,
|
|
153
|
+
system_message=self.system_message,
|
|
154
|
+
roles=self.roles,
|
|
155
|
+
messages=[[x, y] for x, y in self.messages],
|
|
156
|
+
offset=self.offset,
|
|
157
|
+
sep_style=self.sep_style,
|
|
158
|
+
sep=self.sep,
|
|
159
|
+
sep2=self.sep2,
|
|
160
|
+
stop_str=self.stop_str,
|
|
161
|
+
stop_token_ids=self.stop_token_ids,
|
|
162
|
+
)
|
|
163
|
+
|
|
164
|
+
def dict(self):
|
|
165
|
+
return {
|
|
166
|
+
"template_name": self.name,
|
|
167
|
+
"system_message": self.system_message,
|
|
168
|
+
"roles": self.roles,
|
|
169
|
+
"messages": self.messages,
|
|
170
|
+
"offset": self.offset,
|
|
171
|
+
}
|
|
172
|
+
|
|
173
|
+
|
|
174
|
+
# A global registry for all conversation templates
|
|
175
|
+
conv_templates: Dict[str, Conversation] = {}
|
|
176
|
+
|
|
177
|
+
|
|
178
|
+
def register_conv_template(template: Conversation, override: bool = False):
|
|
179
|
+
"""Register a new conversation template."""
|
|
180
|
+
if not override:
|
|
181
|
+
assert template.name not in conv_templates, f"{template.name} has been registered."
|
|
182
|
+
|
|
183
|
+
conv_templates[template.name] = template
|
|
184
|
+
|
|
185
|
+
|
|
186
|
+
def get_conv_template(name: str) -> Conversation:
|
|
187
|
+
"""Get a conversation template."""
|
|
188
|
+
return conv_templates[name].copy()
|
|
189
|
+
|
|
190
|
+
|
|
191
|
+
# register_conv_template(
|
|
192
|
+
# Conversation(
|
|
193
|
+
# name="deepseek",
|
|
194
|
+
# system_template="{system_message}",
|
|
195
|
+
# # system_message="You are a helpful assistant. Please answer truthfully and write out your "
|
|
196
|
+
# # "thinking step by step to be sure you get the right answer.",
|
|
197
|
+
# system_message="",
|
|
198
|
+
# roles=("User", "Assistant"),
|
|
199
|
+
# messages=(),
|
|
200
|
+
# offset=0,
|
|
201
|
+
# sep_style=SeparatorStyle.DeepSeek,
|
|
202
|
+
# sep="\n\n",
|
|
203
|
+
# sep2="<|end▁of▁sentence|>",
|
|
204
|
+
# stop_token_ids=[100001],
|
|
205
|
+
# stop_str=["User:", "<|end▁of▁sentence|>"]
|
|
206
|
+
# )
|
|
207
|
+
# )
|
|
208
|
+
register_conv_template(
|
|
209
|
+
Conversation(
|
|
210
|
+
name="deepseek",
|
|
211
|
+
system_template="{system_message}",
|
|
212
|
+
# system_message="You are a helpful assistant. Please answer truthfully and write out your "
|
|
213
|
+
# "thinking step by step to be sure you get the right answer.",
|
|
214
|
+
system_message="",
|
|
215
|
+
roles=("<|User|>", "<|Assistant|>"),
|
|
216
|
+
messages=(),
|
|
217
|
+
offset=0,
|
|
218
|
+
sep_style=SeparatorStyle.DeepSeek,
|
|
219
|
+
sep="\n\n",
|
|
220
|
+
sep2="<|end▁of▁sentence|>",
|
|
221
|
+
stop_token_ids=[100001],
|
|
222
|
+
stop_str=["User:", "<|end▁of▁sentence|>"]
|
|
223
|
+
)
|
|
224
|
+
)
|
|
225
|
+
# register_conv_template(
|
|
226
|
+
# Conversation(
|
|
227
|
+
# name="deepseekv2",
|
|
228
|
+
# system_template="{system_message}",
|
|
229
|
+
# system_message="",
|
|
230
|
+
# roles=("User", "Assistant"),
|
|
231
|
+
# messages=(),
|
|
232
|
+
# offset=0,
|
|
233
|
+
# sep_style=SeparatorStyle.DeepSeekV2,
|
|
234
|
+
# sep="\n<|sft▁end|>",
|
|
235
|
+
# sep2="<|end▁of▁sentence|>",
|
|
236
|
+
# stop_token_ids=[100001],
|
|
237
|
+
# stop_str=["User:", "<|end▁of▁sentence|>"]
|
|
238
|
+
# )
|
|
239
|
+
# )
|
|
240
|
+
register_conv_template(
|
|
241
|
+
Conversation(
|
|
242
|
+
name="deepseekv2",
|
|
243
|
+
system_template="{system_message}",
|
|
244
|
+
system_message="",
|
|
245
|
+
roles=("|<User>|", "|<Assistant>|"),
|
|
246
|
+
messages=(),
|
|
247
|
+
offset=0,
|
|
248
|
+
sep_style=SeparatorStyle.DeepSeekV2,
|
|
249
|
+
sep="\n<|sft▁end|>",
|
|
250
|
+
sep2="<|end▁of▁sentence|>",
|
|
251
|
+
stop_token_ids=[100001],
|
|
252
|
+
stop_str=["User:", "<|end▁of▁sentence|>"]
|
|
253
|
+
)
|
|
254
|
+
)
|
|
255
|
+
|
|
256
|
+
|
|
257
|
+
register_conv_template(
|
|
258
|
+
Conversation(
|
|
259
|
+
name="plain",
|
|
260
|
+
system_template="",
|
|
261
|
+
system_message="",
|
|
262
|
+
roles=("", ""),
|
|
263
|
+
messages=(),
|
|
264
|
+
offset=0,
|
|
265
|
+
sep_style=SeparatorStyle.PLAIN,
|
|
266
|
+
sep="",
|
|
267
|
+
sep2="",
|
|
268
|
+
stop_token_ids=[100001],
|
|
269
|
+
stop_str=['</s>'],
|
|
270
|
+
)
|
|
271
|
+
)
|
|
272
|
+
|
|
273
|
+
|
|
274
|
+
register_conv_template(
|
|
275
|
+
Conversation(
|
|
276
|
+
name="alignment",
|
|
277
|
+
system_template="",
|
|
278
|
+
system_message="",
|
|
279
|
+
roles=("", ""),
|
|
280
|
+
messages=(),
|
|
281
|
+
offset=0,
|
|
282
|
+
sep_style=SeparatorStyle.ALIGNMENT,
|
|
283
|
+
sep="",
|
|
284
|
+
sep2="",
|
|
285
|
+
stop_token_ids=[100001],
|
|
286
|
+
stop_str=['</s>'],
|
|
287
|
+
)
|
|
288
|
+
)
|
|
289
|
+
|
|
290
|
+
|
|
291
|
+
if __name__ == "__main__":
|
|
292
|
+
print("deepseek template:")
|
|
293
|
+
conv = get_conv_template("deepseek")
|
|
294
|
+
conv.append_message(conv.roles[0], "Hello!")
|
|
295
|
+
conv.append_message(conv.roles[1], "Hi! This is Tony.")
|
|
296
|
+
conv.append_message(conv.roles[0], "Who are you?")
|
|
297
|
+
conv.append_message(conv.roles[1], "I am a helpful assistant.")
|
|
298
|
+
conv.append_message(conv.roles[0], "How are you?")
|
|
299
|
+
conv.append_message(conv.roles[1], None)
|
|
300
|
+
print(conv.get_prompt())
|
|
301
|
+
|
|
302
|
+
print("deepseekv2 template:")
|
|
303
|
+
conv = get_conv_template("deepseekv2")
|
|
304
|
+
conv.append_message(conv.roles[0], "Hello!")
|
|
305
|
+
conv.append_message(conv.roles[1], "Hi! This is Tony.")
|
|
306
|
+
conv.append_message(conv.roles[0], "Who are you?")
|
|
307
|
+
conv.append_message(conv.roles[1], "I am a helpful assistant.")
|
|
308
|
+
conv.append_message(conv.roles[0], "How are you?")
|
|
309
|
+
conv.append_message(conv.roles[1], None)
|
|
310
|
+
print(conv.get_prompt())
|