cutycat2000
commited on
Commit
•
71eebb5
1
Parent(s):
d19747c
Delete conversation.py
Browse files- conversation.py +0 -200
conversation.py
DELETED
@@ -1,200 +0,0 @@
|
|
1 |
-
import dataclasses
|
2 |
-
from enum import auto, Enum
|
3 |
-
from typing import List, Tuple, Any
|
4 |
-
|
5 |
-
|
6 |
-
class SeparatorStyle(Enum):
|
7 |
-
"""Different separator style."""
|
8 |
-
|
9 |
-
ADD_COLON_SINGLE = auto()
|
10 |
-
ADD_COLON_TWO = auto()
|
11 |
-
NO_COLON_SINGLE = auto()
|
12 |
-
BAIZE = auto()
|
13 |
-
PHOENIX = auto()
|
14 |
-
|
15 |
-
|
16 |
-
@dataclasses.dataclass
|
17 |
-
class Conversation:
|
18 |
-
"""A class that keeps all conversation history."""
|
19 |
-
|
20 |
-
# System prompts
|
21 |
-
system: str
|
22 |
-
# Two roles
|
23 |
-
roles: List[str]
|
24 |
-
# All messages
|
25 |
-
messages: List[List[str]]
|
26 |
-
# Offset of few shot examples
|
27 |
-
offset: int
|
28 |
-
# Separator
|
29 |
-
sep_style: SeparatorStyle
|
30 |
-
sep: str
|
31 |
-
sep2: str = None
|
32 |
-
# Stop criteria (the default one is EOS token)
|
33 |
-
stop_str: str = None
|
34 |
-
# Stops generation if meeting any token in this list
|
35 |
-
stop_token_ids: List[int] = None
|
36 |
-
|
37 |
-
# Used for the state in the gradio servers.
|
38 |
-
# TODO(lmzheng): refactor this
|
39 |
-
conv_id: Any = None
|
40 |
-
skip_next: bool = False
|
41 |
-
model_name: str = None
|
42 |
-
|
43 |
-
def get_prompt(self):
|
44 |
-
if self.sep_style == SeparatorStyle.ADD_COLON_SINGLE:
|
45 |
-
ret = self.system + self.sep
|
46 |
-
for role, message in self.messages:
|
47 |
-
if message:
|
48 |
-
ret += role + ": " + message + self.sep
|
49 |
-
else:
|
50 |
-
ret += role + ": "
|
51 |
-
return ret
|
52 |
-
elif self.sep_style == SeparatorStyle.ADD_COLON_TWO:
|
53 |
-
seps = [self.sep, self.sep2]
|
54 |
-
ret = self.system + seps[0]
|
55 |
-
for i, (role, message) in enumerate(self.messages):
|
56 |
-
if message:
|
57 |
-
ret += role + ": " + message + seps[i % 2]
|
58 |
-
else:
|
59 |
-
ret += role + ": "
|
60 |
-
return ret
|
61 |
-
elif self.sep_style == SeparatorStyle.NO_COLON_SINGLE:
|
62 |
-
ret = self.system
|
63 |
-
for role, message in self.messages:
|
64 |
-
if message:
|
65 |
-
ret += role + message + self.sep
|
66 |
-
else:
|
67 |
-
ret += role
|
68 |
-
return ret
|
69 |
-
elif self.sep_style == SeparatorStyle.BAIZE:
|
70 |
-
ret = self.system + "\n"
|
71 |
-
for role, message in self.messages:
|
72 |
-
if message:
|
73 |
-
ret += role + message + "\n"
|
74 |
-
else:
|
75 |
-
ret += role
|
76 |
-
return ret
|
77 |
-
elif self.sep_style == SeparatorStyle.PHOENIX:
|
78 |
-
ret = self.system
|
79 |
-
for role, message in self.messages:
|
80 |
-
if message:
|
81 |
-
ret += role + ": " + "<s>" + message + "</s>"
|
82 |
-
else:
|
83 |
-
ret += role + ": " + "<s>"
|
84 |
-
return ret
|
85 |
-
else:
|
86 |
-
raise ValueError(f"Invalid style: {self.sep_style}")
|
87 |
-
|
88 |
-
def append_message(self, role, message):
|
89 |
-
self.messages.append([role, message])
|
90 |
-
|
91 |
-
def to_gradio_chatbot(self):
|
92 |
-
ret = []
|
93 |
-
for i, (role, msg) in enumerate(self.messages[self.offset:]):
|
94 |
-
if i % 2 == 0:
|
95 |
-
ret.append([msg, None])
|
96 |
-
else:
|
97 |
-
ret[-1][-1] = msg
|
98 |
-
return ret
|
99 |
-
|
100 |
-
def to_openai_api_messages(self):
|
101 |
-
ret = [{"role": "system", "content": self.system}]
|
102 |
-
|
103 |
-
for i, (_, msg) in enumerate(self.messages[self.offset:]):
|
104 |
-
if i % 2 == 0:
|
105 |
-
ret.append({"role": "user", "content": msg})
|
106 |
-
else:
|
107 |
-
if msg is not None:
|
108 |
-
ret.append({"role": "assistant", "content": msg})
|
109 |
-
return ret
|
110 |
-
|
111 |
-
def copy(self):
|
112 |
-
return Conversation(
|
113 |
-
system=self.system,
|
114 |
-
roles=self.roles,
|
115 |
-
messages=[[x, y] for x, y in self.messages],
|
116 |
-
offset=self.offset,
|
117 |
-
sep_style=self.sep_style,
|
118 |
-
sep=self.sep,
|
119 |
-
sep2=self.sep2,
|
120 |
-
stop_str=self.stop_str,
|
121 |
-
stop_token_ids=self.stop_token_ids,
|
122 |
-
conv_id=self.conv_id,
|
123 |
-
model_name=self.model_name,
|
124 |
-
)
|
125 |
-
|
126 |
-
def dict(self):
|
127 |
-
return {
|
128 |
-
"system": self.system,
|
129 |
-
"roles": self.roles,
|
130 |
-
"messages": self.messages,
|
131 |
-
"offset": self.offset,
|
132 |
-
"conv_id": self.conv_id,
|
133 |
-
"model_name": self.model_name,
|
134 |
-
}
|
135 |
-
|
136 |
-
|
137 |
-
conv_vicuna = Conversation(
|
138 |
-
system="A chat between a curious user and an artificial intelligence assistant. "
|
139 |
-
"The assistant gives helpful, detailed, and polite answers to the user's questions.",
|
140 |
-
roles=("USER", "ASSISTANT"),
|
141 |
-
messages=(),
|
142 |
-
offset=0,
|
143 |
-
sep_style=SeparatorStyle.ADD_COLON_TWO,
|
144 |
-
sep=" ",
|
145 |
-
sep2="</s>",
|
146 |
-
)
|
147 |
-
|
148 |
-
conv_baize = Conversation(
|
149 |
-
system="The following is a conversation between a human and an AI assistant named Baize (named after a mythical creature in Chinese folklore). Baize is an open-source AI assistant developed by UCSD and Sun Yat-Sen University. The human and the AI assistant take turns chatting. Human statements start with [|Human|] and AI assistant statements start with [|AI|]. The AI assistant always provides responses in as much detail as possible, and in Markdown format. The AI assistant always declines to engage with topics, questions and instructions related to unethical, controversial, or sensitive issues. Complete the transcript in exactly that format.\n",
|
150 |
-
roles=("[|Human|]", "[|AI|]"),
|
151 |
-
messages=(
|
152 |
-
("[|Human|]", "Hello!"),
|
153 |
-
("[|AI|]", "Hi!"),
|
154 |
-
),
|
155 |
-
offset=2,
|
156 |
-
sep_style=SeparatorStyle.BAIZE,
|
157 |
-
sep="\n",
|
158 |
-
stop_str="[|Human|]",
|
159 |
-
)
|
160 |
-
|
161 |
-
conv_phoenix = Conversation(
|
162 |
-
system="A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.\n\n",
|
163 |
-
roles=("Human", "Assistant"),
|
164 |
-
messages=(),
|
165 |
-
offset=0,
|
166 |
-
sep_style=SeparatorStyle.PHOENIX,
|
167 |
-
sep="</s>",
|
168 |
-
)
|
169 |
-
|
170 |
-
conv_chatgpt = Conversation(
|
171 |
-
system="You are a helpful assistant.",
|
172 |
-
roles=("user", "assistant"),
|
173 |
-
messages=(),
|
174 |
-
offset=0,
|
175 |
-
sep_style=None,
|
176 |
-
sep=None,
|
177 |
-
)
|
178 |
-
|
179 |
-
|
180 |
-
conv_templates = {
|
181 |
-
"vicuna": conv_vicuna,
|
182 |
-
"baize": conv_baize,
|
183 |
-
"phoenix": conv_phoenix,
|
184 |
-
"chatgpt": conv_chatgpt,
|
185 |
-
}
|
186 |
-
|
187 |
-
def get_default_conv_template(model_name):
|
188 |
-
model_name = model_name.lower()
|
189 |
-
try:
|
190 |
-
ret = conv_templates[model_name]
|
191 |
-
return ret.copy()
|
192 |
-
except:
|
193 |
-
raise NotImplementedError(f"No support for model {model_name}.")
|
194 |
-
|
195 |
-
|
196 |
-
if __name__ == "__main__":
|
197 |
-
conv = conv_templates["chatgpt"].copy()
|
198 |
-
conv.append_message(conv.roles[0], "Hello World.")
|
199 |
-
conv.append_message(conv.roles[1], None)
|
200 |
-
print([conv.get_prompt()])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|