20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224 | class MonsterLLM(CustomLLM):
"""MonsterAPI LLM.
Monster Deploy可以让您在MonsterAPI的成本优化GPU云上托管任何vLLM支持的大型语言模型（LLM），如Tinyllama、Mixtral、Phi-2等，作为rest API端点。
通过MonsterAPI与Llama索引的集成，您可以使用部署的LLM API端点来创建RAG系统或RAG机器人，用于以下用例：
- 回答您文档中的问题
- 改进您文档的内容
- 查找您文档中重要的上下文
一旦部署启动，请在部署实时后使用base_url和api_auth_token，并在下面使用它们。
注意：在使用LLama索引访问Monster Deploy LLM时，您需要创建一个带有所需模板的提示，并将编译后的提示作为输入发送。
有关更多详细信息，请参见“LLama索引提示模板使用示例”部分。
有关更多详细信息，请参见（https://developer.monsterapi.ai/docs/monster-deploy-beta）。
一旦部署启动，请在部署实时后使用base_url和api_auth_token，并在下面使用它们。
注意：在使用LLama索引访问Monster Deploy LLM时，您需要创建一个带有所需模板的提示，并将编译后的提示作为输入发送。有关更多详细信息，请参见“LLama索引提示模板使用示例”部分。
示例：
`pip install llama-index-llms-monsterapi`
```python
llm = MonsterLLM(
model="deploy-llm",
base_url="https://ecc7deb6-26e0-419b-a7f2-0deb934af29a.monsterapi.ai",
monster_api_key="a0f8a6ba-c32f-4407-af0c-169f1915490c",
temperature=0.75,
)
response = llm.complete("What is the capital of France?")
print(str(response))
```"""
model: str = Field(description="The MonsterAPI model to use.")
monster_api_key: Optional[str] = Field(description="The MonsterAPI key to use.")
max_new_tokens: int = Field(
default=DEFAULT_NUM_OUTPUTS,
description="The number of tokens to generate.",
gt=0,
)
temperature: float = Field(
default=DEFAULT_MONSTER_TEMP,
description="The temperature to use for sampling.",
gte=0.0,
lte=1.0,
)
context_window: int = Field(
default=DEFAULT_CONTEXT_WINDOW,
description="The number of context tokens available to the LLM.",
gt=0,
)
_client: Any = PrivateAttr()
def __init__(
self,
model: str,
base_url: str = "https://api.monsterapi.ai/v1",
monster_api_key: Optional[str] = None,
max_new_tokens: int = DEFAULT_NUM_OUTPUTS,
temperature: float = DEFAULT_MONSTER_TEMP,
context_window: int = DEFAULT_CONTEXT_WINDOW,
callback_manager: Optional[CallbackManager] = None,
system_prompt: Optional[str] = None,
messages_to_prompt: Optional[Callable[[Sequence[ChatMessage]], str]] = None,
completion_to_prompt: Optional[Callable[[str], str]] = None,
pydantic_program_mode: PydanticProgramMode = PydanticProgramMode.DEFAULT,
output_parser: Optional[BaseOutputParser] = None,
) -> None:
self._client, available_llms = self.initialize_client(monster_api_key, base_url)
# Check if provided model is supported
if model not in available_llms:
error_message = (
f"Model: {model} is not supported. "
f"Supported models are {available_llms}. "
"Please update monsterapiclient to see if any models are added. "
"pip install --upgrade monsterapi"
)
raise RuntimeError(error_message)
super().__init__(
model=model,
monster_api_key=monster_api_key,
max_new_tokens=max_new_tokens,
temperature=temperature,
context_window=context_window,
callback_manager=callback_manager,
system_prompt=system_prompt,
messages_to_prompt=messages_to_prompt,
completion_to_prompt=completion_to_prompt,
pydantic_program_mode=pydantic_program_mode,
output_parser=output_parser,
)
def initialize_client(
self, monster_api_key: Optional[str], base_url: Optional[str]
) -> Any:
try:
from monsterapi import client as MonsterClient
from monsterapi.InputDataModels import MODEL_TYPES
except ImportError:
raise ImportError(
"Could not import Monster API client library."
"Please install it with `pip install monsterapi`"
)
llm_models_enabled = [i for i, j in MODEL_TYPES.items() if j == "LLM"]
return MonsterClient(monster_api_key, base_url), llm_models_enabled
@classmethod
def class_name(cls) -> str:
return "MonsterLLM"
@property
def metadata(self) -> LLMMetadata:
"""获取LLM元数据。"""
return LLMMetadata(
context_window=self.context_window,
num_output=self.max_new_tokens,
model_name=self.model,
)
def _get_input_dict(self, prompt: str, **kwargs: Any) -> Dict[str, Any]:
return {
"prompt": prompt,
"temperature": self.temperature,
"max_length": self.max_new_tokens,
**kwargs,
}
@llm_chat_callback()
def chat(self, messages: Sequence[ChatMessage], **kwargs: Any) -> ChatResponse:
prompt = self.messages_to_prompt(messages)
return self.complete(prompt, formatted=True, **kwargs)
@llm_completion_callback()
def complete(
self, prompt: str, formatted: bool = False, timeout: int = 100, **kwargs: Any
) -> CompletionResponse:
if not formatted:
prompt = self.completion_to_prompt(prompt)
stream = kwargs.pop("stream", False)
if stream is True:
raise NotImplementedError(
"complete method cannot be used with stream=True, please use stream_complete method"
)
# Validate input args against input Pydantic model
input_dict = self._get_input_dict(prompt, **kwargs)
result = self._client.generate(
model=self.model, data=input_dict, timeout=timeout
)
if isinstance(result, Exception):
raise result
if isinstance(result, dict) and "error" in result:
raise RuntimeError(result["error"])
if isinstance(result, dict) and "text" in result:
if isinstance(result["text"], list):
return CompletionResponse(text=result["text"][0])
elif isinstance(result["text"], str):
return CompletionResponse(text=result["text"])
if isinstance(result, list):
return CompletionResponse(text=result[0]["text"])
raise RuntimeError("Unexpected Return please contact monsterapi support!")
@llm_completion_callback()
def stream_complete(self, prompt: str, **kwargs: Any) -> CompletionResponseGen:
if "deploy" not in self.model:
raise NotImplementedError(
"stream_complete method can only be used with deploy models for now. Support for other models will be added soon."
)
# Validate input args against input Pydantic model
input_dict = self._get_input_dict(prompt, **kwargs)
input_dict["stream"] = True
# Starting the stream
result_stream = self._client.generate(model=self.model, data=input_dict)
if isinstance(result_stream, Exception):
raise result_stream
if isinstance(result_stream, dict) and "error" in result_stream:
raise RuntimeError(result_stream["error"])
# Iterating over the generator
try:
for result in result_stream:
yield CompletionResponse(text=result[0])
except StopIteration:
pass
|