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 | class InferenceLLMConfig(BaseModel):
"""Configuration for the inference model."""
model_name: str
base_url: str
api_key: SecretStr
api_version: str = "2024-12-01-preview" # used only if model is from azure openai
model_config = ConfigDict(arbitrary_types_allowed=True)
supports_response_schema: bool = False
temperature: Optional[float] = None
seed: int = 1729
max_tokens: Optional[int] = None
@model_validator(mode="after")
def init_client(self) -> Self:
try:
# check if the model supports structured output
self.supports_response_schema = supports_response_schema(self.model_name.split("/")[-1])
logger.debug(
f"\nModel: {self.model_name} Supports response schema: {self.supports_response_schema}"
)
except Exception as e:
# logger.exception(f"Error in initializing the LLM : {self}")
logger.error(f"Error in initializing the LLM : {e}")
raise e
return self
def load_model(self, prompt: str, schema: Type[BaseModel] = None, *args, **kwargs):
pass
@observe(as_type="generation")
async def a_generate(self, prompt: str, schema: Type[BaseModel] = None, *args, **kwargs):
messages = [{"role": "user", "content": prompt}]
return await self.a_generate_from_messages(
messages=messages, schema=schema, *args, **kwargs
)
@observe(as_type="generation")
@retry(
wait=wait_fixed(60),
stop=stop_after_attempt(6),
retry=retry_if_exception_type(
(litellm.exceptions.RateLimitError, instructor.exceptions.InstructorRetryException)
),
)
async def a_generate_from_messages(
self, messages: list, schema: Type[BaseModel] = None, *args, **kwargs
):
# check if model supports structured output
if schema:
if self.supports_response_schema:
res = await litellm.acompletion(
model=self.model_name,
api_key=self.api_key.get_secret_value(),
base_url=self.base_url,
messages=messages,
response_format=schema,
api_version=self.api_version,
)
if res.choices[0].finish_reason == "content_filter":
raise ValueError(f"Response filtred by content filter")
else:
dict_res = ast.literal_eval(res.choices[0].message.content)
return schema(**dict_res)
else:
client = instructor.from_litellm(acompletion, mode=instructor.Mode.JSON)
res, raw_completion = await client.chat.completions.create_with_completion(
model=self.model_name,
api_key=self.api_key.get_secret_value(),
base_url=self.base_url,
messages=messages,
response_model=schema,
api_version=self.api_version,
)
return res
else:
res = await litellm.acompletion(
model=self.model_name,
api_key=self.api_key.get_secret_value(),
base_url=self.base_url,
messages=messages,
api_version=self.api_version,
)
return res.choices[0].message.content
@observe(as_type="generation")
def generate(self, prompt: str, schema: Type[BaseModel] = None, *args, **kwargs):
messages = [{"role": "user", "content": prompt}]
return self.generate_from_messages(messages=messages, schema=schema, *args, **kwargs)
@observe(as_type="generation")
@retry(
wait=wait_fixed(60),
stop=stop_after_attempt(6),
retry=retry_if_exception_type(
(litellm.exceptions.RateLimitError, instructor.exceptions.InstructorRetryException)
),
)
def generate_from_messages(
self, messages: list, schema: Type[BaseModel] = None, *args, **kwargs
):
try:
# check if model supports structured output
if schema:
if self.supports_response_schema:
res = litellm.completion(
model=self.model_name,
api_key=self.api_key.get_secret_value(),
base_url=self.base_url,
messages=messages,
response_format=schema,
api_version=self.api_version,
)
if res.choices[0].finish_reason == "content_filter":
raise ValueError(f"Response filtred by content filter")
else:
dict_res = ast.literal_eval(res.choices[0].message.content)
return schema(**dict_res)
else:
client = instructor.from_litellm(completion, mode=instructor.Mode.JSON)
res, raw_completion = client.chat.completions.create_with_completion(
model=self.model_name,
api_key=self.api_key.get_secret_value(),
base_url=self.base_url,
messages=messages,
response_model=schema,
api_version=self.api_version,
)
return res
else:
res = litellm.completion(
model=self.model_name,
api_key=self.api_key.get_secret_value(),
base_url=self.base_url,
messages=messages,
api_version=self.api_version,
)
return res.choices[0].message.content
except Exception as e:
# todo handle cost if exception
logger.error(f"Error in generating response from LLM: {e}")
return None
def get_model_name(self, *args, **kwargs) -> str:
return self.model_name
|