1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
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
|
import logging
from typing import Any, Generator, Union
from r2r.base import (
LLMChatCompletion,
LLMChatCompletionChunk,
LLMConfig,
LLMProvider,
)
from r2r.base.abstractions.llm import GenerationConfig
logger = logging.getLogger(__name__)
class LiteLLM(LLMProvider):
"""A concrete class for creating LiteLLM models."""
def __init__(
self,
config: LLMConfig,
*args,
**kwargs,
) -> None:
try:
from litellm import acompletion, completion
self.litellm_completion = completion
self.litellm_acompletion = acompletion
except ImportError:
raise ImportError(
"Error, `litellm` is required to run a LiteLLM. Please install it using `pip install litellm`."
)
super().__init__(config)
def get_completion(
self,
messages: list[dict],
generation_config: GenerationConfig,
**kwargs,
) -> LLMChatCompletion:
if generation_config.stream:
raise ValueError(
"Stream must be set to False to use the `get_completion` method."
)
return self._get_completion(messages, generation_config, **kwargs)
def get_completion_stream(
self,
messages: list[dict],
generation_config: GenerationConfig,
**kwargs,
) -> Generator[LLMChatCompletionChunk, None, None]:
if not generation_config.stream:
raise ValueError(
"Stream must be set to True to use the `get_completion_stream` method."
)
return self._get_completion(messages, generation_config, **kwargs)
def extract_content(self, response: LLMChatCompletion) -> str:
return response.choices[0].message.content
def _get_completion(
self,
messages: list[dict],
generation_config: GenerationConfig,
**kwargs,
) -> Union[
LLMChatCompletion, Generator[LLMChatCompletionChunk, None, None]
]:
# Create a dictionary with the default arguments
args = self._get_base_args(generation_config)
args["messages"] = messages
# Conditionally add the 'functions' argument if it's not None
if generation_config.functions is not None:
args["functions"] = generation_config.functions
args = {**args, **kwargs}
response = self.litellm_completion(**args)
if not generation_config.stream:
return LLMChatCompletion(**response.dict())
else:
return self._get_chat_completion(response)
def _get_chat_completion(
self,
response: Any,
) -> Generator[LLMChatCompletionChunk, None, None]:
for part in response:
yield LLMChatCompletionChunk(**part.dict())
def _get_base_args(
self,
generation_config: GenerationConfig,
prompt=None,
) -> dict:
"""Get the base arguments for the LiteLLM API."""
args = {
"model": generation_config.model,
"temperature": generation_config.temperature,
"top_p": generation_config.top_p,
"stream": generation_config.stream,
# TODO - We need to cap this to avoid potential errors when exceed max allowable context
"max_tokens": generation_config.max_tokens_to_sample,
}
return args
async def aget_completion(
self,
messages: list[dict],
generation_config: GenerationConfig,
**kwargs,
) -> LLMChatCompletion:
if generation_config.stream:
raise ValueError(
"Stream must be set to False to use the `aget_completion` method."
)
return await self._aget_completion(
messages, generation_config, **kwargs
)
async def _aget_completion(
self,
messages: list[dict],
generation_config: GenerationConfig,
**kwargs,
) -> Union[LLMChatCompletion, LLMChatCompletionChunk]:
"""Asynchronously get a completion from the OpenAI API based on the provided messages."""
# Create a dictionary with the default arguments
args = self._get_base_args(generation_config)
args["messages"] = messages
# Conditionally add the 'functions' argument if it's not None
if generation_config.functions is not None:
args["functions"] = generation_config.functions
args = {**args, **kwargs}
# Create the chat completion
return await self.litellm_acompletion(**args)
|
