aboutsummaryrefslogtreecommitdiff
path: root/.venv/lib/python3.12/site-packages/core/base/providers/llm.py
blob: 669dfc4fa58208402a82cfdfc4fdf766da8e8f6a (about) (plain)
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
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
import asyncio
import logging
import random
import time
from abc import abstractmethod
from concurrent.futures import ThreadPoolExecutor
from typing import Any, AsyncGenerator, Generator, Optional

from litellm import AuthenticationError

from core.base.abstractions import (
    GenerationConfig,
    LLMChatCompletion,
    LLMChatCompletionChunk,
)

from .base import Provider, ProviderConfig

logger = logging.getLogger()


class CompletionConfig(ProviderConfig):
    provider: Optional[str] = None
    generation_config: Optional[GenerationConfig] = None
    concurrent_request_limit: int = 256
    max_retries: int = 3
    initial_backoff: float = 1.0
    max_backoff: float = 64.0

    def validate_config(self) -> None:
        if not self.provider:
            raise ValueError("Provider must be set.")
        if self.provider not in self.supported_providers:
            raise ValueError(f"Provider '{self.provider}' is not supported.")

    @property
    def supported_providers(self) -> list[str]:
        return ["anthropic", "litellm", "openai", "r2r"]


class CompletionProvider(Provider):
    def __init__(self, config: CompletionConfig) -> None:
        if not isinstance(config, CompletionConfig):
            raise ValueError(
                "CompletionProvider must be initialized with a `CompletionConfig`."
            )
        logger.info(f"Initializing CompletionProvider with config: {config}")
        super().__init__(config)
        self.config: CompletionConfig = config
        self.semaphore = asyncio.Semaphore(config.concurrent_request_limit)
        self.thread_pool = ThreadPoolExecutor(
            max_workers=config.concurrent_request_limit
        )

    async def _execute_with_backoff_async(self, task: dict[str, Any]):
        retries = 0
        backoff = self.config.initial_backoff
        while retries < self.config.max_retries:
            try:
                async with self.semaphore:
                    return await self._execute_task(task)
            except AuthenticationError:
                raise
            except Exception as e:
                logger.warning(
                    f"Request failed (attempt {retries + 1}): {str(e)}"
                )
                retries += 1
                if retries == self.config.max_retries:
                    raise
                await asyncio.sleep(random.uniform(0, backoff))
                backoff = min(backoff * 2, self.config.max_backoff)

    async def _execute_with_backoff_async_stream(
        self, task: dict[str, Any]
    ) -> AsyncGenerator[Any, None]:
        retries = 0
        backoff = self.config.initial_backoff
        while retries < self.config.max_retries:
            try:
                async with self.semaphore:
                    async for chunk in await self._execute_task(task):
                        yield chunk
                return  # Successful completion of the stream
            except AuthenticationError:
                raise
            except Exception as e:
                logger.warning(
                    f"Streaming request failed (attempt {retries + 1}): {str(e)}"
                )
                retries += 1
                if retries == self.config.max_retries:
                    raise
                await asyncio.sleep(random.uniform(0, backoff))
                backoff = min(backoff * 2, self.config.max_backoff)

    def _execute_with_backoff_sync(self, task: dict[str, Any]):
        retries = 0
        backoff = self.config.initial_backoff
        while retries < self.config.max_retries:
            try:
                return self._execute_task_sync(task)
            except Exception as e:
                logger.warning(
                    f"Request failed (attempt {retries + 1}): {str(e)}"
                )
                retries += 1
                if retries == self.config.max_retries:
                    raise
                time.sleep(random.uniform(0, backoff))
                backoff = min(backoff * 2, self.config.max_backoff)

    def _execute_with_backoff_sync_stream(
        self, task: dict[str, Any]
    ) -> Generator[Any, None, None]:
        retries = 0
        backoff = self.config.initial_backoff
        while retries < self.config.max_retries:
            try:
                yield from self._execute_task_sync(task)
                return  # Successful completion of the stream
            except Exception as e:
                logger.warning(
                    f"Streaming request failed (attempt {retries + 1}): {str(e)}"
                )
                retries += 1
                if retries == self.config.max_retries:
                    raise
                time.sleep(random.uniform(0, backoff))
                backoff = min(backoff * 2, self.config.max_backoff)

    @abstractmethod
    async def _execute_task(self, task: dict[str, Any]):
        pass

    @abstractmethod
    def _execute_task_sync(self, task: dict[str, Any]):
        pass

    async def aget_completion(
        self,
        messages: list[dict],
        generation_config: GenerationConfig,
        **kwargs,
    ) -> LLMChatCompletion:
        task = {
            "messages": messages,
            "generation_config": generation_config,
            "kwargs": kwargs,
        }
        response = await self._execute_with_backoff_async(task)
        return LLMChatCompletion(**response.dict())

    async def aget_completion_stream(
        self,
        messages: list[dict],
        generation_config: GenerationConfig,
        **kwargs,
    ) -> AsyncGenerator[LLMChatCompletionChunk, None]:
        generation_config.stream = True
        task = {
            "messages": messages,
            "generation_config": generation_config,
            "kwargs": kwargs,
        }
        async for chunk in self._execute_with_backoff_async_stream(task):
            if isinstance(chunk, dict):
                yield LLMChatCompletionChunk(**chunk)
                continue

            chunk.choices[0].finish_reason = (
                chunk.choices[0].finish_reason
                if chunk.choices[0].finish_reason != ""
                else None
            )  # handle error output conventions
            chunk.choices[0].finish_reason = (
                chunk.choices[0].finish_reason
                if chunk.choices[0].finish_reason != "eos"
                else "stop"
            )  # hardcode `eos` to `stop` for consistency
            try:
                yield LLMChatCompletionChunk(**(chunk.dict()))
            except Exception as e:
                logger.error(f"Error parsing chunk: {e}")
                yield LLMChatCompletionChunk(**(chunk.as_dict()))

    def get_completion_stream(
        self,
        messages: list[dict],
        generation_config: GenerationConfig,
        **kwargs,
    ) -> Generator[LLMChatCompletionChunk, None, None]:
        generation_config.stream = True
        task = {
            "messages": messages,
            "generation_config": generation_config,
            "kwargs": kwargs,
        }
        for chunk in self._execute_with_backoff_sync_stream(task):
            yield LLMChatCompletionChunk(**chunk.dict())