File size: 14,659 Bytes
77320e4 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 |
import json
import logging
import os
from abc import ABC
from typing import Callable, List
import openai
from tenacity import ( # for exponential backoff
before_sleep_log,
retry,
stop_after_attempt,
wait_random_exponential,
)
from ..base_llm import BaseLLM
from ...schemas import *
logger = logging.getLogger(__name__)
MAX_PROMPT_LENGTH = 7000
@retry(wait=wait_random_exponential(min=1, max=10), stop=stop_after_attempt(10), reraise=True,
before_sleep=before_sleep_log(logger, logging.WARNING))
def chatcompletion_with_backoff(**kwargs):
return openai.ChatCompletion.create(**kwargs)
@retry(wait=wait_random_exponential(min=1, max=10), stop=stop_after_attempt(10), reraise=True,
before_sleep=before_sleep_log(logger, logging.WARNING))
async def async_chatcompletion_with_backoff(**kwargs):
async def _internal_coroutine():
return await openai.ChatCompletion.acreate(**kwargs)
return await _internal_coroutine()
class OptOpenAIClient(BaseLLM, ABC):
"""
Wrapper class for OpenAI GPT API collections.
:param model_name: The name of the model to use.
:type model_name: str
:param params: The parameters for the model.
:type params: OptParamModel
"""
model_name: str
params: OptParamModel = OptParamModel()
def __init__(self, **data):
super().__init__(**data)
openai.api_key = "EMPTY"
openai.api_base = "http://localhost:8000/v1"
@classmethod
async def create(cls, config_data):
return OptOpenAIClient(**config_data)
def get_model_name(self) -> str:
return self.model_name
def get_model_param(self) -> OptParamModel:
return self.params
def completion(self, prompt: str, **kwargs) -> BaseCompletion:
"""
Completion method for OpenAI GPT API.
:param prompt: The prompt to use for completion.
:type prompt: str
:param kwargs: Additional keyword arguments.
:type kwargs: dict
:return: BaseCompletion object.
:rtype: BaseCompletion
"""
response = chatcompletion_with_backoff(
model=self.model_name,
# engine=self.get_model_name(), # GPT-4
messages=[
{"role": "user", "content": prompt[-MAX_PROMPT_LENGTH:]}
],
timeout=1000,
**kwargs
)
return BaseCompletion(state="success",
content=response.choices[0].message["content"],
prompt_token=response.get("usage", {}).get("prompt_tokens", 0),
completion_token=response.get("usage", {}).get("completion_tokens", 0))
async def async_completion(self, prompt: str, **kwargs) -> BaseCompletion:
"""
Completion method for OpenAI GPT API.
:param prompt: The prompt to use for completion.
:type prompt: str
:param kwargs: Additional keyword arguments.
:type kwargs: dict
:return: BaseCompletion object.
:rtype: BaseCompletion
"""
response = await async_chatcompletion_with_backoff(
# engine=self.get_model_name(), # GPT-4
model=self.model_name,
messages=[
{"role": "user", "content": prompt[-MAX_PROMPT_LENGTH:]}
],
timeout=1000,
**kwargs
)
return BaseCompletion(state="success",
content=response.choices[0].message["content"],
prompt_token=response.get("usage", {}).get("prompt_tokens", 0),
completion_token=response.get("usage", {}).get("completion_tokens", 0))
def chat_completion(self, message: List[dict]) -> ChatCompletion:
"""
Chat completion method for OpenAI GPT API.
:param message: The message to use for completion.
:type message: List[dict]
:return: ChatCompletion object.
:rtype: ChatCompletion
"""
try:
# response = openai.ChatCompletion.create(
# engine=self.get_model_name(), # GPT-4
# messages=message,
# timeout=1000,
# )
response = openai.ChatCompletion.create(
n=self.params.n,
model=self.model_name,
messages=message,
temperature=self.params.temperature,
max_tokens=self.params.max_tokens,
top_p=self.params.top_p,
frequency_penalty=self.params.frequency_penalty,
presence_penalty=self.params.presence_penalty,
)
return ChatCompletion(
state="success",
role=response.choices[0].message["role"],
content=response.choices[0].message["content"],
prompt_token=response.get("usage", {}).get("prompt_tokens", 0),
completion_token=response.get("usage", {}).get("completion_tokens", 0),
)
except Exception as exception:
print("Exception:", exception)
return ChatCompletion(state="error", content=exception)
def stream_chat_completion(self, message: List[dict], **kwargs):
"""
Stream output chat completion for OpenAI GPT API.
:param message: The message (scratchpad) to use for completion. Usually contains json of role and content.
:type message: List[dict]
:param kwargs: Additional keyword arguments.
:type kwargs: dict
:return: ChatCompletion object.
:rtype: ChatCompletion
"""
try:
# response = openai.ChatCompletion.create(
# engine=self.get_model_name(), # GPT-4
# messages=message,
# timeout=1000,
# **kwargs,
# )
response = openai.ChatCompletion.create(
n=self.params.n,
model=self.model_name,
messages=message,
temperature=self.params.temperature,
max_tokens=self.params.max_tokens,
top_p=self.params.top_p,
frequency_penalty=self.params.frequency_penalty,
presence_penalty=self.params.presence_penalty,
stream=True,
**kwargs
)
role = next(response).choices[0].delta["role"]
messages = []
## TODO: Calculate prompt_token and for stream mode
for resp in response:
messages.append(resp.choices[0].delta.get("content", ""))
yield ChatCompletion(
state="success",
role=role,
content=messages[-1],
prompt_token=0,
completion_token=0,
)
except Exception as exception:
print("Exception:", exception)
return ChatCompletion(state="error", content=exception)
def function_chat_completion(
self,
message: List[dict],
function_map: Dict[str, Callable],
function_schema: List[Dict],
) -> ChatCompletionWithHistory:
"""
Chat completion method for OpenAI GPT API.
:param message: The message to use for completion.
:type message: List[dict]
:param function_map: The function map to use for completion.
:type function_map: Dict[str, Callable]
:param function_schema: The function schema to use for completion.
:type function_schema: List[Dict]
:return: ChatCompletionWithHistory object.
:rtype: ChatCompletionWithHistory
"""
assert len(function_schema) == len(function_map)
try:
# response = openai.ChatCompletion.create(
# engine=self.get_model_name(), # GPT-4
# messages=message,
# functions=function_schema,
# timeout=1000,
# )
response = openai.ChatCompletion.create(
n=self.params.n,
model=self.model_name,
messages=message,
functions=function_schema,
temperature=self.params.temperature,
max_tokens=self.params.max_tokens,
top_p=self.params.top_p,
frequency_penalty=self.params.frequency_penalty,
presence_penalty=self.params.presence_penalty,
)
response_message = response.choices[0]["message"]
if response_message.get("function_call"):
function_name = response_message["function_call"]["name"]
fuction_to_call = function_map[function_name]
function_args = json.loads(
response_message["function_call"]["arguments"]
)
function_response = fuction_to_call(**function_args)
# Postprocess function response
if isinstance(function_response, str):
plugin_cost = 0
plugin_token = 0
elif isinstance(function_response, AgentOutput):
plugin_cost = function_response.cost
plugin_token = function_response.token_usage
function_response = function_response.output
else:
raise Exception(
"Invalid tool response type. Must be on of [AgentOutput, str]"
)
message.append(dict(response_message))
message.append(
{
"role": "function",
"name": function_name,
"content": function_response,
}
)
second_response = openai.ChatCompletion.create(
model=self.get_model_name(),
messages=message,
)
message.append(dict(second_response.choices[0].message))
return ChatCompletionWithHistory(
state="success",
role=second_response.choices[0].message["role"],
content=second_response.choices[0].message["content"],
prompt_token=response.get("usage", {}).get("prompt_tokens", 0)
+ second_response.get("usage", {}).get("prompt_tokens", 0),
completion_token=response.get("usage", {}).get(
"completion_tokens", 0
)
+ second_response.get("usage", {}).get("completion_tokens", 0),
message_scratchpad=message,
plugin_cost=plugin_cost,
plugin_token=plugin_token,
)
else:
message.append(dict(response_message))
return ChatCompletionWithHistory(
state="success",
role=response.choices[0].message["role"],
content=response.choices[0].message["content"],
prompt_token=response.get("usage", {}).get("prompt_tokens", 0),
completion_token=response.get("usage", {}).get(
"completion_tokens", 0
),
message_scratchpad=message,
)
except Exception as exception:
print("Exception:", exception)
return ChatCompletionWithHistory(state="error", content=str(exception))
def function_chat_stream_completion(
self,
message: List[dict],
function_map: Dict[str, Callable],
function_schema: List[Dict],
) -> ChatCompletionWithHistory:
assert len(function_schema) == len(function_map)
try:
response = openai.ChatCompletion.create(
n=self.params.n,
model=self.get_model_name(),
messages=message,
functions=function_schema,
temperature=self.params.temperature,
max_tokens=self.params.max_tokens,
top_p=self.params.top_p,
frequency_penalty=self.params.frequency_penalty,
presence_penalty=self.params.presence_penalty,
stream=True,
)
tmp = next(response)
role = tmp.choices[0].delta["role"]
_type = (
"function_call"
if tmp.choices[0].delta["content"] is None
else "content"
)
if _type == "function_call":
name = tmp.choices[0].delta["function_call"]["name"]
yield _type, ChatCompletionWithHistory(
state="success",
role=role,
content="{" + f'"name":"{name}", "arguments":',
message_scratchpad=message,
)
for resp in response:
# print(resp)
content = resp.choices[0].delta.get(_type, "")
if isinstance(content, dict):
content = content["arguments"]
yield _type, ChatCompletionWithHistory(
state="success",
role=role,
content=content,
message_scratchpad=message,
)
# result = ''.join(messages)
# if _type == "function_call":
# result = json.loads(result)
# function_name = result["name"]
# fuction_to_call = function_map[function_name]
# function_args = result["arguments"]
# function_response = fuction_to_call(**function_args)
#
# # Postprocess function response
# if isinstance(function_response, AgentOutput):
# function_response = function_response.output
# message.append({"role": "function",
# "name": function_name,
# "content": function_response})
# second_response = self.function_chat_stream_completion(message=message,function_map=function_map,function_schema=function_schema)
# message.append(dict(second_response.choices[0].message))
except Exception as e:
logger.error(f"Failed to get response {str(e)}", exc_info=True)
raise e
|