Spaces:
Build error
Build error
File size: 13,955 Bytes
01523b5 |
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 |
import logging
import json
import ast
import os
import numpy as np
from aiohttp import ClientSession
from typing import Dict, List, Optional, Union
from tenacity import retry, stop_after_attempt, wait_exponential
from pydantic import BaseModel, Field
from agentverse.llms.base import LLMResult
from agentverse.logging import logger
from agentverse.message import Message
from . import llm_registry
from .base import BaseChatModel, BaseCompletionModel, BaseModelArgs
from .utils.jsonrepair import JsonRepair
try:
import openai
from openai.error import OpenAIError
except ImportError:
is_openai_available = False
logging.warning("openai package is not installed")
else:
# openai.proxy = os.environ.get("http_proxy")
# if openai.proxy is None:
# openai.proxy = os.environ.get("HTTP_PROXY")
if os.environ.get("OPENAI_API_KEY") != None:
openai.api_key = os.environ.get("OPENAI_API_KEY")
is_openai_available = True
elif os.environ.get("AZURE_OPENAI_API_KEY") != None:
openai.api_type = "azure"
openai.api_key = os.environ.get("AZURE_OPENAI_API_KEY")
openai.api_base = os.environ.get("AZURE_OPENAI_API_BASE")
openai.api_version = "2023-05-15"
is_openai_available = True
else:
logging.warning(
"OpenAI API key is not set. Please set the environment variable OPENAI_API_KEY"
)
is_openai_available = False
class OpenAIChatArgs(BaseModelArgs):
model: str = Field(default="gpt-3.5-turbo")
deployment_id: str = Field(default=None)
max_tokens: int = Field(default=2048)
temperature: float = Field(default=1.0)
top_p: int = Field(default=1)
n: int = Field(default=1)
stop: Optional[Union[str, List]] = Field(default=None)
presence_penalty: int = Field(default=0)
frequency_penalty: int = Field(default=0)
# class OpenAICompletionArgs(OpenAIChatArgs):
# model: str = Field(default="text-davinci-003")
# suffix: str = Field(default="")
# best_of: int = Field(default=1)
# @llm_registry.register("text-davinci-003")
# class OpenAICompletion(BaseCompletionModel):
# args: OpenAICompletionArgs = Field(default_factory=OpenAICompletionArgs)
# def __init__(self, max_retry: int = 3, **kwargs):
# args = OpenAICompletionArgs()
# args = args.dict()
# for k, v in args.items():
# args[k] = kwargs.pop(k, v)
# if len(kwargs) > 0:
# logging.warning(f"Unused arguments: {kwargs}")
# super().__init__(args=args, max_retry=max_retry)
# def generate_response(self, prompt: str) -> LLMResult:
# response = openai.Completion.create(prompt=prompt, **self.args.dict())
# return LLMResult(
# content=response["choices"][0]["text"],
# send_tokens=response["usage"]["prompt_tokens"],
# recv_tokens=response["usage"]["completion_tokens"],
# total_tokens=response["usage"]["total_tokens"],
# )
# async def agenerate_response(self, prompt: str) -> LLMResult:
# response = await openai.Completion.acreate(prompt=prompt, **self.args.dict())
# return LLMResult(
# content=response["choices"][0]["text"],
# send_tokens=response["usage"]["prompt_tokens"],
# recv_tokens=response["usage"]["completion_tokens"],
# total_tokens=response["usage"]["total_tokens"],
# )
@llm_registry.register("gpt-35-turbo")
@llm_registry.register("gpt-3.5-turbo")
@llm_registry.register("gpt-4")
class OpenAIChat(BaseChatModel):
args: OpenAIChatArgs = Field(default_factory=OpenAIChatArgs)
total_prompt_tokens: int = 0
total_completion_tokens: int = 0
def __init__(self, max_retry: int = 3, **kwargs):
args = OpenAIChatArgs()
args = args.dict()
for k, v in args.items():
args[k] = kwargs.pop(k, v)
if len(kwargs) > 0:
logging.warning(f"Unused arguments: {kwargs}")
super().__init__(args=args, max_retry=max_retry)
# def _construct_messages(self, history: List[Message]):
# return history + [{"role": "user", "content": query}]
@retry(
stop=stop_after_attempt(20),
wait=wait_exponential(multiplier=1, min=4, max=10),
reraise=True,
)
def generate_response(
self,
prepend_prompt: str = "",
history: List[dict] = [],
append_prompt: str = "",
functions: List[dict] = [],
) -> LLMResult:
messages = self.construct_messages(prepend_prompt, history, append_prompt)
logger.log_prompt(messages)
try:
# Execute function call
if functions != []:
response = openai.ChatCompletion.create(
messages=messages,
functions=functions,
**self.args.dict(),
)
if response["choices"][0]["message"].get("function_call") is not None:
self.collect_metrics(response)
return LLMResult(
content=response["choices"][0]["message"].get("content", ""),
function_name=response["choices"][0]["message"][
"function_call"
]["name"],
function_arguments=ast.literal_eval(
response["choices"][0]["message"]["function_call"][
"arguments"
]
),
send_tokens=response["usage"]["prompt_tokens"],
recv_tokens=response["usage"]["completion_tokens"],
total_tokens=response["usage"]["total_tokens"],
)
else:
self.collect_metrics(response)
return LLMResult(
content=response["choices"][0]["message"]["content"],
send_tokens=response["usage"]["prompt_tokens"],
recv_tokens=response["usage"]["completion_tokens"],
total_tokens=response["usage"]["total_tokens"],
)
else:
response = openai.ChatCompletion.create(
messages=messages,
**self.args.dict(),
)
self.collect_metrics(response)
return LLMResult(
content=response["choices"][0]["message"]["content"],
send_tokens=response["usage"]["prompt_tokens"],
recv_tokens=response["usage"]["completion_tokens"],
total_tokens=response["usage"]["total_tokens"],
)
except (OpenAIError, KeyboardInterrupt, json.decoder.JSONDecodeError) as error:
raise
@retry(
stop=stop_after_attempt(20),
wait=wait_exponential(multiplier=1, min=4, max=10),
reraise=True,
)
async def agenerate_response(
self,
prepend_prompt: str = "",
history: List[dict] = [],
append_prompt: str = "",
functions: List[dict] = [],
) -> LLMResult:
messages = self.construct_messages(prepend_prompt, history, append_prompt)
logger.log_prompt(messages)
try:
if functions != []:
async with ClientSession(trust_env=True) as session:
openai.aiosession.set(session)
response = await openai.ChatCompletion.acreate(
messages=messages,
functions=functions,
**self.args.dict(),
)
if response["choices"][0]["message"].get("function_call") is not None:
function_name = response["choices"][0]["message"]["function_call"][
"name"
]
valid_function = False
if function_name.startswith("function."):
function_name = function_name.replace("function.", "")
elif function_name.startswith("functions."):
function_name = function_name.replace("functions.", "")
for function in functions:
if function["name"] == function_name:
valid_function = True
break
if not valid_function:
logger.warn(
f"The returned function name {function_name} is not in the list of valid functions. Retrying..."
)
raise ValueError(
f"The returned function name {function_name} is not in the list of valid functions."
)
try:
arguments = ast.literal_eval(
response["choices"][0]["message"]["function_call"][
"arguments"
]
)
except:
try:
arguments = ast.literal_eval(
JsonRepair(
response["choices"][0]["message"]["function_call"][
"arguments"
]
).repair()
)
except:
logger.warn(
"The returned argument in function call is not valid json. Retrying..."
)
raise ValueError(
"The returned argument in function call is not valid json."
)
self.collect_metrics(response)
return LLMResult(
function_name=function_name,
function_arguments=arguments,
send_tokens=response["usage"]["prompt_tokens"],
recv_tokens=response["usage"]["completion_tokens"],
total_tokens=response["usage"]["total_tokens"],
)
else:
self.collect_metrics(response)
return LLMResult(
content=response["choices"][0]["message"]["content"],
send_tokens=response["usage"]["prompt_tokens"],
recv_tokens=response["usage"]["completion_tokens"],
total_tokens=response["usage"]["total_tokens"],
)
else:
async with ClientSession(trust_env=True) as session:
openai.aiosession.set(session)
response = await openai.ChatCompletion.acreate(
messages=messages,
**self.args.dict(),
)
self.collect_metrics(response)
return LLMResult(
content=response["choices"][0]["message"]["content"],
send_tokens=response["usage"]["prompt_tokens"],
recv_tokens=response["usage"]["completion_tokens"],
total_tokens=response["usage"]["total_tokens"],
)
except (OpenAIError, KeyboardInterrupt, json.decoder.JSONDecodeError) as error:
raise
def construct_messages(
self, prepend_prompt: str, history: List[dict], append_prompt: str
):
messages = []
if prepend_prompt != "":
messages.append({"role": "system", "content": prepend_prompt})
if len(history) > 0:
messages += history
if append_prompt != "":
messages.append({"role": "user", "content": append_prompt})
return messages
def collect_metrics(self, response):
self.total_prompt_tokens += response["usage"]["prompt_tokens"]
self.total_completion_tokens += response["usage"]["completion_tokens"]
def get_spend(self) -> int:
input_cost_map = {
"gpt-3.5-turbo": 0.0015,
"gpt-3.5-turbo-16k": 0.003,
"gpt-3.5-turbo-0613": 0.0015,
"gpt-3.5-turbo-16k-0613": 0.003,
"gpt-4": 0.03,
"gpt-4-0613": 0.03,
"gpt-4-32k": 0.06,
}
output_cost_map = {
"gpt-3.5-turbo": 0.002,
"gpt-3.5-turbo-16k": 0.004,
"gpt-3.5-turbo-0613": 0.002,
"gpt-3.5-turbo-16k-0613": 0.004,
"gpt-4": 0.06,
"gpt-4-0613": 0.06,
"gpt-4-32k": 0.12,
}
model = self.args.model
if model not in input_cost_map or model not in output_cost_map:
raise ValueError(f"Model type {model} not supported")
return (
self.total_prompt_tokens * input_cost_map[model] / 1000.0
+ self.total_completion_tokens * output_cost_map[model] / 1000.0
)
@retry(
stop=stop_after_attempt(3),
wait=wait_exponential(multiplier=1, min=4, max=10),
reraise=True,
)
def get_embedding(text: str, attempts=3) -> np.array:
try:
text = text.replace("\n", " ")
if openai.api_type == "azure":
embedding = openai.Embedding.create(
input=[text], deployment_id="text-embedding-ada-002"
)["data"][0]["embedding"]
else:
embedding = openai.Embedding.create(
input=[text], model="text-embedding-ada-002"
)["data"][0]["embedding"]
return tuple(embedding)
except Exception as e:
attempt += 1
logger.error(f"Error {e} when requesting openai models. Retrying")
raise
|