File size: 26,790 Bytes
e3278e4 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 |
import asyncio
import os
from typing import TYPE_CHECKING, Any, Callable, Optional
import httpx
import openai
import litellm
from litellm import get_secret, get_secret_str
from litellm._logging import verbose_router_logger
from litellm.llms.azure.azure import get_azure_ad_token_from_oidc
from litellm.llms.azure.common_utils import (
get_azure_ad_token_from_entrata_id,
get_azure_ad_token_from_username_password,
)
from litellm.secret_managers.get_azure_ad_token_provider import (
get_azure_ad_token_provider,
)
from litellm.utils import calculate_max_parallel_requests
if TYPE_CHECKING:
from litellm.router import Router as _Router
LitellmRouter = _Router
else:
LitellmRouter = Any
class InitalizeOpenAISDKClient:
@staticmethod
def should_initialize_sync_client(
litellm_router_instance: LitellmRouter,
) -> bool:
"""
Returns if Sync OpenAI, Azure Clients should be initialized.
Do not init sync clients when router.router_general_settings.async_only_mode is True
"""
if litellm_router_instance is None:
return False
if litellm_router_instance.router_general_settings is not None:
if (
hasattr(litellm_router_instance, "router_general_settings")
and hasattr(
litellm_router_instance.router_general_settings, "async_only_mode"
)
and litellm_router_instance.router_general_settings.async_only_mode
is True
):
return False
return True
@staticmethod
def set_client( # noqa: PLR0915
litellm_router_instance: LitellmRouter, model: dict
):
"""
- Initializes Azure/OpenAI clients. Stores them in cache, b/c of this - https://github.com/BerriAI/litellm/issues/1278
- Initializes Semaphore for client w/ rpm. Stores them in cache. b/c of this - https://github.com/BerriAI/litellm/issues/2994
"""
client_ttl = litellm_router_instance.client_ttl
litellm_params = model.get("litellm_params", {})
model_name = litellm_params.get("model")
model_id = model["model_info"]["id"]
# ### IF RPM SET - initialize a semaphore ###
rpm = litellm_params.get("rpm", None)
tpm = litellm_params.get("tpm", None)
max_parallel_requests = litellm_params.get("max_parallel_requests", None)
calculated_max_parallel_requests = calculate_max_parallel_requests(
rpm=rpm,
max_parallel_requests=max_parallel_requests,
tpm=tpm,
default_max_parallel_requests=litellm_router_instance.default_max_parallel_requests,
)
if calculated_max_parallel_requests:
semaphore = asyncio.Semaphore(calculated_max_parallel_requests)
cache_key = f"{model_id}_max_parallel_requests_client"
litellm_router_instance.cache.set_cache(
key=cache_key,
value=semaphore,
local_only=True,
)
#### for OpenAI / Azure we need to initalize the Client for High Traffic ########
custom_llm_provider = litellm_params.get("custom_llm_provider")
custom_llm_provider = custom_llm_provider or model_name.split("/", 1)[0] or ""
default_api_base = None
default_api_key = None
if custom_llm_provider in litellm.openai_compatible_providers:
_, custom_llm_provider, api_key, api_base = litellm.get_llm_provider(
model=model_name
)
default_api_base = api_base
default_api_key = api_key
if (
model_name in litellm.open_ai_chat_completion_models
or custom_llm_provider in litellm.openai_compatible_providers
or custom_llm_provider == "azure"
or custom_llm_provider == "azure_text"
or custom_llm_provider == "custom_openai"
or custom_llm_provider == "openai"
or custom_llm_provider == "text-completion-openai"
or "ft:gpt-3.5-turbo" in model_name
or model_name in litellm.open_ai_embedding_models
):
is_azure_ai_studio_model: bool = False
if custom_llm_provider == "azure":
if litellm.utils._is_non_openai_azure_model(model_name):
is_azure_ai_studio_model = True
custom_llm_provider = "openai"
# remove azure prefx from model_name
model_name = model_name.replace("azure/", "")
# glorified / complicated reading of configs
# user can pass vars directly or they can pas os.environ/AZURE_API_KEY, in which case we will read the env
# we do this here because we init clients for Azure, OpenAI and we need to set the right key
api_key = litellm_params.get("api_key") or default_api_key
if (
api_key
and isinstance(api_key, str)
and api_key.startswith("os.environ/")
):
api_key_env_name = api_key.replace("os.environ/", "")
api_key = get_secret_str(api_key_env_name)
litellm_params["api_key"] = api_key
api_base = litellm_params.get("api_base")
base_url: Optional[str] = litellm_params.get("base_url")
api_base = (
api_base or base_url or default_api_base
) # allow users to pass in `api_base` or `base_url` for azure
if api_base and api_base.startswith("os.environ/"):
api_base_env_name = api_base.replace("os.environ/", "")
api_base = get_secret_str(api_base_env_name)
litellm_params["api_base"] = api_base
## AZURE AI STUDIO MISTRAL CHECK ##
"""
Make sure api base ends in /v1/
if not, add it - https://github.com/BerriAI/litellm/issues/2279
"""
if (
is_azure_ai_studio_model is True
and api_base is not None
and isinstance(api_base, str)
and not api_base.endswith("/v1/")
):
# check if it ends with a trailing slash
if api_base.endswith("/"):
api_base += "v1/"
elif api_base.endswith("/v1"):
api_base += "/"
else:
api_base += "/v1/"
api_version = litellm_params.get("api_version")
if api_version and api_version.startswith("os.environ/"):
api_version_env_name = api_version.replace("os.environ/", "")
api_version = get_secret_str(api_version_env_name)
litellm_params["api_version"] = api_version
timeout: Optional[float] = (
litellm_params.pop("timeout", None) or litellm.request_timeout
)
if isinstance(timeout, str) and timeout.startswith("os.environ/"):
timeout_env_name = timeout.replace("os.environ/", "")
timeout = get_secret(timeout_env_name) # type: ignore
litellm_params["timeout"] = timeout
stream_timeout: Optional[float] = litellm_params.pop(
"stream_timeout", timeout
) # if no stream_timeout is set, default to timeout
if isinstance(stream_timeout, str) and stream_timeout.startswith(
"os.environ/"
):
stream_timeout_env_name = stream_timeout.replace("os.environ/", "")
stream_timeout = get_secret(stream_timeout_env_name) # type: ignore
litellm_params["stream_timeout"] = stream_timeout
max_retries: Optional[int] = litellm_params.pop(
"max_retries", 0
) # router handles retry logic
if isinstance(max_retries, str) and max_retries.startswith("os.environ/"):
max_retries_env_name = max_retries.replace("os.environ/", "")
max_retries = get_secret(max_retries_env_name) # type: ignore
litellm_params["max_retries"] = max_retries
organization = litellm_params.get("organization", None)
if isinstance(organization, str) and organization.startswith("os.environ/"):
organization_env_name = organization.replace("os.environ/", "")
organization = get_secret_str(organization_env_name)
litellm_params["organization"] = organization
azure_ad_token_provider: Optional[Callable[[], str]] = None
if litellm_params.get("tenant_id"):
verbose_router_logger.debug(
"Using Azure AD Token Provider for Azure Auth"
)
azure_ad_token_provider = get_azure_ad_token_from_entrata_id(
tenant_id=litellm_params.get("tenant_id"),
client_id=litellm_params.get("client_id"),
client_secret=litellm_params.get("client_secret"),
)
if litellm_params.get("azure_username") and litellm_params.get(
"azure_password"
):
azure_ad_token_provider = get_azure_ad_token_from_username_password(
azure_username=litellm_params.get("azure_username"),
azure_password=litellm_params.get("azure_password"),
client_id=litellm_params.get("client_id"),
)
if custom_llm_provider == "azure" or custom_llm_provider == "azure_text":
if api_base is None or not isinstance(api_base, str):
filtered_litellm_params = {
k: v
for k, v in model["litellm_params"].items()
if k != "api_key"
}
_filtered_model = {
"model_name": model["model_name"],
"litellm_params": filtered_litellm_params,
}
raise ValueError(
f"api_base is required for Azure OpenAI. Set it on your config. Model - {_filtered_model}"
)
azure_ad_token = litellm_params.get("azure_ad_token")
if azure_ad_token is not None:
if azure_ad_token.startswith("oidc/"):
azure_ad_token = get_azure_ad_token_from_oidc(azure_ad_token)
elif (
azure_ad_token_provider is None
and litellm.enable_azure_ad_token_refresh is True
):
try:
azure_ad_token_provider = get_azure_ad_token_provider()
except ValueError:
verbose_router_logger.debug(
"Azure AD Token Provider could not be used."
)
if api_version is None:
api_version = os.getenv(
"AZURE_API_VERSION", litellm.AZURE_DEFAULT_API_VERSION
)
if "gateway.ai.cloudflare.com" in api_base:
if not api_base.endswith("/"):
api_base += "/"
azure_model = model_name.replace("azure/", "")
api_base += f"{azure_model}"
cache_key = f"{model_id}_async_client"
_client = openai.AsyncAzureOpenAI(
api_key=api_key,
azure_ad_token=azure_ad_token,
azure_ad_token_provider=azure_ad_token_provider,
base_url=api_base,
api_version=api_version,
timeout=timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.AsyncClient(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
if InitalizeOpenAISDKClient.should_initialize_sync_client(
litellm_router_instance=litellm_router_instance
):
cache_key = f"{model_id}_client"
_client = openai.AzureOpenAI( # type: ignore
api_key=api_key,
azure_ad_token=azure_ad_token,
azure_ad_token_provider=azure_ad_token_provider,
base_url=api_base,
api_version=api_version,
timeout=timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.Client(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
# streaming clients can have diff timeouts
cache_key = f"{model_id}_stream_async_client"
_client = openai.AsyncAzureOpenAI( # type: ignore
api_key=api_key,
azure_ad_token=azure_ad_token,
azure_ad_token_provider=azure_ad_token_provider,
base_url=api_base,
api_version=api_version,
timeout=stream_timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.AsyncClient(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
if InitalizeOpenAISDKClient.should_initialize_sync_client(
litellm_router_instance=litellm_router_instance
):
cache_key = f"{model_id}_stream_client"
_client = openai.AzureOpenAI( # type: ignore
api_key=api_key,
azure_ad_token=azure_ad_token,
azure_ad_token_provider=azure_ad_token_provider,
base_url=api_base,
api_version=api_version,
timeout=stream_timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.Client(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
else:
_api_key = api_key
if _api_key is not None and isinstance(_api_key, str):
# only show first 5 chars of api_key
_api_key = _api_key[:8] + "*" * 15
verbose_router_logger.debug(
f"Initializing Azure OpenAI Client for {model_name}, Api Base: {str(api_base)}, Api Key:{_api_key}"
)
azure_client_params = {
"api_key": api_key,
"azure_endpoint": api_base,
"api_version": api_version,
"azure_ad_token": azure_ad_token,
"azure_ad_token_provider": azure_ad_token_provider,
}
if azure_ad_token_provider is not None:
azure_client_params["azure_ad_token_provider"] = (
azure_ad_token_provider
)
from litellm.llms.azure.azure import (
select_azure_base_url_or_endpoint,
)
# this decides if we should set azure_endpoint or base_url on Azure OpenAI Client
# required to support GPT-4 vision enhancements, since base_url needs to be set on Azure OpenAI Client
azure_client_params = select_azure_base_url_or_endpoint(
azure_client_params
)
cache_key = f"{model_id}_async_client"
_client = openai.AsyncAzureOpenAI( # type: ignore
**azure_client_params,
timeout=timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.AsyncClient(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
if InitalizeOpenAISDKClient.should_initialize_sync_client(
litellm_router_instance=litellm_router_instance
):
cache_key = f"{model_id}_client"
_client = openai.AzureOpenAI( # type: ignore
**azure_client_params,
timeout=timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.Client(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
# streaming clients should have diff timeouts
cache_key = f"{model_id}_stream_async_client"
_client = openai.AsyncAzureOpenAI( # type: ignore
**azure_client_params,
timeout=stream_timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.AsyncClient(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
),
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
if InitalizeOpenAISDKClient.should_initialize_sync_client(
litellm_router_instance=litellm_router_instance
):
cache_key = f"{model_id}_stream_client"
_client = openai.AzureOpenAI( # type: ignore
**azure_client_params,
timeout=stream_timeout, # type: ignore
max_retries=max_retries, # type: ignore
http_client=httpx.Client(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
),
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
else:
_api_key = api_key # type: ignore
if _api_key is not None and isinstance(_api_key, str):
# only show first 5 chars of api_key
_api_key = _api_key[:8] + "*" * 15
verbose_router_logger.debug(
f"Initializing OpenAI Client for {model_name}, Api Base:{str(api_base)}, Api Key:{_api_key}"
)
cache_key = f"{model_id}_async_client"
_client = openai.AsyncOpenAI( # type: ignore
api_key=api_key,
base_url=api_base,
timeout=timeout, # type: ignore
max_retries=max_retries, # type: ignore
organization=organization,
http_client=httpx.AsyncClient(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
if InitalizeOpenAISDKClient.should_initialize_sync_client(
litellm_router_instance=litellm_router_instance
):
cache_key = f"{model_id}_client"
_client = openai.OpenAI( # type: ignore
api_key=api_key,
base_url=api_base,
timeout=timeout, # type: ignore
max_retries=max_retries, # type: ignore
organization=organization,
http_client=httpx.Client(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
# streaming clients should have diff timeouts
cache_key = f"{model_id}_stream_async_client"
_client = openai.AsyncOpenAI( # type: ignore
api_key=api_key,
base_url=api_base,
timeout=stream_timeout, # type: ignore
max_retries=max_retries, # type: ignore
organization=organization,
http_client=httpx.AsyncClient(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
if InitalizeOpenAISDKClient.should_initialize_sync_client(
litellm_router_instance=litellm_router_instance
):
# streaming clients should have diff timeouts
cache_key = f"{model_id}_stream_client"
_client = openai.OpenAI( # type: ignore
api_key=api_key,
base_url=api_base,
timeout=stream_timeout, # type: ignore
max_retries=max_retries, # type: ignore
organization=organization,
http_client=httpx.Client(
limits=httpx.Limits(
max_connections=1000, max_keepalive_connections=100
),
verify=litellm.ssl_verify,
), # type: ignore
)
litellm_router_instance.cache.set_cache(
key=cache_key,
value=_client,
ttl=client_ttl,
local_only=True,
) # cache for 1 hr
|