File size: 16,928 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 |
# What is this?
## Log success + failure events to Braintrust
import copy
import os
from datetime import datetime
from typing import Optional, Dict
import httpx
from pydantic import BaseModel
import litellm
from litellm import verbose_logger
from litellm.integrations.custom_logger import CustomLogger
from litellm.llms.custom_httpx.http_handler import (
HTTPHandler,
get_async_httpx_client,
httpxSpecialProvider,
)
from litellm.utils import print_verbose
global_braintrust_http_handler = get_async_httpx_client(llm_provider=httpxSpecialProvider.LoggingCallback)
global_braintrust_sync_http_handler = HTTPHandler()
API_BASE = "https://api.braintrustdata.com/v1"
def get_utc_datetime():
import datetime as dt
from datetime import datetime
if hasattr(dt, "UTC"):
return datetime.now(dt.UTC) # type: ignore
else:
return datetime.utcnow() # type: ignore
class BraintrustLogger(CustomLogger):
def __init__(self, api_key: Optional[str] = None, api_base: Optional[str] = None) -> None:
super().__init__()
self.validate_environment(api_key=api_key)
self.api_base = api_base or API_BASE
self.default_project_id = None
self.api_key: str = api_key or os.getenv("BRAINTRUST_API_KEY") # type: ignore
self.headers = {
"Authorization": "Bearer " + self.api_key,
"Content-Type": "application/json",
}
self._project_id_cache: Dict[str, str] = {} # Cache mapping project names to IDs
def validate_environment(self, api_key: Optional[str]):
"""
Expects
BRAINTRUST_API_KEY
in the environment
"""
missing_keys = []
if api_key is None and os.getenv("BRAINTRUST_API_KEY", None) is None:
missing_keys.append("BRAINTRUST_API_KEY")
if len(missing_keys) > 0:
raise Exception("Missing keys={} in environment.".format(missing_keys))
def get_project_id_sync(self, project_name: str) -> str:
"""
Get project ID from name, using cache if available.
If project doesn't exist, creates it.
"""
if project_name in self._project_id_cache:
return self._project_id_cache[project_name]
try:
response = global_braintrust_sync_http_handler.post(
f"{self.api_base}/project", headers=self.headers, json={"name": project_name}
)
project_dict = response.json()
project_id = project_dict["id"]
self._project_id_cache[project_name] = project_id
return project_id
except httpx.HTTPStatusError as e:
raise Exception(f"Failed to register project: {e.response.text}")
async def get_project_id_async(self, project_name: str) -> str:
"""
Async version of get_project_id_sync
"""
if project_name in self._project_id_cache:
return self._project_id_cache[project_name]
try:
response = await global_braintrust_http_handler.post(
f"{self.api_base}/project/register", headers=self.headers, json={"name": project_name}
)
project_dict = response.json()
project_id = project_dict["id"]
self._project_id_cache[project_name] = project_id
return project_id
except httpx.HTTPStatusError as e:
raise Exception(f"Failed to register project: {e.response.text}")
@staticmethod
def add_metadata_from_header(litellm_params: dict, metadata: dict) -> dict:
"""
Adds metadata from proxy request headers to Langfuse logging if keys start with "langfuse_"
and overwrites litellm_params.metadata if already included.
For example if you want to append your trace to an existing `trace_id` via header, send
`headers: { ..., langfuse_existing_trace_id: your-existing-trace-id }` via proxy request.
"""
if litellm_params is None:
return metadata
if litellm_params.get("proxy_server_request") is None:
return metadata
if metadata is None:
metadata = {}
proxy_headers = litellm_params.get("proxy_server_request", {}).get("headers", {}) or {}
for metadata_param_key in proxy_headers:
if metadata_param_key.startswith("braintrust"):
trace_param_key = metadata_param_key.replace("braintrust", "", 1)
if trace_param_key in metadata:
verbose_logger.warning(f"Overwriting Braintrust `{trace_param_key}` from request header")
else:
verbose_logger.debug(f"Found Braintrust `{trace_param_key}` in request header")
metadata[trace_param_key] = proxy_headers.get(metadata_param_key)
return metadata
async def create_default_project_and_experiment(self):
project = await global_braintrust_http_handler.post(
f"{self.api_base}/project", headers=self.headers, json={"name": "litellm"}
)
project_dict = project.json()
self.default_project_id = project_dict["id"]
def create_sync_default_project_and_experiment(self):
project = global_braintrust_sync_http_handler.post(
f"{self.api_base}/project", headers=self.headers, json={"name": "litellm"}
)
project_dict = project.json()
self.default_project_id = project_dict["id"]
def log_success_event( # noqa: PLR0915
self, kwargs, response_obj, start_time, end_time
):
verbose_logger.debug("REACHES BRAINTRUST SUCCESS")
try:
litellm_call_id = kwargs.get("litellm_call_id")
prompt = {"messages": kwargs.get("messages")}
output = None
choices = []
if response_obj is not None and (
kwargs.get("call_type", None) == "embedding" or isinstance(response_obj, litellm.EmbeddingResponse)
):
output = None
elif response_obj is not None and isinstance(response_obj, litellm.ModelResponse):
output = response_obj["choices"][0]["message"].json()
choices = response_obj["choices"]
elif response_obj is not None and isinstance(response_obj, litellm.TextCompletionResponse):
output = response_obj.choices[0].text
choices = response_obj.choices
elif response_obj is not None and isinstance(response_obj, litellm.ImageResponse):
output = response_obj["data"]
litellm_params = kwargs.get("litellm_params", {})
metadata = litellm_params.get("metadata", {}) or {} # if litellm_params['metadata'] == None
metadata = self.add_metadata_from_header(litellm_params, metadata)
clean_metadata = {}
try:
metadata = copy.deepcopy(metadata) # Avoid modifying the original metadata
except Exception:
new_metadata = {}
for key, value in metadata.items():
if (
isinstance(value, list)
or isinstance(value, dict)
or isinstance(value, str)
or isinstance(value, int)
or isinstance(value, float)
):
new_metadata[key] = copy.deepcopy(value)
metadata = new_metadata
# Get project_id from metadata or create default if needed
project_id = metadata.get("project_id")
if project_id is None:
project_name = metadata.get("project_name")
project_id = self.get_project_id_sync(project_name) if project_name else None
if project_id is None:
if self.default_project_id is None:
self.create_sync_default_project_and_experiment()
project_id = self.default_project_id
tags = []
if isinstance(metadata, dict):
for key, value in metadata.items():
# generate langfuse tags - Default Tags sent to Langfuse from LiteLLM Proxy
if (
litellm.langfuse_default_tags is not None
and isinstance(litellm.langfuse_default_tags, list)
and key in litellm.langfuse_default_tags
):
tags.append(f"{key}:{value}")
# clean litellm metadata before logging
if key in [
"headers",
"endpoint",
"caching_groups",
"previous_models",
]:
continue
else:
clean_metadata[key] = value
cost = kwargs.get("response_cost", None)
if cost is not None:
clean_metadata["litellm_response_cost"] = cost
metrics: Optional[dict] = None
usage_obj = getattr(response_obj, "usage", None)
if usage_obj and isinstance(usage_obj, litellm.Usage):
litellm.utils.get_logging_id(start_time, response_obj)
metrics = {
"prompt_tokens": usage_obj.prompt_tokens,
"completion_tokens": usage_obj.completion_tokens,
"total_tokens": usage_obj.total_tokens,
"total_cost": cost,
"time_to_first_token": end_time.timestamp() - start_time.timestamp(),
"start": start_time.timestamp(),
"end": end_time.timestamp(),
}
request_data = {
"id": litellm_call_id,
"input": prompt["messages"],
"metadata": clean_metadata,
"tags": tags,
"span_attributes": {"name": "Chat Completion", "type": "llm"},
}
if choices is not None:
request_data["output"] = [choice.dict() for choice in choices]
else:
request_data["output"] = output
if metrics is not None:
request_data["metrics"] = metrics
try:
print_verbose(f"global_braintrust_sync_http_handler.post: {global_braintrust_sync_http_handler.post}")
global_braintrust_sync_http_handler.post(
url=f"{self.api_base}/project_logs/{project_id}/insert",
json={"events": [request_data]},
headers=self.headers,
)
except httpx.HTTPStatusError as e:
raise Exception(e.response.text)
except Exception as e:
raise e # don't use verbose_logger.exception, if exception is raised
async def async_log_success_event( # noqa: PLR0915
self, kwargs, response_obj, start_time, end_time
):
verbose_logger.debug("REACHES BRAINTRUST SUCCESS")
try:
litellm_call_id = kwargs.get("litellm_call_id")
prompt = {"messages": kwargs.get("messages")}
output = None
choices = []
if response_obj is not None and (
kwargs.get("call_type", None) == "embedding" or isinstance(response_obj, litellm.EmbeddingResponse)
):
output = None
elif response_obj is not None and isinstance(response_obj, litellm.ModelResponse):
output = response_obj["choices"][0]["message"].json()
choices = response_obj["choices"]
elif response_obj is not None and isinstance(response_obj, litellm.TextCompletionResponse):
output = response_obj.choices[0].text
choices = response_obj.choices
elif response_obj is not None and isinstance(response_obj, litellm.ImageResponse):
output = response_obj["data"]
litellm_params = kwargs.get("litellm_params", {})
metadata = litellm_params.get("metadata", {}) or {} # if litellm_params['metadata'] == None
metadata = self.add_metadata_from_header(litellm_params, metadata)
clean_metadata = {}
new_metadata = {}
for key, value in metadata.items():
if (
isinstance(value, list)
or isinstance(value, str)
or isinstance(value, int)
or isinstance(value, float)
):
new_metadata[key] = value
elif isinstance(value, BaseModel):
new_metadata[key] = value.model_dump_json()
elif isinstance(value, dict):
for k, v in value.items():
if isinstance(v, datetime):
value[k] = v.isoformat()
new_metadata[key] = value
# Get project_id from metadata or create default if needed
project_id = metadata.get("project_id")
if project_id is None:
project_name = metadata.get("project_name")
project_id = await self.get_project_id_async(project_name) if project_name else None
if project_id is None:
if self.default_project_id is None:
await self.create_default_project_and_experiment()
project_id = self.default_project_id
tags = []
if isinstance(metadata, dict):
for key, value in metadata.items():
# generate langfuse tags - Default Tags sent to Langfuse from LiteLLM Proxy
if (
litellm.langfuse_default_tags is not None
and isinstance(litellm.langfuse_default_tags, list)
and key in litellm.langfuse_default_tags
):
tags.append(f"{key}:{value}")
# clean litellm metadata before logging
if key in [
"headers",
"endpoint",
"caching_groups",
"previous_models",
]:
continue
else:
clean_metadata[key] = value
cost = kwargs.get("response_cost", None)
if cost is not None:
clean_metadata["litellm_response_cost"] = cost
metrics: Optional[dict] = None
usage_obj = getattr(response_obj, "usage", None)
if usage_obj and isinstance(usage_obj, litellm.Usage):
litellm.utils.get_logging_id(start_time, response_obj)
metrics = {
"prompt_tokens": usage_obj.prompt_tokens,
"completion_tokens": usage_obj.completion_tokens,
"total_tokens": usage_obj.total_tokens,
"total_cost": cost,
"start": start_time.timestamp(),
"end": end_time.timestamp(),
}
api_call_start_time = kwargs.get("api_call_start_time")
completion_start_time = kwargs.get("completion_start_time")
if api_call_start_time is not None and completion_start_time is not None:
metrics["time_to_first_token"] = completion_start_time.timestamp() - api_call_start_time.timestamp()
request_data = {
"id": litellm_call_id,
"input": prompt["messages"],
"output": output,
"metadata": clean_metadata,
"tags": tags,
"span_attributes": {"name": "Chat Completion", "type": "llm"},
}
if choices is not None:
request_data["output"] = [choice.dict() for choice in choices]
else:
request_data["output"] = output
if metrics is not None:
request_data["metrics"] = metrics
if metrics is not None:
request_data["metrics"] = metrics
try:
await global_braintrust_http_handler.post(
url=f"{self.api_base}/project_logs/{project_id}/insert",
json={"events": [request_data]},
headers=self.headers,
)
except httpx.HTTPStatusError as e:
raise Exception(e.response.text)
except Exception as e:
raise e # don't use verbose_logger.exception, if exception is raised
def log_failure_event(self, kwargs, response_obj, start_time, end_time):
return super().log_failure_event(kwargs, response_obj, start_time, end_time)
|