phoenix/litellm
1
0
Fork 0
forked from phoenix/litellm-mirror
Code Pull requests Activity

Merge branch 'main' into litellm_aporio_integration

Browse source
This commit is contained in:
Krish Dholakia 2024-07-17 22:14:29 -07:00 • committed by GitHub
commit 77656d9f11
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
19 changed files with 512 additions and 142 deletions
Download patch file Download diff file Expand all files Collapse all files

10
docs/my-website/docs/observability/helicone_integration.md
View file

@ -72,7 +72,7 @@ Helicone's proxy provides [advanced functionality](https://docs.helicone.ai/gett
To use Helicone as a proxy for your LLM requests:
1. Set Helicone as your base URL via: litellm.api_base
2. Pass in Helicone request headers via: litellm.headers
2. Pass in Helicone request headers via: litellm.metadata
Complete Code:
@ -99,7 +99,7 @@ print(response)
You can add custom metadata and properties to your requests using Helicone headers. Here are some examples:
```python
litellm.headers = {
litellm.metadata = {
"Helicone-Auth": f"Bearer {os.getenv('HELICONE_API_KEY')}", # Authenticate to send requests to Helicone API
"Helicone-User-Id": "user-abc", # Specify the user making the request
"Helicone-Property-App": "web", # Custom property to add additional information
@ -127,7 +127,7 @@ litellm.headers = {
Enable caching and set up rate limiting policies:
```python
litellm.headers = {
litellm.metadata = {
"Helicone-Auth": f"Bearer {os.getenv('HELICONE_API_KEY')}", # Authenticate to send requests to Helicone API
"Helicone-Cache-Enabled": "true", # Enable caching of responses
"Cache-Control": "max-age=3600", # Set cache limit to 1 hour
@ -140,7 +140,7 @@ litellm.headers = {
Track multi-step and agentic LLM interactions using session IDs and paths:
```python
litellm.headers = {
litellm.metadata = {
"Helicone-Auth": f"Bearer {os.getenv('HELICONE_API_KEY')}", # Authenticate to send requests to Helicone API
"Helicone-Session-Id": "session-abc-123", # The session ID you want to track
"Helicone-Session-Path": "parent-trace/child-trace", # The path of the session
@ -157,7 +157,7 @@ By using these two headers, you can effectively group and visualize multi-step L
Set up retry mechanisms and fallback options:
```python
litellm.headers = {
litellm.metadata = {
"Helicone-Auth": f"Bearer {os.getenv('HELICONE_API_KEY')}", # Authenticate to send requests to Helicone API
"Helicone-Retry-Enabled": "true", # Enable retry mechanism
"helicone-retry-num": "3", # Set number of retries

2
docs/my-website/docs/observability/langsmith_integration.md
View file

@ -14,7 +14,7 @@ https://github.com/BerriAI/litellm
An all-in-one developer platform for every step of the application lifecycle
https://smith.langchain.com/
<Image img={require('../../img/langsmith.png')} />
<Image img={require('../../img/langsmith_new.png')} />
:::info
We want to learn how we can make the callbacks better! Meet the LiteLLM [founders](https://calendly.com/d/4mp-gd3-k5k/berriai-1-1-onboarding-litellm-hosted-version) or

45
docs/my-website/docs/proxy/logging.md
View file

@ -5,6 +5,7 @@ Log Proxy input, output, and exceptions using:
- Langfuse
- OpenTelemetry
- Custom Callbacks
- Langsmith
- DataDog
- DynamoDB
- s3 Bucket
@ -1086,6 +1087,50 @@ litellm_settings:
Start the LiteLLM Proxy and make a test request to verify the logs reached your callback API
## Logging LLM IO to Langsmith
1. Set `success_callback: ["langsmith"]` on litellm config.yaml
If you're using a custom LangSmith instance, you can set the
`LANGSMITH_BASE_URL` environment variable to point to your instance.
```yaml
litellm_settings:
success_callback: ["langsmith"]
environment_variables:
LANGSMITH_API_KEY: "lsv2_pt_xxxxxxxx"
LANGSMITH_PROJECT: "litellm-proxy"
LANGSMITH_BASE_URL: "https://api.smith.langchain.com" # (Optional - only needed if you have a custom Langsmith instance)
```
2. Start Proxy
```
litellm --config /path/to/config.yaml
```
3. Test it!
```bash
curl --location 'http://0.0.0.0:4000/chat/completions' \
--header 'Content-Type: application/json' \
--data ' {
"model": "fake-openai-endpoint",
"messages": [
{
"role": "user",
"content": "Hello, Claude gm!"
}
],
}
'
```
Expect to see your log on Langfuse
<Image img={require('../../img/langsmith_new.png')} />
## Logging LLM IO to Galileo
[BETA]

BIN
docs/my-website/img/langsmith_new.png Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 353 KiB

2
litellm/__init__.py
View file

@ -38,7 +38,7 @@ success_callback: List[Union[str, Callable]] = []
failure_callback: List[Union[str, Callable]] = []
service_callback: List[Union[str, Callable]] = []
_custom_logger_compatible_callbacks_literal = Literal[
"lago", "openmeter", "logfire", "dynamic_rate_limiter"
"lago", "openmeter", "logfire", "dynamic_rate_limiter", "langsmith", "galileo"
]
callbacks: List[Union[Callable, _custom_logger_compatible_callbacks_literal]] = []
_langfuse_default_tags: Optional[

211
litellm/integrations/langsmith.py
View file

@ -1,13 +1,43 @@
#### What this does ####
# On success, logs events to Langsmith
import dotenv, os # type: ignore
import requests # type: ignore
from datetime import datetime
import traceback
import asyncio
import os
import traceback
import types
from datetime import datetime
from typing import Any, List, Optional, Union
import dotenv # type: ignore
import httpx
import requests # type: ignore
from pydantic import BaseModel # type: ignore
import litellm
from litellm._logging import verbose_logger
from litellm.integrations.custom_logger import CustomLogger
from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler
class LangsmithInputs(BaseModel):
model: Optional[str] = None
messages: Optional[List[Any]] = None
stream: Optional[bool] = None
call_type: Optional[str] = None
litellm_call_id: Optional[str] = None
completion_start_time: Optional[datetime] = None
temperature: Optional[float] = None
max_tokens: Optional[int] = None
custom_llm_provider: Optional[str] = None
input: Optional[List[Any]] = None
log_event_type: Optional[str] = None
original_response: Optional[Any] = None
response_cost: Optional[float] = None
# LiteLLM Virtual Key specific fields
user_api_key: Optional[str] = None
user_api_key_user_id: Optional[str] = None
user_api_key_team_alias: Optional[str] = None
def is_serializable(value):
non_serializable_types = (
@ -19,7 +49,7 @@ def is_serializable(value):
return not isinstance(value, non_serializable_types)
class LangsmithLogger:
class LangsmithLogger(CustomLogger):
# Class variables or attributes
def __init__(self):
self.langsmith_api_key = os.getenv("LANGSMITH_API_KEY")
@ -27,71 +57,121 @@ class LangsmithLogger:
self.langsmith_default_run_name = os.getenv(
"LANGSMITH_DEFAULT_RUN_NAME", "LLMRun"
)
self.langsmith_base_url = os.getenv(
"LANGSMITH_BASE_URL", "https://api.smith.langchain.com"
)
self.async_httpx_client = AsyncHTTPHandler(
timeout=httpx.Timeout(timeout=600.0, connect=5.0)
)
def log_event(self, kwargs, response_obj, start_time, end_time, print_verbose):
# Method definition
# inspired by Langsmith http api here: https://github.com/langchain-ai/langsmith-cookbook/blob/main/tracing-examples/rest/rest.ipynb
metadata = (
kwargs.get("litellm_params", {}).get("metadata", {}) or {}
) # if metadata is None
def _prepare_log_data(self, kwargs, response_obj, start_time, end_time):
import datetime
from datetime import timezone
metadata = kwargs.get("litellm_params", {}).get("metadata", {}) or {}
kwargs["user_api_key"] = metadata.get("user_api_key", None)
kwargs["user_api_key_user_id"] = metadata.get("user_api_key_user_id", None)
kwargs["user_api_key_team_alias"] = metadata.get(
"user_api_key_team_alias", None
)
# set project name and run_name for langsmith logging
# users can pass project_name and run name to litellm.completion()
# Example: litellm.completion(model, messages, metadata={"project_name": "my-litellm-project", "run_name": "my-langsmith-run"})
# if not set litellm will fallback to the environment variable LANGSMITH_PROJECT, then to the default project_name = litellm-completion, run_name = LLMRun
project_name = metadata.get("project_name", self.langsmith_project)
run_name = metadata.get("run_name", self.langsmith_default_run_name)
print_verbose(
run_id = metadata.get("id", None)
verbose_logger.debug(
f"Langsmith Logging - project_name: {project_name}, run_name {run_name}"
)
langsmith_base_url = os.getenv(
"LANGSMITH_BASE_URL", "https://api.smith.langchain.com"
)
try:
print_verbose(
f"Langsmith Logging - Enters logging function for model {kwargs}"
)
import requests
import datetime
from datetime import timezone
start_time = kwargs["start_time"].astimezone(timezone.utc).isoformat()
end_time = kwargs["end_time"].astimezone(timezone.utc).isoformat()
except:
start_time = datetime.datetime.utcnow().isoformat()
end_time = datetime.datetime.utcnow().isoformat()
# filter out kwargs to not include any dicts, langsmith throws an erros when trying to log kwargs
logged_kwargs = LangsmithInputs(**kwargs)
kwargs = logged_kwargs.model_dump()
new_kwargs = {}
for key in kwargs:
value = kwargs[key]
if key == "start_time" or key == "end_time" or value is None:
pass
elif key == "original_response" and not isinstance(value, str):
new_kwargs[key] = str(value)
elif type(value) == datetime.datetime:
new_kwargs[key] = value.isoformat()
elif type(value) != dict and is_serializable(value=value):
new_kwargs[key] = value
elif not is_serializable(value=value):
continue
if isinstance(response_obj, BaseModel):
try:
start_time = kwargs["start_time"].astimezone(timezone.utc).isoformat()
end_time = kwargs["end_time"].astimezone(timezone.utc).isoformat()
response_obj = response_obj.model_dump()
except:
start_time = datetime.datetime.utcnow().isoformat()
end_time = datetime.datetime.utcnow().isoformat()
response_obj = response_obj.dict() # type: ignore
# filter out kwargs to not include any dicts, langsmith throws an erros when trying to log kwargs
new_kwargs = {}
for key in kwargs:
value = kwargs[key]
if key == "start_time" or key == "end_time" or value is None:
pass
elif type(value) == datetime.datetime:
new_kwargs[key] = value.isoformat()
elif type(value) != dict and is_serializable(value=value):
new_kwargs[key] = value
data = {
"name": run_name,
"run_type": "llm", # this should always be llm, since litellm always logs llm calls. Langsmith allow us to log "chain"
"inputs": new_kwargs,
"outputs": response_obj,
"session_name": project_name,
"start_time": start_time,
"end_time": end_time,
}
if isinstance(response_obj, BaseModel):
try:
response_obj = response_obj.model_dump()
except:
response_obj = response_obj.dict() # type: ignore
if run_id:
data["id"] = run_id
data = {
"name": run_name,
"run_type": "llm", # this should always be llm, since litellm always logs llm calls. Langsmith allow us to log "chain"
"inputs": new_kwargs,
"outputs": response_obj,
"session_name": project_name,
"start_time": start_time,
"end_time": end_time,
}
verbose_logger.debug("Langsmith Logging data on langsmith: %s", data)
return data
async def async_log_success_event(self, kwargs, response_obj, start_time, end_time):
try:
verbose_logger.debug(
"Langsmith Async Layer Logging - kwargs: %s, response_obj: %s",
kwargs,
response_obj,
)
data = self._prepare_log_data(kwargs, response_obj, start_time, end_time)
url = f"{self.langsmith_base_url}/runs"
verbose_logger.debug(f"Langsmith Logging - About to send data to {url} ...")
headers = {"x-api-key": self.langsmith_api_key}
response = await self.async_httpx_client.post(
url=url, json=data, headers=headers
)
if response.status_code >= 300:
verbose_logger.error(
f"Langmsith Error: {response.status_code} - {response.text}"
)
else:
verbose_logger.debug(
"Run successfully created, response=%s", response.text
)
verbose_logger.debug(
f"Langsmith Layer Logging - final response object: {response_obj}. Response text from langsmith={response.text}"
)
except:
verbose_logger.error(f"Langsmith Layer Error - {traceback.format_exc()}")
def log_success_event(self, kwargs, response_obj, start_time, end_time):
try:
verbose_logger.debug(
"Langsmith Sync Layer Logging - kwargs: %s, response_obj: %s",
kwargs,
response_obj,
)
data = self._prepare_log_data(kwargs, response_obj, start_time, end_time)
url = f"{self.langsmith_base_url}/runs"
verbose_logger.debug(f"Langsmith Logging - About to send data to {url} ...")
url = f"{langsmith_base_url}/runs"
print_verbose(f"Langsmith Logging - About to send data to {url} ...")
response = requests.post(
url=url,
json=data,
@ -99,12 +179,21 @@ class LangsmithLogger:
)
if response.status_code >= 300:
print_verbose(f"Error: {response.status_code}")
verbose_logger.error(f"Error: {response.status_code} - {response.text}")
else:
print_verbose("Run successfully created")
print_verbose(
f"Langsmith Layer Logging - final response object: {response_obj}"
verbose_logger.debug("Run successfully created")
verbose_logger.debug(
f"Langsmith Layer Logging - final response object: {response_obj}. Response text from langsmith={response.text}"
)
except:
print_verbose(f"Langsmith Layer Error - {traceback.format_exc()}")
pass
verbose_logger.error(f"Langsmith Layer Error - {traceback.format_exc()}")
def get_run_by_id(self, run_id):
url = f"{self.langsmith_base_url}/runs/{run_id}"
response = requests.get(
url=url,
headers={"x-api-key": self.langsmith_api_key},
)
return response.json()

50
litellm/litellm_core_utils/litellm_logging.py
View file

@ -39,7 +39,6 @@ from litellm.utils import (
add_breadcrumb,
capture_exception,
customLogger,
langsmithLogger,
liteDebuggerClient,
logfireLogger,
lunaryLogger,
@ -89,7 +88,6 @@ alerts_channel = None
heliconeLogger = None
athinaLogger = None
promptLayerLogger = None
langsmithLogger = None
logfireLogger = None
weightsBiasesLogger = None
customLogger = None
@ -136,7 +134,7 @@ in_memory_trace_id_cache = ServiceTraceIDCache()
class Logging:
global supabaseClient, liteDebuggerClient, promptLayerLogger, weightsBiasesLogger, langsmithLogger, logfireLogger, capture_exception, add_breadcrumb, lunaryLogger, logfireLogger, prometheusLogger, slack_app
global supabaseClient, liteDebuggerClient, promptLayerLogger, weightsBiasesLogger, logfireLogger, capture_exception, add_breadcrumb, lunaryLogger, logfireLogger, prometheusLogger, slack_app
custom_pricing: bool = False
stream_options = None
@ -738,23 +736,6 @@ class Logging:
end_time=end_time,
print_verbose=print_verbose,
)
if callback == "langsmith":
print_verbose("reaches langsmith for logging!")
if self.stream:
if "complete_streaming_response" not in kwargs:
continue
else:
print_verbose(
"reaches langsmith for streaming logging!"
)
result = kwargs["complete_streaming_response"]
langsmithLogger.log_event(
kwargs=self.model_call_details,
response_obj=result,
start_time=start_time,
end_time=end_time,
print_verbose=print_verbose,
)
if callback == "logfire":
global logfireLogger
verbose_logger.debug("reaches logfire for success logging!")
@ -1337,7 +1318,14 @@ class Logging:
if kwargs.get("no-log", False) == True:
print_verbose("no-log request, skipping logging")
continue
if callback == "cache" and litellm.cache is not None:
if (
callback == "cache"
and litellm.cache is not None
and self.model_call_details.get("litellm_params", {}).get(
"acompletion", False
)
is True
):
# set_cache once complete streaming response is built
print_verbose("async success_callback: reaches cache for logging!")
kwargs = self.model_call_details
@ -1417,6 +1405,9 @@ class Logging:
end_time=end_time,
)
if callable(callback): # custom logger functions
global customLogger
if customLogger is None:
customLogger = CustomLogger()
if self.stream:
if (
"async_complete_streaming_response"
@ -1822,7 +1813,7 @@ def set_callbacks(callback_list, function_id=None):
"""
Globally sets the callback client
"""
global sentry_sdk_instance, capture_exception, add_breadcrumb, posthog, slack_app, alerts_channel, traceloopLogger, athinaLogger, heliconeLogger, aispendLogger, berrispendLogger, supabaseClient, liteDebuggerClient, lunaryLogger, promptLayerLogger, langFuseLogger, customLogger, weightsBiasesLogger, langsmithLogger, logfireLogger, dynamoLogger, s3Logger, dataDogLogger, prometheusLogger, greenscaleLogger, openMeterLogger
global sentry_sdk_instance, capture_exception, add_breadcrumb, posthog, slack_app, alerts_channel, traceloopLogger, athinaLogger, heliconeLogger, aispendLogger, berrispendLogger, supabaseClient, liteDebuggerClient, lunaryLogger, promptLayerLogger, langFuseLogger, customLogger, weightsBiasesLogger, logfireLogger, dynamoLogger, s3Logger, dataDogLogger, prometheusLogger, greenscaleLogger, openMeterLogger
try:
for callback in callback_list:
@ -1903,8 +1894,6 @@ def set_callbacks(callback_list, function_id=None):
s3Logger = S3Logger()
elif callback == "wandb":
weightsBiasesLogger = WeightsBiasesLogger()
elif callback == "langsmith":
langsmithLogger = LangsmithLogger()
elif callback == "logfire":
logfireLogger = LogfireLogger()
elif callback == "aispend":
@ -1957,6 +1946,15 @@ def _init_custom_logger_compatible_class(
_in_memory_loggers.append(_openmeter_logger)
return _openmeter_logger # type: ignore
elif logging_integration == "langsmith":
for callback in _in_memory_loggers:
if isinstance(callback, LangsmithLogger):
return callback # type: ignore
_langsmith_logger = LangsmithLogger()
_in_memory_loggers.append(_langsmith_logger)
return _langsmith_logger # type: ignore
elif logging_integration == "galileo":
for callback in _in_memory_loggers:
if isinstance(callback, GalileoObserve):
@ -2025,6 +2023,10 @@ def get_custom_logger_compatible_class(
for callback in _in_memory_loggers:
if isinstance(callback, GalileoObserve):
return callback
elif logging_integration == "langsmith":
for callback in _in_memory_loggers:
if isinstance(callback, LangsmithLogger):
return callback
elif logging_integration == "logfire":
if "LOGFIRE_TOKEN" not in os.environ:
raise ValueError("LOGFIRE_TOKEN not found in environment variables")

40
litellm/model_prices_and_context_window_backup.json
View file

@ -1020,6 +1020,26 @@
"mode": "chat",
"supports_function_calling": true
},
"groq/llama3-groq-70b-8192-tool-use-preview": {
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00000089,
"output_cost_per_token": 0.00000089,
"litellm_provider": "groq",
"mode": "chat",
"supports_function_calling": true
},
"groq/llama3-groq-8b-8192-tool-use-preview": {
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00000019,
"output_cost_per_token": 0.00000019,
"litellm_provider": "groq",
"mode": "chat",
"supports_function_calling": true
},
"friendliai/mixtral-8x7b-instruct-v0-1": {
"max_tokens": 32768,
"max_input_tokens": 32768,
@ -1800,6 +1820,26 @@
"supports_vision": true,
"source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"
},
"medlm-medium": {
"max_tokens": 8192,
"max_input_tokens": 32768,
"max_output_tokens": 8192,
"input_cost_per_character": 0.0000005,
"output_cost_per_character": 0.000001,
"litellm_provider": "vertex_ai-language-models",
"mode": "chat",
"source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"
},
"medlm-large": {
"max_tokens": 1024,
"max_input_tokens": 8192,
"max_output_tokens": 1024,
"input_cost_per_character": 0.000005,
"output_cost_per_character": 0.000015,
"litellm_provider": "vertex_ai-language-models",
"mode": "chat",
"source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"
},
"vertex_ai/claude-3-sonnet@20240229": {
"max_tokens": 4096,
"max_input_tokens": 200000,

11
litellm/proxy/_new_secret_config.yaml
View file

@ -1,10 +1,5 @@
model_list:
- model_name: "*"
- model_name: llama-3
litellm_params:
model: openai/*
litellm_settings:
guardrails:
- prompt_injection:
callbacks: ["aporio_prompt_injection"]
default_on: true
model: gpt-4
request_timeout: 1

4
litellm/proxy/proxy_config.yaml
View file

@ -17,9 +17,7 @@ model_list:
general_settings:
master_key: sk-1234
litellm_key_header_name: "X-Litellm-Key"
litellm_settings:
cache: true
callbacks: ["otel"]
success_callback: ["langsmith"]

3
litellm/router.py
View file

@ -718,6 +718,9 @@ class Router:
data.get(
"timeout", None
) # timeout set on litellm_params for this deployment
or data.get(
"request_timeout", None
) # timeout set on litellm_params for this deployment
or self.timeout # timeout set on router
or kwargs.get(
"timeout", None

20
litellm/tests/test_caching.py
View file

@ -1579,18 +1579,21 @@ async def test_redis_semantic_cache_acompletion():
assert response1.id == response2.id
def test_caching_redis_simple(caplog):
def test_caching_redis_simple(caplog, capsys):
"""
Relevant issue - https://github.com/BerriAI/litellm/issues/4511
"""
litellm.set_verbose = True ## REQUIRED FOR TEST.
litellm.cache = Cache(
type="redis", url=os.getenv("REDIS_SSL_URL")
) # passing `supported_call_types = ["completion"]` has no effect
s = time.time()
uuid_str = str(uuid.uuid4())
x = completion(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello, how are you? Wink"}],
model="gpt-3.5-turbo",
messages=[{"role": "user", "content": f"Hello, how are you? Wink {uuid_str}"}],
stream=True,
)
for m in x:
@ -1599,8 +1602,8 @@ def test_caching_redis_simple(caplog):
s2 = time.time()
x = completion(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello, how are you? Wink"}],
model="gpt-3.5-turbo",
messages=[{"role": "user", "content": f"Hello, how are you? Wink {uuid_str}"}],
stream=True,
)
for m in x:
@ -1609,11 +1612,15 @@ def test_caching_redis_simple(caplog):
redis_async_caching_error = False
redis_service_logging_error = False
captured = capsys.readouterr()
captured_logs = [rec.message for rec in caplog.records]
print(f"captured_logs: {captured_logs}")
for item in captured_logs:
if "Error connecting to Async Redis client" in item:
if (
"Error connecting to Async Redis client" in item
or "Set ASYNC Redis Cache" in item
):
redis_async_caching_error = True
if "ServiceLogging.async_service_success_hook" in item:
@ -1621,3 +1628,4 @@ def test_caching_redis_simple(caplog):
assert redis_async_caching_error is False
assert redis_service_logging_error is False
assert "async success_callback: reaches cache for logging" not in captured.out

2
litellm/tests/test_completion.py
View file

@ -23,7 +23,7 @@ from litellm import RateLimitError, Timeout, completion, completion_cost, embedd
from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
from litellm.llms.prompt_templates.factory import anthropic_messages_pt
# litellm.num_retries = 3
# litellm.num_retries=3
litellm.cache = None
litellm.success_callback = []
user_message = "Write a short poem about the sky"

27
litellm/tests/test_completion_cost.py
View file

@ -706,6 +706,33 @@ def test_vertex_ai_completion_cost():
print("calculated_input_cost: {}".format(calculated_input_cost))
# @pytest.mark.skip(reason="new test - WIP, working on fixing this")
def test_vertex_ai_medlm_completion_cost():
"""Test for medlm completion cost."""
with pytest.raises(Exception) as e:
model = "vertex_ai/medlm-medium"
messages = [{"role": "user", "content": "Test MedLM completion cost."}]
predictive_cost = completion_cost(
model=model, messages=messages, custom_llm_provider="vertex_ai"
)
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
litellm.model_cost = litellm.get_model_cost_map(url="")
model = "vertex_ai/medlm-medium"
messages = [{"role": "user", "content": "Test MedLM completion cost."}]
predictive_cost = completion_cost(
model=model, messages=messages, custom_llm_provider="vertex_ai"
)
assert predictive_cost > 0
model = "vertex_ai/medlm-large"
messages = [{"role": "user", "content": "Test MedLM completion cost."}]
predictive_cost = completion_cost(model=model, messages=messages)
assert predictive_cost > 0
def test_vertex_ai_claude_completion_cost():
from litellm import Choices, Message, ModelResponse
from litellm.utils import Usage

2
litellm/tests/test_embedding.py
View file

@ -589,7 +589,7 @@ async def test_triton_embeddings():
print(f"response: {response}")
# stubbed endpoint is setup to return this
assert response.data[0]["embedding"] == [0.1, 0.2, 0.3]
assert response.data[0]["embedding"] == [0.1, 0.2]
except Exception as e:
pytest.fail(f"Error occurred: {e}")

168
litellm/tests/test_langsmith.py
View file

@ -1,70 +1,176 @@
import sys
import os
import io
import os
import sys
sys.path.insert(0, os.path.abspath("../.."))
from litellm import completion
import litellm
import asyncio
import logging
import uuid
import pytest
import litellm
from litellm import completion
from litellm._logging import verbose_logger
from litellm.integrations.langsmith import LangsmithLogger
from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler
verbose_logger.setLevel(logging.DEBUG)
litellm.success_callback = ["langsmith"]
litellm.set_verbose = True
import time
def test_langsmith_logging():
@pytest.mark.asyncio()
async def test_async_langsmith_logging():
try:
response = completion(
test_langsmith_logger = LangsmithLogger()
run_id = str(uuid.uuid4())
litellm.set_verbose = True
litellm.callbacks = ["langsmith"]
response = await litellm.acompletion(
model="claude-instant-1.2",
messages=[{"role": "user", "content": "what llm are u"}],
max_tokens=10,
temperature=0.2,
metadata={
"id": run_id,
"user_api_key": "6eb81e014497d89f3cc1aa9da7c2b37bda6b7fea68e4b710d33d94201e68970c",
"user_api_key_alias": "ishaans-langmsith-key",
"user_api_end_user_max_budget": None,
"litellm_api_version": "1.40.19",
"global_max_parallel_requests": None,
"user_api_key_user_id": "admin",
"user_api_key_org_id": None,
"user_api_key_team_id": "dbe2f686-a686-4896-864a-4c3924458709",
"user_api_key_team_alias": "testing-team",
},
)
print(response)
await asyncio.sleep(3)
print("run_id", run_id)
logged_run_on_langsmith = test_langsmith_logger.get_run_by_id(run_id=run_id)
print("logged_run_on_langsmith", logged_run_on_langsmith)
print("fields in logged_run_on_langsmith", logged_run_on_langsmith.keys())
input_fields_on_langsmith = logged_run_on_langsmith.get("inputs")
extra_fields_on_langsmith = logged_run_on_langsmith.get("extra").get(
"invocation_params"
)
print("\nLogged INPUT ON LANGSMITH", input_fields_on_langsmith)
print("\nextra fields on langsmith", extra_fields_on_langsmith)
assert isinstance(input_fields_on_langsmith, dict)
assert "api_key" not in input_fields_on_langsmith
assert "api_key" not in extra_fields_on_langsmith
# assert user_api_key in extra_fields_on_langsmith
assert "user_api_key" in extra_fields_on_langsmith
assert "user_api_key_user_id" in extra_fields_on_langsmith
assert "user_api_key_team_alias" in extra_fields_on_langsmith
for cb in litellm.callbacks:
if isinstance(cb, LangsmithLogger):
await cb.async_httpx_client.client.aclose()
# test_langsmith_logger.async_httpx_client.close()
except Exception as e:
print(e)
pytest.fail(f"Error occurred: {e}")
# test_langsmith_logging()
def test_langsmith_logging_with_metadata():
def test_async_langsmith_logging_with_metadata():
try:
litellm.success_callback = ["langsmith"]
litellm.set_verbose = True
response = completion(
model="gpt-3.5-turbo",
messages=[{"role": "user", "content": "what llm are u"}],
max_tokens=10,
temperature=0.2,
metadata={
"run_name": "litellmRUN",
"project_name": "litellm-completion",
},
)
print(response)
time.sleep(3)
for cb in litellm.callbacks:
if isinstance(cb, LangsmithLogger):
cb.async_httpx_client.close()
except Exception as e:
pytest.fail(f"Error occurred: {e}")
print(e)
# test_langsmith_logging_with_metadata()
def test_langsmith_logging_with_streaming_and_metadata():
@pytest.mark.parametrize("sync_mode", [False, True])
@pytest.mark.asyncio
async def test_async_langsmith_logging_with_streaming_and_metadata(sync_mode):
try:
response = completion(
model="gpt-3.5-turbo",
messages=[{"role": "user", "content": "what llm are u"}],
max_tokens=10,
temperature=0.2,
metadata={
"run_name": "litellmRUN",
"project_name": "litellm-completion",
},
stream=True,
test_langsmith_logger = LangsmithLogger()
litellm.success_callback = ["langsmith"]
litellm.set_verbose = True
run_id = str(uuid.uuid4())
messages = [{"role": "user", "content": "what llm are u"}]
if sync_mode is True:
response = completion(
model="gpt-3.5-turbo",
messages=messages,
max_tokens=10,
temperature=0.2,
stream=True,
metadata={"id": run_id},
)
for cb in litellm.callbacks:
if isinstance(cb, LangsmithLogger):
cb.async_httpx_client = AsyncHTTPHandler()
for chunk in response:
continue
time.sleep(3)
else:
response = await litellm.acompletion(
model="gpt-3.5-turbo",
messages=messages,
max_tokens=10,
temperature=0.2,
mock_response="This is a mock request",
stream=True,
metadata={"id": run_id},
)
for cb in litellm.callbacks:
if isinstance(cb, LangsmithLogger):
cb.async_httpx_client = AsyncHTTPHandler()
async for chunk in response:
continue
await asyncio.sleep(3)
print("run_id", run_id)
logged_run_on_langsmith = test_langsmith_logger.get_run_by_id(run_id=run_id)
print("logged_run_on_langsmith", logged_run_on_langsmith)
print("fields in logged_run_on_langsmith", logged_run_on_langsmith.keys())
input_fields_on_langsmith = logged_run_on_langsmith.get("inputs")
extra_fields_on_langsmith = logged_run_on_langsmith.get("extra").get(
"invocation_params"
)
for chunk in response:
continue
assert logged_run_on_langsmith.get("run_type") == "llm"
print("\nLogged INPUT ON LANGSMITH", input_fields_on_langsmith)
print("\nextra fields on langsmith", extra_fields_on_langsmith)
assert isinstance(input_fields_on_langsmith, dict)
except Exception as e:
pytest.fail(f"Error occurred: {e}")
print(e)
test_langsmith_logging_with_streaming_and_metadata()

2
litellm/tests/test_streaming.py
View file

@ -515,6 +515,7 @@ async def test_completion_predibase_streaming(sync_mode):
response = completion(
model="predibase/llama-3-8b-instruct",
tenant_id="c4768f95",
max_tokens=10,
api_base="https://serving.app.predibase.com",
api_key=os.getenv("PREDIBASE_API_KEY"),
messages=[{"role": "user", "content": "What is the meaning of life?"}],
@ -539,6 +540,7 @@ async def test_completion_predibase_streaming(sync_mode):
response = await litellm.acompletion(
model="predibase/llama-3-8b-instruct",
tenant_id="c4768f95",
max_tokens=10,
api_base="https://serving.app.predibase.com",
api_key=os.getenv("PREDIBASE_API_KEY"),
messages=[{"role": "user", "content": "What is the meaning of life?"}],

15
litellm/utils.py
View file

@ -417,6 +417,21 @@ def function_setup(
# we only support async dynamo db logging for acompletion/aembedding since that's used on proxy
litellm._async_success_callback.append(callback)
removed_async_items.append(index)
elif callback == "langsmith":
callback_class = litellm.litellm_core_utils.litellm_logging._init_custom_logger_compatible_class( # type: ignore
callback, internal_usage_cache=None, llm_router=None
)
# don't double add a callback
if not any(
isinstance(cb, type(callback_class)) for cb in litellm.callbacks
):
litellm.callbacks.append(callback_class) # type: ignore
litellm.input_callback.append(callback_class) # type: ignore
litellm.success_callback.append(callback_class) # type: ignore
litellm.failure_callback.append(callback_class) # type: ignore
litellm._async_success_callback.append(callback_class) # type: ignore
litellm._async_failure_callback.append(callback_class) # type: ignore
# Pop the async items from success_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):

40
model_prices_and_context_window.json
View file

@ -1020,6 +1020,26 @@
"mode": "chat",
"supports_function_calling": true
},
"groq/llama3-groq-70b-8192-tool-use-preview": {
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00000089,
"output_cost_per_token": 0.00000089,
"litellm_provider": "groq",
"mode": "chat",
"supports_function_calling": true
},
"groq/llama3-groq-8b-8192-tool-use-preview": {
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00000019,
"output_cost_per_token": 0.00000019,
"litellm_provider": "groq",
"mode": "chat",
"supports_function_calling": true
},
"friendliai/mixtral-8x7b-instruct-v0-1": {
"max_tokens": 32768,
"max_input_tokens": 32768,
@ -1800,6 +1820,26 @@
"supports_vision": true,
"source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"
},
"medlm-medium": {
"max_tokens": 8192,
"max_input_tokens": 32768,
"max_output_tokens": 8192,
"input_cost_per_character": 0.0000005,
"output_cost_per_character": 0.000001,
"litellm_provider": "vertex_ai-language-models",
"mode": "chat",
"source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"
},
"medlm-large": {
"max_tokens": 1024,
"max_input_tokens": 8192,
"max_output_tokens": 1024,
"input_cost_per_character": 0.000005,
"output_cost_per_character": 0.000015,
"litellm_provider": "vertex_ai-language-models",
"mode": "chat",
"source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"
},
"vertex_ai/claude-3-sonnet@20240229": {
"max_tokens": 4096,
"max_input_tokens": 200000,