litellm/tests/local_testing/test_router_get_deployments.py

593 lines
21 KiB
Python

# Tests for router.get_available_deployment
# specifically test if it can pick the correct LLM when rpm/tpm set
# These are fast Tests, and make no API calls
import asyncio
import os
import sys
import time
import traceback
import pytest
sys.path.insert(
0, os.path.abspath("../..")
) # Adds the parent directory to the system path
from collections import defaultdict
from concurrent.futures import ThreadPoolExecutor
from dotenv import load_dotenv
import litellm
from litellm import Router
load_dotenv()
def test_weighted_selection_router():
# this tests if load balancing works based on the provided rpms in the router
# it's a fast test, only tests get_available_deployment
# users can pass rpms as a litellm_param
try:
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
"rpm": 6,
},
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
"rpm": 1440,
},
},
]
router = Router(
model_list=model_list,
)
selection_counts = defaultdict(int)
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
for _ in range(1000):
selected_model = router.get_available_deployment("gpt-3.5-turbo")
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["azure/chatgpt-v-2"] / total_requests > 0.89
), f"Assertion failed: 'azure/chatgpt-v-2' does not have about 90% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")
# test_weighted_selection_router()
def test_weighted_selection_router_tpm():
# this tests if load balancing works based on the provided tpms in the router
# it's a fast test, only tests get_available_deployment
# users can pass rpms as a litellm_param
try:
print("\ntest weighted selection based on TPM\n")
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
"tpm": 5,
},
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
"tpm": 90,
},
},
]
router = Router(
model_list=model_list,
)
selection_counts = defaultdict(int)
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
for _ in range(1000):
selected_model = router.get_available_deployment("gpt-3.5-turbo")
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["azure/chatgpt-v-2"] / total_requests > 0.89
), f"Assertion failed: 'azure/chatgpt-v-2' does not have about 90% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")
# test_weighted_selection_router_tpm()
def test_weighted_selection_router_tpm_as_router_param():
# this tests if load balancing works based on the provided tpms in the router
# it's a fast test, only tests get_available_deployment
# users can pass rpms as a litellm_param
try:
print("\ntest weighted selection based on TPM\n")
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
},
"tpm": 5,
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
},
"tpm": 90,
},
]
router = Router(
model_list=model_list,
)
selection_counts = defaultdict(int)
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
for _ in range(1000):
selected_model = router.get_available_deployment("gpt-3.5-turbo")
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["azure/chatgpt-v-2"] / total_requests > 0.89
), f"Assertion failed: 'azure/chatgpt-v-2' does not have about 90% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")
# test_weighted_selection_router_tpm_as_router_param()
def test_weighted_selection_router_rpm_as_router_param():
# this tests if load balancing works based on the provided tpms in the router
# it's a fast test, only tests get_available_deployment
# users can pass rpms as a litellm_param
try:
print("\ntest weighted selection based on RPM\n")
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
},
"rpm": 5,
"tpm": 5,
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
},
"rpm": 90,
"tpm": 90,
},
]
router = Router(
model_list=model_list,
)
selection_counts = defaultdict(int)
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
for _ in range(1000):
selected_model = router.get_available_deployment("gpt-3.5-turbo")
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["azure/chatgpt-v-2"] / total_requests > 0.89
), f"Assertion failed: 'azure/chatgpt-v-2' does not have about 90% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")
# test_weighted_selection_router_tpm_as_router_param()
def test_weighted_selection_router_no_rpm_set():
# this tests if we can do selection when no rpm is provided too
# it's a fast test, only tests get_available_deployment
# users can pass rpms as a litellm_param
try:
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
"rpm": 6,
},
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
"rpm": 1440,
},
},
{
"model_name": "claude-1",
"litellm_params": {
"model": "bedrock/claude1.2",
"rpm": 1440,
},
},
]
router = Router(
model_list=model_list,
)
selection_counts = defaultdict(int)
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
for _ in range(1000):
selected_model = router.get_available_deployment("claude-1")
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["bedrock/claude1.2"] / total_requests == 1
), f"Assertion failed: Selection counts {selection_counts}"
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")
# test_weighted_selection_router_no_rpm_set()
def test_model_group_aliases():
try:
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
"tpm": 1,
},
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
"tpm": 99,
},
},
{
"model_name": "claude-1",
"litellm_params": {
"model": "bedrock/claude1.2",
"tpm": 1,
},
},
]
router = Router(
model_list=model_list,
model_group_alias={
"gpt-4": "gpt-3.5-turbo"
}, # gpt-4 requests sent to gpt-3.5-turbo
)
# test that gpt-4 requests are sent to gpt-3.5-turbo
for _ in range(20):
selected_model = router.get_available_deployment("gpt-4")
print("\n selected model", selected_model)
selected_model_name = selected_model.get("model_name")
if selected_model_name != "gpt-3.5-turbo":
pytest.fail(
f"Selected model {selected_model_name} is not gpt-3.5-turbo"
)
# test that
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
selection_counts = defaultdict(int)
for _ in range(1000):
selected_model = router.get_available_deployment("gpt-3.5-turbo")
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["azure/chatgpt-v-2"] / total_requests > 0.89
), f"Assertion failed: 'azure/chatgpt-v-2' does not have about 90% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")
# test_model_group_aliases()
def test_usage_based_routing():
"""
in this test we, have a model group with two models in it, model-a and model-b.
Then at some point, we exceed the TPM limit (set in the litellm_params)
for model-a only; but for model-b we are still under the limit
"""
try:
def get_azure_params(deployment_name: str):
params = {
"model": f"azure/{deployment_name}",
"api_key": os.environ["AZURE_API_KEY"],
"api_version": os.environ["AZURE_API_VERSION"],
"api_base": os.environ["AZURE_API_BASE"],
}
return params
model_list = [
{
"model_name": "azure/gpt-4",
"litellm_params": get_azure_params("chatgpt-low-tpm"),
"tpm": 100,
},
{
"model_name": "azure/gpt-4",
"litellm_params": get_azure_params("chatgpt-high-tpm"),
"tpm": 1000,
},
]
router = Router(
model_list=model_list,
set_verbose=True,
debug_level="DEBUG",
routing_strategy="usage-based-routing",
redis_host=os.environ["REDIS_HOST"],
redis_port=os.environ["REDIS_PORT"],
)
messages = [
{"content": "Tell me a joke.", "role": "user"},
]
selection_counts = defaultdict(int)
for _ in range(25):
response = router.completion(
model="azure/gpt-4",
messages=messages,
timeout=5,
mock_response="good morning",
)
# print("response", response)
selection_counts[response["model"]] += 1
print("selection counts", selection_counts)
total_requests = sum(selection_counts.values())
# Assert that 'chatgpt-low-tpm' has more than 2 requests
assert (
selection_counts["chatgpt-low-tpm"] > 2
), f"Assertion failed: 'chatgpt-low-tpm' does not have more than 2 request in the weighted load balancer. Selection counts {selection_counts}"
# Assert that 'chatgpt-high-tpm' has about 70% of the total requests [DO NOT MAKE THIS LOWER THAN 70%]
assert (
selection_counts["chatgpt-high-tpm"] / total_requests > 0.70
), f"Assertion failed: 'chatgpt-high-tpm' does not have about 80% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
except Exception as e:
pytest.fail(f"Error occurred: {e}")
@pytest.mark.asyncio
async def test_wildcard_openai_routing():
"""
Initialize router with *, all models go through * and use OPENAI_API_KEY
"""
try:
model_list = [
{
"model_name": "*",
"litellm_params": {
"model": "openai/*",
"api_key": os.getenv("OPENAI_API_KEY"),
},
"tpm": 100,
},
]
router = Router(
model_list=model_list,
)
messages = [
{"content": "Tell me a joke.", "role": "user"},
]
selection_counts = defaultdict(int)
for _ in range(25):
response = await router.acompletion(
model="gpt-4",
messages=messages,
mock_response="good morning",
)
# print("response1", response)
selection_counts[response["model"]] += 1
response = await router.acompletion(
model="gpt-3.5-turbo",
messages=messages,
mock_response="good morning",
)
# print("response2", response)
selection_counts[response["model"]] += 1
response = await router.acompletion(
model="gpt-4-turbo-preview",
messages=messages,
mock_response="good morning",
)
# print("response3", response)
# print("response", response)
selection_counts[response["model"]] += 1
assert selection_counts["gpt-4"] == 25
assert selection_counts["gpt-3.5-turbo"] == 25
assert selection_counts["gpt-4-turbo-preview"] == 25
except Exception as e:
pytest.fail(f"Error occurred: {e}")
"""
Test async router get deployment (Simpl-shuffle)
"""
rpm_list = [[None, None], [6, 1440]]
tpm_list = [[None, None], [6, 1440]]
@pytest.mark.asyncio
@pytest.mark.parametrize(
"rpm_list, tpm_list",
[(rpm, tpm) for rpm in rpm_list for tpm in tpm_list],
)
async def test_weighted_selection_router_async(rpm_list, tpm_list):
# this tests if load balancing works based on the provided rpms in the router
# it's a fast test, only tests get_available_deployment
# users can pass rpms as a litellm_param
try:
litellm.set_verbose = False
model_list = [
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "gpt-3.5-turbo",
"api_key": os.getenv("OPENAI_API_KEY"),
"rpm": rpm_list[0],
"tpm": tpm_list[0],
},
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {
"model": "azure/chatgpt-v-2",
"api_key": os.getenv("AZURE_API_KEY"),
"api_base": os.getenv("AZURE_API_BASE"),
"api_version": os.getenv("AZURE_API_VERSION"),
"rpm": rpm_list[1],
"tpm": tpm_list[1],
},
},
]
router = Router(
model_list=model_list,
)
selection_counts = defaultdict(int)
# call get_available_deployment 1k times, it should pick azure/chatgpt-v-2 about 90% of the time
for _ in range(1000):
selected_model = await router.async_get_available_deployment(
"gpt-3.5-turbo"
)
selected_model_id = selected_model["litellm_params"]["model"]
selected_model_name = selected_model_id
selection_counts[selected_model_name] += 1
print(selection_counts)
total_requests = sum(selection_counts.values())
if rpm_list[0] is not None or tpm_list[0] is not None:
# Assert that 'azure/chatgpt-v-2' has about 90% of the total requests
assert (
selection_counts["azure/chatgpt-v-2"] / total_requests > 0.89
), f"Assertion failed: 'azure/chatgpt-v-2' does not have about 90% of the total requests in the weighted load balancer. Selection counts {selection_counts}"
else:
# Assert both are used
assert selection_counts["azure/chatgpt-v-2"] > 0
assert selection_counts["gpt-3.5-turbo"] > 0
router.reset()
except Exception as e:
traceback.print_exc()
pytest.fail(f"Error occurred: {e}")