Merge branch 'main' into litellm_global_spend_updates

This commit is contained in:
Krish Dholakia 2024-01-24 20:20:15 -08:00 committed by GitHub
commit 6501fdb76e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 166 additions and 24 deletions

View file

@ -888,6 +888,9 @@ def test_call_with_key_over_budget(prisma_client):
# update spend using track_cost callback, make 2nd request, it should fail
from litellm.proxy.proxy_server import track_cost_callback
from litellm import ModelResponse, Choices, Message, Usage
from litellm.caching import Cache
litellm.cache = Cache()
import time
request_id = f"chatcmpl-e41836bb-bb8b-4df2-8e70-8f3e160155ac{time.time()}"
@ -935,6 +938,10 @@ def test_call_with_key_over_budget(prisma_client):
assert spend_log.request_id == request_id
assert spend_log.spend == float("2e-05")
assert spend_log.model == "chatgpt-v-2"
assert (
spend_log.cache_key
== "a61ae14fe4a8b8014a61e6ae01a100c8bc6770ac37c293242afed954bc69207d"
)
# use generated key to auth in
result = await user_api_key_auth(request=request, api_key=bearer_token)
@ -948,6 +955,76 @@ def test_call_with_key_over_budget(prisma_client):
print(vars(e))
@pytest.mark.asyncio()
async def test_call_with_key_never_over_budget(prisma_client):
# Make a call with a key with budget=None, it should never fail
setattr(litellm.proxy.proxy_server, "prisma_client", prisma_client)
setattr(litellm.proxy.proxy_server, "master_key", "sk-1234")
try:
await litellm.proxy.proxy_server.prisma_client.connect()
request = GenerateKeyRequest(max_budget=None)
key = await generate_key_fn(request)
print(key)
generated_key = key.key
user_id = key.user_id
bearer_token = "Bearer " + generated_key
request = Request(scope={"type": "http"})
request._url = URL(url="/chat/completions")
# use generated key to auth in
result = await user_api_key_auth(request=request, api_key=bearer_token)
print("result from user auth with new key", result)
# update spend using track_cost callback, make 2nd request, it should fail
from litellm.proxy.proxy_server import track_cost_callback
from litellm import ModelResponse, Choices, Message, Usage
import time
request_id = f"chatcmpl-e41836bb-bb8b-4df2-8e70-8f3e160155ac{time.time()}"
resp = ModelResponse(
id=request_id,
choices=[
Choices(
finish_reason=None,
index=0,
message=Message(
content=" Sure! Here is a short poem about the sky:\n\nA canvas of blue, a",
role="assistant",
),
)
],
model="gpt-35-turbo", # azure always has model written like this
usage=Usage(
prompt_tokens=210000, completion_tokens=200000, total_tokens=41000
),
)
await track_cost_callback(
kwargs={
"model": "chatgpt-v-2",
"stream": False,
"litellm_params": {
"metadata": {
"user_api_key": generated_key,
"user_api_key_user_id": user_id,
}
},
"response_cost": 200000,
},
completion_response=resp,
start_time=datetime.now(),
end_time=datetime.now(),
)
# use generated key to auth in
result = await user_api_key_auth(request=request, api_key=bearer_token)
print("result from user auth with new key", result)
except Exception as e:
pytest.fail(f"This should have not failed!. They key uses max_budget=None. {e}")
@pytest.mark.asyncio()
async def test_call_with_key_over_budget_stream(prisma_client):
# 14. Make a call with a key over budget, expect to fail