phoenix/litellm-mirror
0
0
Fork 1
mirror of https://github.com/BerriAI/litellm.git synced 2025-04-26 11:14:04 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

inly write model tpm/rpm tracking when user set it

Browse source
This commit is contained in:
Ishaan Jaff 2024-08-18 09:57:31 -07:00
parent db30aa6382
commit 94e74b9ede
2 changed files with 33 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

14
litellm/proxy/hooks/parallel_request_limiter.py
View file

@ -400,6 +400,11 @@ class _PROXY_MaxParallelRequestsHandler(CustomLogger):
)
user_api_key_end_user_id = kwargs.get("user")
user_api_key_metadata = (
kwargs["litellm_params"]["metadata"].get("user_api_key_metadata", {})
or {}
)
# ------------
# Setup values
# ------------
@ -456,7 +461,14 @@ class _PROXY_MaxParallelRequestsHandler(CustomLogger):
# Update usage - model group + API Key
# ------------
model_group = get_model_group_from_litellm_kwargs(kwargs)
if user_api_key is not None and model_group is not None:
if (
user_api_key is not None
and model_group is not None
and (
"model_rpm_limit" in user_api_key_metadata
or "model_tpm_limit" in user_api_key_metadata
)
):
request_count_api_key = (
f"{user_api_key}::{model_group}::{precise_minute}::request_count"
)