fix(router.py): handle initial scenario for tpm/rpm routing

This commit is contained in:
Krrish Dholakia 2023-12-30 07:28:13 +05:30
parent 05795ead59
commit a34de56289
3 changed files with 9 additions and 4 deletions

View file

@ -1606,7 +1606,12 @@ class Router:
self.routing_strategy == "usage-based-routing"
and self.lowesttpm_logger is not None
):
return self.lowesttpm_logger.get_available_deployments(model_group=model)
min_deployment = self.lowesttpm_logger.get_available_deployments(
model_group=model
)
if min_deployment is None:
min_deployment = random.choice(healthy_deployments)
return min_deployment
raise ValueError("No models available.")

View file

@ -136,6 +136,9 @@ class LowestTPMLoggingHandler(CustomLogger):
lowest_tpm = float("inf")
deployment = None
if tpm_dict is None: # base case
return
for item, item_tpm in tpm_dict.items():
## get the item from model list
_deployment = None

View file

@ -128,6 +128,3 @@ def test_router_get_available_deployments():
assert return_dict[1] == 10
assert return_dict[2] == 54
assert return_dict[3] == 100
# test_router_get_available_deployments()