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

fix(router.py): handle 1 deployment being picked

Browse source
This commit is contained in:
Krrish Dholakia 2024-04-10 18:32:54 -07:00
parent 52462e8bac
commit 266dba65e7
1 changed files with 21 additions and 24 deletions
Download patch file Download diff file Expand all files Collapse all files

45
litellm/router.py
View file

@ -2434,8 +2434,22 @@ class Router:
):
"""
Common checks for 'get_available_deployment' across sync + async call.
If 'healthy_deployments' returned is None, this means the user chose a specific deployment
"""
# check if aliases set on litellm model alias map
if specific_deployment == True:
# users can also specify a specific deployment name. At this point we should check if they are just trying to call a specific deployment
for deployment in self.model_list:
deployment_model = deployment.get("litellm_params").get("model")
if deployment_model == model:
# User Passed a specific deployment name on their config.yaml, example azure/chat-gpt-v-2
# return the first deployment where the `model` matches the specificed deployment name
return deployment, None
raise ValueError(
f"LiteLLM Router: Trying to call specific deployment, but Model:{model} does not exist in Model List: {self.model_list}"
)
if model in self.model_group_alias:
verbose_router_logger.debug(
f"Using a model alias. Got Request for {model}, sending requests to {self.model_group_alias.get(model)}"
@ -2447,7 +2461,7 @@ class Router:
self.default_deployment
) # self.default_deployment
updated_deployment["litellm_params"]["model"] = model
return updated_deployment
return updated_deployment, None
## get healthy deployments
### get all deployments
@ -2496,18 +2510,6 @@ class Router:
specific_deployment=specific_deployment,
)
if specific_deployment == True:
# users can also specify a specific deployment name. At this point we should check if they are just trying to call a specific deployment
for deployment in self.model_list:
deployment_model = deployment.get("litellm_params").get("model")
if deployment_model == model:
# User Passed a specific deployment name on their config.yaml, example azure/chat-gpt-v-2
# return the first deployment where the `model` matches the specificed deployment name
return deployment
raise ValueError(
f"LiteLLM Router: Trying to call specific deployment, but Model:{model} does not exist in Model List: {self.model_list}"
)
model, healthy_deployments = self._common_checks_available_deployment(
model=model,
messages=messages,
@ -2515,6 +2517,9 @@ class Router:
specific_deployment=specific_deployment,
)
if healthy_deployments is None:
return model
# filter out the deployments currently cooling down
deployments_to_remove = []
# cooldown_deployments is a list of model_id's cooling down, cooldown_deployments = ["16700539-b3cd-42f4-b426-6a12a1bb706a", "16700539-b3cd-42f4-b426-7899"]
@ -2572,17 +2577,6 @@ class Router:
"""
# users need to explicitly call a specific deployment, by setting `specific_deployment = True` as completion()/embedding() kwarg
# When this was no explicit we had several issues with fallbacks timing out
if specific_deployment == True:
# users can also specify a specific deployment name. At this point we should check if they are just trying to call a specific deployment
for deployment in self.model_list:
deployment_model = deployment.get("litellm_params").get("model")
if deployment_model == model:
# User Passed a specific deployment name on their config.yaml, example azure/chat-gpt-v-2
# return the first deployment where the `model` matches the specificed deployment name
return deployment
raise ValueError(
f"LiteLLM Router: Trying to call specific deployment, but Model:{model} does not exist in Model List: {self.model_list}"
)
model, healthy_deployments = self._common_checks_available_deployment(
model=model,
@ -2591,6 +2585,9 @@ class Router:
specific_deployment=specific_deployment,
)
if healthy_deployments is None:
return model
# filter out the deployments currently cooling down
deployments_to_remove = []
# cooldown_deployments is a list of model_id's cooling down, cooldown_deployments = ["16700539-b3cd-42f4-b426-6a12a1bb706a", "16700539-b3cd-42f4-b426-7899"]