🔥 Load Balancing, Fallbacks, Retries, Timeouts

Quick Start load balancing
Quick Start client side fallbacks

Quick Start - Load Balancing

Step 1 - Set deployments on config

Example config below. Here requests with model=gpt-3.5-turbo will be routed across multiple instances of azure/gpt-3.5-turbo

model_list:
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/<your-deployment-name>
      api_base: <your-azure-endpoint>
      api_key: <your-azure-api-key>
      rpm: 6      # Rate limit for this deployment: in requests per minute (rpm)
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-small-ca
      api_base: https://my-endpoint-canada-berri992.openai.azure.com/
      api_key: <your-azure-api-key>
      rpm: 6
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-large
      api_base: https://openai-france-1234.openai.azure.com/
      api_key: <your-azure-api-key>
      rpm: 1440
routing_strategy: simple-shuffle # Literal["simple-shuffle", "least-busy", "usage-based-routing","latency-based-routing"], default="simple-shuffle"
  model_group_alias: {"gpt-4": "gpt-3.5-turbo"} # all requests with `gpt-4` will be routed to models with `gpt-3.5-turbo`
  num_retries: 2
  timeout: 30                                  # 30 seconds
  redis_host: <your redis host>                # set this when using multiple litellm proxy deployments, load balancing state stored in redis
  redis_password: <your redis password>
  redis_port: 1992

info

Detailed information about routing strategies can be found here

Step 2: Start Proxy with config

$ litellm --config /path/to/config.yaml

Test - Simple Call

Here requests with model=gpt-3.5-turbo will be routed across multiple instances of azure/gpt-3.5-turbo

👉 Key Change: model="gpt-3.5-turbo"

Check the model_id in Response Headers to make sure the requests are being load balanced

OpenAI Python v1.0.0+
Curl Request
Langchain

import openai
client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages = [
        {
            "role": "user",
            "content": "this is a test request, write a short poem"
        }
    ]
)

print(response)

Pass metadata as part of the request body

curl --location 'http://0.0.0.0:4000/chat/completions' \
    --header 'Content-Type: application/json' \
    --data '{
    "model": "gpt-3.5-turbo",
    "messages": [
        {
        "role": "user",
        "content": "what llm are you"
        }
    ]
}'

from langchain.chat_models import ChatOpenAI
from langchain.prompts.chat import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    SystemMessagePromptTemplate,
)
from langchain.schema import HumanMessage, SystemMessage
import os 

os.environ["OPENAI_API_KEY"] = "anything"

chat = ChatOpenAI(
    openai_api_base="http://0.0.0.0:4000",
    model="gpt-3.5-turbo",
)

messages = [
    SystemMessage(
        content="You are a helpful assistant that im using to make a test request to."
    ),
    HumanMessage(
        content="test from litellm. tell me why it's amazing in 1 sentence"
    ),
]
response = chat(messages)

print(response)

Test - Loadbalancing

In this request, the following will occur:

A rate limit exception will be raised
LiteLLM proxy will retry the request on the model group (default is 3).

curl -X POST 'http://0.0.0.0:4000/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer sk-1234' \
-d '{
  "model": "gpt-3.5-turbo",
  "messages": [
        {"role": "user", "content": "Hi there!"}
    ],
    "mock_testing_rate_limit_error": true
}'

See Code

Test - Client Side Fallbacks

In this request the following will occur:

The request to model="zephyr-beta" will fail
litellm proxy will loop through all the model_groups specified in fallbacks=["gpt-3.5-turbo"]
The request to model="gpt-3.5-turbo" will succeed and the client making the request will get a response from gpt-3.5-turbo

👉 Key Change: "fallbacks": ["gpt-3.5-turbo"]

OpenAI Python v1.0.0+
Curl Request
Langchain

import openai
client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

response = client.chat.completions.create(
    model="zephyr-beta",
    messages = [
        {
            "role": "user",
            "content": "this is a test request, write a short poem"
        }
    ],
    extra_body={
        "fallbacks": ["gpt-3.5-turbo"]
    }
)

print(response)

Pass metadata as part of the request body

curl --location 'http://0.0.0.0:4000/chat/completions' \
    --header 'Content-Type: application/json' \
    --data '{
    "model": "zephyr-beta"",
    "messages": [
        {
        "role": "user",
        "content": "what llm are you"
        }
    ],
    "fallbacks": ["gpt-3.5-turbo"]
}'

from langchain.chat_models import ChatOpenAI
from langchain.prompts.chat import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    SystemMessagePromptTemplate,
)
from langchain.schema import HumanMessage, SystemMessage
import os 

os.environ["OPENAI_API_KEY"] = "anything"

chat = ChatOpenAI(
    openai_api_base="http://0.0.0.0:4000",
    model="zephyr-beta",
    extra_body={
        "fallbacks": ["gpt-3.5-turbo"]
    }
)

messages = [
    SystemMessage(
        content="You are a helpful assistant that im using to make a test request to."
    ),
    HumanMessage(
        content="test from litellm. tell me why it's amazing in 1 sentence"
    ),
]
response = chat(messages)

print(response)

Advanced

Fallbacks + Retries + Timeouts + Cooldowns

Set via config

model_list:
  - model_name: zephyr-beta
    litellm_params:
        model: huggingface/HuggingFaceH4/zephyr-7b-beta
        api_base: http://0.0.0.0:8001
  - model_name: zephyr-beta
    litellm_params:
        model: huggingface/HuggingFaceH4/zephyr-7b-beta
        api_base: http://0.0.0.0:8002
  - model_name: zephyr-beta
    litellm_params:
        model: huggingface/HuggingFaceH4/zephyr-7b-beta
        api_base: http://0.0.0.0:8003
  - model_name: gpt-3.5-turbo
    litellm_params:
        model: gpt-3.5-turbo
        api_key: <my-openai-key>
  - model_name: gpt-3.5-turbo-16k
    litellm_params:
        model: gpt-3.5-turbo-16k
        api_key: <my-openai-key>

litellm_settings:
  num_retries: 3 # retry call 3 times on each model_name (e.g. zephyr-beta)
  request_timeout: 10 # raise Timeout error if call takes longer than 10s. Sets litellm.request_timeout 
  fallbacks: [{"zephyr-beta": ["gpt-3.5-turbo"]}] # fallback to gpt-3.5-turbo if call fails num_retries 
  context_window_fallbacks: [{"zephyr-beta": ["gpt-3.5-turbo-16k"]}, {"gpt-3.5-turbo": ["gpt-3.5-turbo-16k"]}] # fallback to gpt-3.5-turbo-16k if context window error
  allowed_fails: 3 # cooldown model if it fails > 1 call in a minute. 
  cooldown_time: 30 # how long to cooldown model if fails/min > allowed_fails

Context Window Fallbacks (Pre-Call Checks + Fallbacks)

Before call is made check if a call is within model context window with enable_pre_call_checks: true.

See Code

1. Setup config

For azure deployments, set the base model. Pick the base model from this list, all the azure models start with azure/.

Same Group
Context Window Fallbacks (Different Groups)

Filter older instances of a model (e.g. gpt-3.5-turbo) with smaller context windows

router_settings:
    enable_pre_call_checks: true # 1. Enable pre-call checks

model_list:
    - model_name: gpt-3.5-turbo
      litellm_params:
        model: azure/chatgpt-v-2
        api_base: os.environ/AZURE_API_BASE
        api_key: os.environ/AZURE_API_KEY
        api_version: "2023-07-01-preview"
      model_info:
        base_model: azure/gpt-4-1106-preview # 2. 👈 (azure-only) SET BASE MODEL
    
    - model_name: gpt-3.5-turbo
      litellm_params:
        model: gpt-3.5-turbo-1106
        api_key: os.environ/OPENAI_API_KEY

2. Start proxy

litellm --config /path/to/config.yaml

# RUNNING on http://0.0.0.0:4000

3. Test it!

import openai
client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

text = "What is the meaning of 42?" * 5000

# request sent to model set on litellm proxy, `litellm --model`
response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages = [
        {"role": "system", "content": text},
        {"role": "user", "content": "Who was Alexander?"},
    ],
)

print(response)

Fallback to larger models if current model is too small.

router_settings:
    enable_pre_call_checks: true # 1. Enable pre-call checks

model_list:
    - model_name: gpt-3.5-turbo-small
      litellm_params:
        model: azure/chatgpt-v-2
      api_base: os.environ/AZURE_API_BASE
      api_key: os.environ/AZURE_API_KEY
      api_version: "2023-07-01-preview"
      model_info:
      base_model: azure/gpt-4-1106-preview # 2. 👈 (azure-only) SET BASE MODEL
    
    - model_name: gpt-3.5-turbo-large
      litellm_params:
      model: gpt-3.5-turbo-1106
      api_key: os.environ/OPENAI_API_KEY

  - model_name: claude-opus
    litellm_params:
      model: claude-3-opus-20240229
      api_key: os.environ/ANTHROPIC_API_KEY

litellm_settings:
  context_window_fallbacks: [{"gpt-3.5-turbo-small": ["gpt-3.5-turbo-large", "claude-opus"]}]

2. Start proxy

litellm --config /path/to/config.yaml

# RUNNING on http://0.0.0.0:4000

3. Test it!

import openai
client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

text = "What is the meaning of 42?" * 5000

# request sent to model set on litellm proxy, `litellm --model`
response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages = [
        {"role": "system", "content": text},
        {"role": "user", "content": "Who was Alexander?"},
    ],
)

print(response)

Content Policy Fallbacks

Fallback across providers (e.g. from Azure OpenAI to Anthropic) if you hit content policy violation errors.

model_list:
    - model_name: gpt-3.5-turbo-small
      litellm_params:
        model: azure/chatgpt-v-2
        api_base: os.environ/AZURE_API_BASE
        api_key: os.environ/AZURE_API_KEY
        api_version: "2023-07-01-preview"

    - model_name: claude-opus
      litellm_params:
        model: claude-3-opus-20240229
        api_key: os.environ/ANTHROPIC_API_KEY

litellm_settings:
  content_policy_fallbacks: [{"gpt-3.5-turbo-small": ["claude-opus"]}]

Default Fallbacks

You can also set default_fallbacks, in case a specific model group is misconfigured / bad.

model_list:
    - model_name: gpt-3.5-turbo-small
      litellm_params:
        model: azure/chatgpt-v-2
        api_base: os.environ/AZURE_API_BASE
        api_key: os.environ/AZURE_API_KEY
        api_version: "2023-07-01-preview"

    - model_name: claude-opus
      litellm_params:
        model: claude-3-opus-20240229
        api_key: os.environ/ANTHROPIC_API_KEY

litellm_settings:
  default_fallbacks: ["claude-opus"]

This will default to claude-opus in case any model fails.

A model-specific fallbacks (e.g. {"gpt-3.5-turbo-small": ["claude-opus"]}) overrides default fallback.

Test Fallbacks!

Check if your fallbacks are working as expected.

Regular Fallbacks

curl -X POST 'http://0.0.0.0:4000/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer sk-1234' \
-D '{
  "model": "my-bad-model",
  "messages": [
    {
      "role": "user",
      "content": "ping"
    }
  ],
  "mock_testing_fallbacks": true # 👈 KEY CHANGE
}
'

Content Policy Fallbacks

curl -X POST 'http://0.0.0.0:4000/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer sk-1234' \
-D '{
  "model": "my-bad-model",
  "messages": [
    {
      "role": "user",
      "content": "ping"
    }
  ],
  "mock_testing_content_policy_fallbacks": true # 👈 KEY CHANGE
}
'

Context Window Fallbacks

curl -X POST 'http://0.0.0.0:4000/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer sk-1234' \
-D '{
  "model": "my-bad-model",
  "messages": [
    {
      "role": "user",
      "content": "ping"
    }
  ],
  "mock_testing_context_window_fallbacks": true # 👈 KEY CHANGE
}
'

EU-Region Filtering (Pre-Call Checks)

Before call is made check if a call is within model context window with enable_pre_call_checks: true.

Set 'region_name' of deployment.

Note: LiteLLM can automatically infer region_name for Vertex AI, Bedrock, and IBM WatsonxAI based on your litellm params. For Azure, set litellm.enable_preview = True.

1. Set Config

router_settings:
    enable_pre_call_checks: true # 1. Enable pre-call checks

model_list:
- model_name: gpt-3.5-turbo
  litellm_params:
    model: azure/chatgpt-v-2
    api_base: os.environ/AZURE_API_BASE
    api_key: os.environ/AZURE_API_KEY
    api_version: "2023-07-01-preview"
    region_name: "eu" # 👈 SET EU-REGION

- model_name: gpt-3.5-turbo
  litellm_params:
    model: gpt-3.5-turbo-1106
    api_key: os.environ/OPENAI_API_KEY

- model_name: gemini-pro
  litellm_params:
    model: vertex_ai/gemini-pro-1.5
    vertex_project: adroit-crow-1234
    vertex_location: us-east1 # 👈 AUTOMATICALLY INFERS 'region_name'

2. Start proxy

litellm --config /path/to/config.yaml

# RUNNING on http://0.0.0.0:4000

3. Test it!

import openai
client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

# request sent to model set on litellm proxy, `litellm --model`
response = client.chat.completions.with_raw_response.create(
    model="gpt-3.5-turbo",
    messages = [{"role": "user", "content": "Who was Alexander?"}]
)

print(response)

print(f"response.headers.get('x-litellm-model-api-base')")

Custom Timeouts, Stream Timeouts - Per Model

For each model you can set timeout & stream_timeout under litellm_params

model_list:
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-small-eu
      api_base: https://my-endpoint-europe-berri-992.openai.azure.com/
      api_key: <your-key>
      timeout: 0.1                      # timeout in (seconds)
      stream_timeout: 0.01              # timeout for stream requests (seconds)
      max_retries: 5
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-small-ca
      api_base: https://my-endpoint-canada-berri992.openai.azure.com/
      api_key: 
      timeout: 0.1                      # timeout in (seconds)
      stream_timeout: 0.01              # timeout for stream requests (seconds)
      max_retries: 5

Start Proxy

$ litellm --config /path/to/config.yaml

Setting Dynamic Timeouts - Per Request

LiteLLM Proxy supports setting a timeout per request

Example Usage

Curl Request
OpenAI v1.0.0+

curl --location 'http://0.0.0.0:4000/chat/completions' \
     --header 'Content-Type: application/json' \
     --data-raw '{
        "model": "gpt-3.5-turbo",
        "messages": [
            {"role": "user", "content": "what color is red"}
        ],
        "logit_bias": {12481: 100},
        "timeout": 1
     }'

import openai


client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=[
        {"role": "user", "content": "what color is red"}
    ],
    logit_bias={12481: 100},
    timeout=1
)

print(response)

🔥 Load Balancing, Fallbacks, Retries, Timeouts

Quick Start - Load Balancing​

Step 1 - Set deployments on config​

Step 2: Start Proxy with config​

Test - Simple Call​

Test - Loadbalancing​

Test - Client Side Fallbacks​

Advanced​

Fallbacks + Retries + Timeouts + Cooldowns​

Context Window Fallbacks (Pre-Call Checks + Fallbacks)​

Content Policy Fallbacks​

Default Fallbacks​

Test Fallbacks!​

Regular Fallbacks​

Content Policy Fallbacks​

Context Window Fallbacks​

EU-Region Filtering (Pre-Call Checks)​

Custom Timeouts, Stream Timeouts - Per Model​

Start Proxy​

Setting Dynamic Timeouts - Per Request​

Quick Start - Load Balancing

Step 1 - Set deployments on config

Step 2: Start Proxy with config

Test - Simple Call

Test - Loadbalancing

Test - Client Side Fallbacks

Advanced

Fallbacks + Retries + Timeouts + Cooldowns

Context Window Fallbacks (Pre-Call Checks + Fallbacks)

Content Policy Fallbacks

Default Fallbacks

Test Fallbacks!

Regular Fallbacks

Content Policy Fallbacks

Context Window Fallbacks

EU-Region Filtering (Pre-Call Checks)

Custom Timeouts, Stream Timeouts - Per Model

Start Proxy

Setting Dynamic Timeouts - Per Request