AlexsanderHamir

import os
import uuid
from locust import HttpUser, task, between, events

# Custom metric to track LiteLLM overhead duration
overhead_durations = []

@events.request.add_listener
def on_request(**kwargs):
    response = kwargs.get('response')

AlexsanderHamir

model_list:
  - model_name: db-openai-endpoint
    litellm_params:
      model: openai/*
      api_base: https://exampleopenaiendpoint-production-0ee2.up.railway.app/

AlexsanderHamir

Local Load Test: 10 Concurrent Requests

Summary

Load testing proxy with 100 total requests
Concurrency level: 10
Client config: Allows unlimited reuse (tests server behavior)

AlexsanderHamir

Iteration 1

Prompt: Say 'hello' in Spanish
Response: Hola
Timing: Client 1581.63ms | LiteLLM Overhead 5.39ms (0.34%) | Provider 1576.23ms
Tokens: 36 prompt + 3 completion = 39

Iteration 2

Prompt: What is 2+2?
Response: 4
Timing: Client 624.66ms | LiteLLM Overhead 6.49ms (1.04%) | Provider 618.17ms

AlexsanderHamir

for i in {1..10}; do                                             
  echo "Request #$i"
  response=$(curl -s -L -X POST 'http://0.0.0.0:4000/v1/chat/completions' \
    -H 'Content-Type: application/json' \
    -H 'Authorization: Bearer sk-...' \
    -H 'x-litellm-num-retries: 0' \
    -D /dev/stderr \
    -d '{
      "model": "gemini-flash-lite",
      "messages": [

AlexsanderHamir

#!/bin/bash

# Runs 10 times, every 5 seconds, and saves to memory_log.json
echo "[" > memory_log.json
for i in {1..10}; do
  echo "Run $i..."
  data=$(curl -s "http://localhost:4000/debug/memory/details" \
    -H "Authorization: Bearer sk-1234")

  if [ $i -gt 1 ]; then echo "," >> memory_log.json; fi

AlexsanderHamir

Hi all,

Sharing our public roadmap on LiteLLM performance overheads:

As of v1.78.5, the LiteLLM AI Gateway adds a 8 ms median overhead and 45 ms P99 overhead at 1 K concurrent requests with 4 LiteLLM instances.

This is an ~80% improvement over v1.76.0. This roadmap has 3 key components we plan on achieving by end of 2025:

1. Achieve 8 ms median across all major LLM endpoints.
2. Resolve open memory leak issues.

AlexsanderHamir

model_list:
  ### RESPONSES
  - model_name: gpt-5-codex
    litellm_params:
      model: openai/*
      api_base: https://exampleopenaiendpoint-production-0ee2.up.railway.app/

  ### EMBEDDINGS
  - model_name: text-embedding-3-large
    litellm_params:

AlexsanderHamir

import os
import uuid
from locust import HttpUser, task, between, events

# Custom metric to track LiteLLM overhead duration
overhead_durations = []

@events.request.add_listener
def on_request(**kwargs):
    response = kwargs.get('response')

AlexsanderHamir

import json
import time
import uuid

from locust import HttpUser, between, events, task
from websocket import (
    WebSocketBadStatusException,
    WebSocketConnectionClosedException,
    create_connection,
)