Problems / LLM serving, medium

An LLM judge on every answer

Every answer is graded inline by an LLM judge with 20 slots before it returns. The judge is now the slowest part.

Solve it in the labTake it as a mock interview

Unlock it once for 15 credits; everything inside it is free (runs, changes, Submit, results, hints). Sign in with Google to start; new accounts get free credits.

Your goals

Success ≥ 99%, p99 ≤ 5 s.

Test cases

Like a coding problem, a solution has to pass every case, not just the one it was built for. Submit in the lab runs all 4; 1 of them are hidden until then.

  1. As given. The problem’s own traffic, scheduled failures and seed.
  2. A slow dependency. The main data store or model answers twice as slowly for part of the run.
  3. A server dies. One replica of the service with the most replicas is killed for part of the run.

The traffic

40 requests a second for 30 seconds: 100% request (reads).

The system

What each component does in this design, and what calls it. The simulator mimics each one from these settings.

ComponentIts job hereSettings
Load balancerLoad balancerTraffic enters here.Calls AI app.Layer 4 balancers route by IP and port; layer 7 balancers read HTTP and can route by path or header.–
AI appApp serviceLoad balancer forwards to it.Calls GPU serving pool, LLM judge.Little's law: throughput ≈ concurrency ÷ latency. 200 slots at 50 ms serve about 4,000 req/s.3 replicas × 500 slots · timeout 30 s
GPU serving poolSelf-hosted serving poolAI app calls it.GPU servers batch requests together; throughput rises with batch size until memory runs out.400 slots
LLM judgeLLM-as-judgeAI app calls it.An LLM judge grades outputs at scale, but it is another model call with its own latency, cost and biases.20 slots

More llm serving problems