Back to projects

Project details

Slim vLLM Benchmark Client

A local-only benchmark client for sending controlled text or image request profiles to OpenAI-compatible model services and collecting latency metrics.

Category
Data / benchmark
Status
Active
Visibility
Local only
Role
Personal project

Technology stack

  • Python
  • vLLM
  • OpenAI API

Prompt datasets, tokenizer settings, concurrency, sampling options, and optional image inputs enter an asynchronous request client. It produces success counts, throughput, time to first token, total latency, and grouped error results.

The client supports repeatable gateway checks, migration comparisons, and service load tests without carrying an inference engine. It remains an active local-only project with mock-service coverage for request shape, scheduling, and metrics.