JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Chinese implementation and benchmark guide

Qwen Choice

A Chinese-language guide to classifying one image with a local Qwen model by restricted option logits, with separate model-stage and full-request timings on a fixed input.

Original by RJMSWDEvaluationGitHub repositorySource reviewed

Before you dive in

What you’ll find in the original

  1. Read allowed option logits directly rather than generating and parsing a textual answer.
  2. Report end-to-end request latency separately from the GPU model stage.
  3. Fix hardware, precision, batch size, image dimensions, visual-token count, and total input tokens in the benchmark record.
Worth knowing

The author reports a 168.694 ms median and 173.989 ms p95 for one RTX A6000, BF16, batch one, and one 1280×720 image. These are local Qwen measurements, not hosted Jev results.