GaviZhou/qwen3-vl-8b-hallu

GaviZhou/qwen3-vl-8b-hallu

Current evaluation candidate: benchmark-update-step120_9-8.

This repository contains only inference artifacts exported from checkpoint-120. It excludes optimizer, scheduler, DeepSpeed, RNG, trainer, and log files.

Evaluation

Use the exact commit SHA printed by the publish script, rather than a moving main revision. Record that SHA with the evaluator and benchmark versions.