ttrv-test-time-rl-vision-language

Adapt vision-language models at test time without labels by extracting implicit reward signals (prediction frequency and entropy) and optimizing via GRPO.

Independent GitHub snapshot · Automated classification · Original documentation governs