I want YOLOWorldDetector to support prompt caching with reparameterize(texts: List List str ) - None …

I want YOLOWorldDetector to support prompt caching with reparameterize(texts: List List str ) - None …: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). After that, model.predict(batch inputs, batch data samples, rescale=True) should work even when the provided data…

The task

After that, `model.predict(batch_inputs, batch_data_samples, rescale=True)` should work even when the provided data samples do not carry a `texts` attribute or `texts` key. In that cached session, prediction should run the image-only backbone path, reuse the stored text feature tensor as the class prompts, pass those…

Part of FineEnvs/MiMo-V2.6-RL-harbor-code.