SD3.5-Controlnet-8b does not use encoder_hidden_states

In https://github.com/huggingface/diffusers/blob/main/examples/controlnet/train_controlnet_sd3.py#L1338
the encoder_hidden_states are set to prompt_embeds, which is incorrect if we are finetuning sd3.5-controlnet-8b since it does not use encoder_hidden_states as stated in https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/controlnets/controlnet_sd3.py#L406

Start at examples/controlnet/train_controlnet_sd3.py around line 1338, then read the SD3 ControlNet implementation at src/diffusers/models/controlnets/controlnet_sd3.py around line 406. Compare the training call's inputs with the model's documented behavior for SD3.5-Controlnet-8b. Done means the training entry point no longer relies on an encoder_hidden_states input that this model does not use.

Scritto dal modello di indicizzazione a partire dal testo della issue.