about implement of prompt

[CodestChan]

optimizer = optimizer_class(
list(brushnet.parameters()) + list(text_encoder.get_input_embeddings().parameters()),
lr=args.learning_rate,
betas=(args.adam_beta1, args.adam_beta2),
weight_decay=args.adam_weight_decay,
eps=args.adam_epsilon,
)

predict the noise residual

if task_class is not None:
noise_pred = self.unet(
sample=latent_model_input,
timestep=t,
encoder_hidden_states=prompt_embeds,
cross_attention_kwargs=cross_attention_kwargs,
return_dict=False,
task_class=task_class,
)[0]

Did the project re-implement get_input_embeddings and unet? Why can't I find the implementation of some ideas in the paper?