INNER CODE UNIT · Python

repeat

jdh-algo/JoyHallo · scripts/app.py:148

                repeat=ref_image_latents.size(0) // ref_timesteps.size(0),
            )
            self.reference_unet(
                ref_image_latents,
                ref_timesteps,
                encoder_hidden_states=face_emb,
                return_dict=False,
            )
            self.reference_control_reader.update(self.reference_control_writer)

        if uncond_audio_fwd:
            audio_emb = torch.zeros_like(audio_emb).to(
                device=audio_emb.device, dtype=audio_emb.dtype
            )

        model_pred = self.denoising_unet(
            noisy_latents,
            timesteps,

View source record →