Fix small inconsistency in output dimension of "_get_t5_prompt_embeds" function in sd3 pipeline (huggingface#12531)

* Fix small inconsistency in output dimension of t5 embeds when text_encoder_3 is None

* first commit

---------

Co-authored-by: Alireza Farashah <[email protected]>
Co-authored-by: Alireza Farashah <[email protected]>

Author: 3 people

src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet.py

@@ -266,7 +266,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,

src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py

@@ -284,7 +284,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,

src/diffusers/pipelines/pag/pipeline_pag_sd_3.py

@@ -237,7 +237,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,

src/diffusers/pipelines/pag/pipeline_pag_sd_3_img2img.py

@@ -253,7 +253,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -248,7 +248,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

@@ -272,7 +272,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py

@@ -278,7 +278,7 @@ def _get_t5_prompt_embeds(
             return torch.zeros(
                 (
                     batch_size * num_images_per_prompt,
-                    self.tokenizer_max_length,
+                    max_sequence_length,
                     self.transformer.config.joint_attention_dim,
                 ),
                 device=device,