Various Bugfixes + Spaces Demo

src/diffusers/models/transformers/transformer_lumina2.py

@@ -448,4 +448,4 @@ def forward(
         if not return_dict:
             return (output,)
 
-        return Transformer2DModelOutput(sample=output)
+        return Transformer2DModelOutput(sample=output)

src/diffusers/pipelines/lumina2/pipeline_lumina2.py

@@ -208,7 +208,6 @@ def __init__(
         )
         self.vae_scale_factor = 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
-        self.max_sequence_length = 256
         self.default_sample_size = (
             self.transformer.config.sample_size
             if hasattr(self, "transformer") and self.transformer is not None
@@ -222,7 +221,7 @@ def _get_gemma_prompt_embeds(
         prompt: Union[str, List[str]],
         num_images_per_prompt: int = 1,
         device: Optional[torch.device] = None,
-        max_length: Optional[int] = None,
+        max_length: int = 256,
     ):
         device = device or self._execution_device
         prompt = [prompt] if isinstance(prompt, str) else prompt
@@ -231,7 +230,7 @@ def _get_gemma_prompt_embeds(
         text_inputs = self.tokenizer(
             prompt,
             pad_to_multiple_of=8,
-            max_length=self.max_sequence_length,
+            max_length=max_length,
             truncation=True,
             padding=True,
             return_tensors="pt",
@@ -240,10 +239,10 @@ def _get_gemma_prompt_embeds(
         untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids.to(device)
 
         if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
-            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.max_sequence_length - 1 : -1])
+            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, max_length - 1 : -1])
             logger.warning(
                 "The following part of your input was truncated because Gemma can only handle sequences up to"
-                f" {self.max_sequence_length} tokens: {removed_text}"
+                f" {max_length} tokens: {removed_text}"
             )
 
         prompt_attention_mask = text_inputs.attention_mask.to(device)
@@ -283,6 +282,7 @@ def encode_prompt(
         prompt_attention_mask: Optional[torch.Tensor] = None,
         negative_prompt_attention_mask: Optional[torch.Tensor] = None,
         system_prompt: Optional[str] = None,
+        max_sequence_length: int = 256,
         **kwargs,
     ):
         r"""
@@ -326,6 +326,7 @@ def encode_prompt(
                 prompt=prompt,
                 num_images_per_prompt=num_images_per_prompt,
                 device=device,
+                max_length=max_sequence_length,
             )
 
         # Get negative embeddings for classifier free guidance
@@ -791,8 +792,6 @@ def __call__(
                         # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
                         latents = latents.to(latents_dtype)
 
-                progress_bar.update()
-
                 if callback_on_step_end is not None:
                     callback_kwargs = {}
                     for k in callback_on_step_end_tensor_inputs:
@@ -822,4 +821,4 @@ def __call__(
         if not return_dict:
             return (image,)
 
-        return ImagePipelineOutput(images=image)
+        return ImagePipelineOutput(images=image)