Extend Support for callback_on_step_end for AuraFlow and LuminaText2Img Pipelines

src/diffusers/pipelines/aura_flow/pipeline_aura_flow.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import inspect
-from typing import List, Optional, Tuple, Union
+from typing import List, Optional, Tuple, Union, Dict, Callable
 
 import torch
 from transformers import T5Tokenizer, UMT5EncoderModel
@@ -24,6 +24,7 @@
 from ...utils import is_torch_xla_available, logging, replace_example_docstring
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 
 
 if is_torch_xla_available():
@@ -131,6 +132,12 @@ class AuraFlowPipeline(DiffusionPipeline):
 
     _optional_components = []
     model_cpu_offload_seq = "text_encoder->transformer->vae"
+    _callback_tensor_inputs = [
+        "latents",
+        "prompt_embeds",
+        "add_text_embeds",
+        "add_time_ids",
+    ]
 
     def __init__(
         self,
@@ -159,12 +166,19 @@ def check_inputs(
         negative_prompt_embeds=None,
         prompt_attention_mask=None,
         negative_prompt_attention_mask=None,
+        callback_on_step_end_tensor_inputs=None,
     ):
         if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
             raise ValueError(
                 f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}."
             )
-
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
         if prompt is not None and prompt_embeds is not None:
             raise ValueError(
                 f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
@@ -408,6 +422,8 @@ def __call__(
         max_sequence_length: int = 256,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
     ) -> Union[ImagePipelineOutput, Tuple]:
         r"""
         Function invoked when calling the pipeline for generation.
@@ -462,6 +478,15 @@ def __call__(
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
                 of a plain tuple.
+            callback_on_step_end (`Callable`, *optional*):
+                A function that calls at the end of each denoising steps during the inference. The function is called
+                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+                `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
             max_sequence_length (`int` defaults to 256): Maximum sequence length to use with the `prompt`.
 
         Examples:
@@ -483,6 +508,7 @@ def __call__(
             negative_prompt_embeds,
             prompt_attention_mask,
             negative_prompt_attention_mask,
+            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
         )
 
         # 2. Determine batch size.
@@ -567,6 +593,15 @@ def __call__(
                 # compute the previous noisy sample x_t -> x_t-1
                 latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()

src/diffusers/pipelines/lumina/pipeline_lumina.py

@@ -17,7 +17,7 @@
 import math
 import re
 import urllib.parse as ul
-from typing import List, Optional, Tuple, Union
+from typing import List, Optional, Tuple, Union, Dict, Callable
 
 import torch
 from transformers import AutoModel, AutoTokenizer
@@ -37,6 +37,7 @@
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 
 
 if is_torch_xla_available():
@@ -174,6 +175,13 @@ class LuminaText2ImgPipeline(DiffusionPipeline):
 
     _optional_components = []
     model_cpu_offload_seq = "text_encoder->transformer->vae"
+    _callback_tensor_inputs = [
+        "latents",
+        "prompt_embeds",
+        "add_text_embeds",
+        "add_time_ids",
+    ]
+
 
     def __init__(
         self,
@@ -395,11 +403,19 @@ def check_inputs(
         negative_prompt_embeds=None,
         prompt_attention_mask=None,
         negative_prompt_attention_mask=None,
+        callback_on_step_end_tensor_inputs=None,
     ):
         if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
             raise ValueError(
                 f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}."
             )
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
 
         if prompt is not None and prompt_embeds is not None:
             raise ValueError(
@@ -644,6 +660,8 @@ def __call__(
         max_sequence_length: int = 256,
         scaling_watershed: Optional[float] = 1.0,
         proportional_attn: Optional[bool] = True,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
     ) -> Union[ImagePipelineOutput, Tuple]:
         """
         Function invoked when calling the pipeline for generation.
@@ -734,6 +752,7 @@ def __call__(
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=negative_prompt_embeds,
             prompt_attention_mask=prompt_attention_mask,
+            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
             negative_prompt_attention_mask=negative_prompt_attention_mask,
         )
         cross_attention_kwargs = {}
@@ -886,6 +905,15 @@ def __call__(
 
                 progress_bar.update()
 
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
                 if XLA_AVAILABLE:
                     xm.mark_step()