add support for pre-calculated prompt embeds to Stable Diffusion ONNX pipelines (huggingface#2597)

* add support for prompt embeds to SD ONNX pipeline

* fix up the pipeline copies

* add prompt embeds param to other ONNX pipelines

* fix up prompt embeds param for SD upscaling ONNX pipeline

* add missing type annotations to ONNX pipes

Author: ssube

pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py

@@ -111,7 +111,15 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    def _encode_prompt(self, prompt, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
+    def _encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        num_images_per_prompt: Optional[int],
+        do_classifier_free_guidance: bool,
+        negative_prompt: Optional[str],
+        prompt_embeds: Optional[np.ndarray] = None,
+        negative_prompt_embeds: Optional[np.ndarray] = None,
+    ):
         r"""
         Encodes the prompt into text encoder hidden states.
 
@@ -125,32 +133,48 @@ def _encode_prompt(self, prompt, num_images_per_prompt, do_classifier_free_guida
             negative_prompt (`str` or `List[str]`):
                 The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
                 if `guidance_scale` is less than `1`).
+            prompt_embeds (`np.ndarray`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`np.ndarray`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
         """
-        batch_size = len(prompt) if isinstance(prompt, list) else 1
-
-        # get prompt text embeddings
-        text_inputs = self.tokenizer(
-            prompt,
-            padding="max_length",
-            max_length=self.tokenizer.model_max_length,
-            truncation=True,
-            return_tensors="np",
-        )
-        text_input_ids = text_inputs.input_ids
-        untruncated_ids = self.tokenizer(prompt, padding="max_length", return_tensors="np").input_ids
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
 
-        if not np.array_equal(text_input_ids, untruncated_ids):
-            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
-            logger.warning(
-                "The following part of your input was truncated because CLIP can only handle sequences up to"
-                f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+        if prompt_embeds is None:
+            # get prompt text embeddings
+            text_inputs = self.tokenizer(
+                prompt,
+                padding="max_length",
+                max_length=self.tokenizer.model_max_length,
+                truncation=True,
+                return_tensors="np",
             )
+            text_input_ids = text_inputs.input_ids
+            untruncated_ids = self.tokenizer(prompt, padding="max_length", return_tensors="np").input_ids
+
+            if not np.array_equal(text_input_ids, untruncated_ids):
+                removed_text = self.tokenizer.batch_decode(
+                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                )
+                logger.warning(
+                    "The following part of your input was truncated because CLIP can only handle sequences up to"
+                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                )
+
+            prompt_embeds = self.text_encoder(input_ids=text_input_ids.astype(np.int32))[0]
 
-        prompt_embeds = self.text_encoder(input_ids=text_input_ids.astype(np.int32))[0]
         prompt_embeds = np.repeat(prompt_embeds, num_images_per_prompt, axis=0)
 
         # get unconditional embeddings for classifier free guidance
-        if do_classifier_free_guidance:
+        if do_classifier_free_guidance and negative_prompt_embeds is None:
             uncond_tokens: List[str]
             if negative_prompt is None:
                 uncond_tokens = [""] * batch_size
@@ -170,7 +194,7 @@ def _encode_prompt(self, prompt, num_images_per_prompt, do_classifier_free_guida
             else:
                 uncond_tokens = negative_prompt
 
-            max_length = text_input_ids.shape[-1]
+            max_length = prompt_embeds.shape[1]
             uncond_input = self.tokenizer(
                 uncond_tokens,
                 padding="max_length",
@@ -179,6 +203,8 @@ def _encode_prompt(self, prompt, num_images_per_prompt, do_classifier_free_guida
                 return_tensors="np",
             )
             negative_prompt_embeds = self.text_encoder(input_ids=uncond_input.input_ids.astype(np.int32))[0]
+
+        if do_classifier_free_guidance:
             negative_prompt_embeds = np.repeat(negative_prompt_embeds, num_images_per_prompt, axis=0)
 
             # For classifier free guidance, we need to do two forward passes.
@@ -188,9 +214,56 @@ def _encode_prompt(self, prompt, num_images_per_prompt, do_classifier_free_guida
 
         return prompt_embeds
 
-    def __call__(
+    def check_inputs(
         self,
         prompt: Union[str, List[str]],
+        height: Optional[int],
+        width: Optional[int],
+        callback_steps: int,
+        negative_prompt: Optional[str] = None,
+        prompt_embeds: Optional[np.ndarray] = None,
+        negative_prompt_embeds: Optional[np.ndarray] = None,
+    ):
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+        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"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and negative_prompt_embeds is not None:
+            if prompt_embeds.shape != negative_prompt_embeds.shape:
+                raise ValueError(
+                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+                    f" {negative_prompt_embeds.shape}."
+                )
+
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
         height: Optional[int] = 512,
         width: Optional[int] = 512,
         num_inference_steps: Optional[int] = 50,
@@ -200,28 +273,86 @@ def __call__(
         eta: Optional[float] = 0.0,
         generator: Optional[np.random.RandomState] = None,
         latents: Optional[np.ndarray] = None,
+        prompt_embeds: Optional[np.ndarray] = None,
+        negative_prompt_embeds: Optional[np.ndarray] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         callback: Optional[Callable[[int, int, np.ndarray], None]] = None,
         callback_steps: int = 1,
     ):
-        if isinstance(prompt, str):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            image (`PIL.Image.Image` or List[`PIL.Image.Image`] or `torch.FloatTensor`):
+                `Image`, or tensor representing an image batch which will be upscaled. *
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds`. instead. Ignored when not using guidance (i.e., ignored if `guidance_scale`
+                is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`np.random.RandomState`, *optional*):
+                One or a list of [numpy generator(s)](TODO) to make generation deterministic.
+            latents (`np.ndarray`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            prompt_embeds (`np.ndarray`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`np.ndarray`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
+            When returning a tuple, the first element is a list with the generated images, and the second element is a
+            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, according to the `safety_checker`.
+        """
+
+        # check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds
+        )
+
+        # define call parameters
+        if prompt is not None and isinstance(prompt, str):
             batch_size = 1
-        elif isinstance(prompt, list):
+        elif prompt is not None and isinstance(prompt, list):
             batch_size = len(prompt)
         else:
-            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
-
-        if height % 8 != 0 or width % 8 != 0:
-            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
-
-        if (callback_steps is None) or (
-            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
-        ):
-            raise ValueError(
-                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
-                f" {type(callback_steps)}."
-            )
+            batch_size = prompt_embeds.shape[0]
 
         if generator is None:
             generator = np.random
@@ -232,7 +363,12 @@ def __call__(
         do_classifier_free_guidance = guidance_scale > 1.0
 
         prompt_embeds = self._encode_prompt(
-            prompt, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
+            prompt,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
         )
 
         # get the initial random noise unless the user supplied it