InvokeAI/invokeai/app/invocations/onnx.py

# Copyright (c) 2023 Borisov Sergey (https://github.com/StAlKeR7779)

import inspect

# from contextlib import ExitStack
from typing import List, Literal, Union

import numpy as np
import torch
from diffusers.image_processor import VaeImageProcessor
from pydantic import BaseModel, ConfigDict, Field, field_validator
from tqdm import tqdm

from invokeai.app.invocations.primitives import ConditioningField, ConditioningOutput, ImageField, ImageOutput
from invokeai.app.services.image_records.image_records_common import ImageCategory, ResourceOrigin
from invokeai.app.shared.fields import FieldDescriptions
from invokeai.app.util.step_callback import stable_diffusion_step_callback
from invokeai.backend import BaseModelType, ModelType, SubModelType

from ...backend.model_management import ONNXModelPatcher
from ...backend.stable_diffusion import PipelineIntermediateState
from ...backend.util import choose_torch_device
from ..util.ti_utils import extract_ti_triggers_from_prompt
from .baseinvocation import (
    BaseInvocation,
    BaseInvocationOutput,
    Input,
    InputField,
    InvocationContext,
    OutputField,
    UIComponent,
    UIType,
    WithMetadata,
    invocation,
    invocation_output,
)
from .controlnet_image_processors import ControlField
from .latent import SAMPLER_NAME_VALUES, LatentsField, LatentsOutput, build_latents_output, get_scheduler
from .model import ClipField, ModelInfo, UNetField, VaeField

ORT_TO_NP_TYPE = {
    "tensor(bool)": np.bool_,
    "tensor(int8)": np.int8,
    "tensor(uint8)": np.uint8,
    "tensor(int16)": np.int16,
    "tensor(uint16)": np.uint16,
    "tensor(int32)": np.int32,
    "tensor(uint32)": np.uint32,
    "tensor(int64)": np.int64,
    "tensor(uint64)": np.uint64,
    "tensor(float16)": np.float16,
    "tensor(float)": np.float32,
    "tensor(double)": np.float64,
}

PRECISION_VALUES = Literal[tuple(ORT_TO_NP_TYPE.keys())]


@invocation("prompt_onnx", title="ONNX Prompt (Raw)", tags=["prompt", "onnx"], category="conditioning", version="1.0.0")
class ONNXPromptInvocation(BaseInvocation):
    prompt: str = InputField(default="", description=FieldDescriptions.raw_prompt, ui_component=UIComponent.Textarea)
    clip: ClipField = InputField(description=FieldDescriptions.clip, input=Input.Connection)

    def invoke(self, context: InvocationContext) -> ConditioningOutput:
        tokenizer_info = context.services.model_manager.get_model(
            **self.clip.tokenizer.model_dump(),
        )
        text_encoder_info = context.services.model_manager.get_model(
            **self.clip.text_encoder.model_dump(),
        )
        with tokenizer_info as orig_tokenizer, text_encoder_info as text_encoder:  # , ExitStack() as stack:
            loras = [
                (
                    context.services.model_manager.get_model(**lora.model_dump(exclude={"weight"})).context.model,
                    lora.weight,
                )
                for lora in self.clip.loras
            ]

            ti_list = []
            for trigger in extract_ti_triggers_from_prompt(self.prompt):
                name = trigger[1:-1]
                try:
                    ti_list.append(
                        (
                            name,
                            context.services.model_manager.get_model(
                                model_name=name,
                                base_model=self.clip.text_encoder.base_model,
                                model_type=ModelType.TextualInversion,
                            ).context.model,
                        )
                    )
                except Exception:
                    # print(e)
                    # import traceback
                    # print(traceback.format_exc())
                    print(f'Warn: trigger: "{trigger}" not found')
            if loras or ti_list:
                text_encoder.release_session()
            with (
                ONNXModelPatcher.apply_lora_text_encoder(text_encoder, loras),
                ONNXModelPatcher.apply_ti(orig_tokenizer, text_encoder, ti_list) as (tokenizer, ti_manager),
            ):
                text_encoder.create_session()

                # copy from
                # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L153
                text_inputs = tokenizer(
                    self.prompt,
                    padding="max_length",
                    max_length=tokenizer.model_max_length,
                    truncation=True,
                    return_tensors="np",
                )
                text_input_ids = text_inputs.input_ids
                """
                untruncated_ids = 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 = text_encoder(input_ids=text_input_ids.astype(np.int32))[0]

        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"

        # TODO: hacky but works ;D maybe rename latents somehow?
        context.services.latents.save(conditioning_name, (prompt_embeds, None))

        return ConditioningOutput(
            conditioning=ConditioningField(
                conditioning_name=conditioning_name,
            ),
        )


# Text to image
@invocation(
    "t2l_onnx",
    title="ONNX Text to Latents",
    tags=["latents", "inference", "txt2img", "onnx"],
    category="latents",
    version="1.0.0",
)
class ONNXTextToLatentsInvocation(BaseInvocation):
    """Generates latents from conditionings."""

    positive_conditioning: ConditioningField = InputField(
        description=FieldDescriptions.positive_cond,
        input=Input.Connection,
    )
    negative_conditioning: ConditioningField = InputField(
        description=FieldDescriptions.negative_cond,
        input=Input.Connection,
    )
    noise: LatentsField = InputField(
        description=FieldDescriptions.noise,
        input=Input.Connection,
    )
    steps: int = InputField(default=10, gt=0, description=FieldDescriptions.steps)
    cfg_scale: Union[float, List[float]] = InputField(
        default=7.5,
        ge=1,
        description=FieldDescriptions.cfg_scale,
    )
    scheduler: SAMPLER_NAME_VALUES = InputField(
        default="euler", description=FieldDescriptions.scheduler, input=Input.Direct, ui_type=UIType.Scheduler
    )
    precision: PRECISION_VALUES = InputField(default="tensor(float16)", description=FieldDescriptions.precision)
    unet: UNetField = InputField(
        description=FieldDescriptions.unet,
        input=Input.Connection,
    )
    control: Union[ControlField, list[ControlField]] = InputField(
        default=None,
        description=FieldDescriptions.control,
    )
    # seamless:   bool = InputField(default=False, description="Whether or not to generate an image that can tile without seams", )
    # seamless_axes: str = InputField(default="", description="The axes to tile the image on, 'x' and/or 'y'")

    @field_validator("cfg_scale")
    def ge_one(cls, v):
        """validate that all cfg_scale values are >= 1"""
        if isinstance(v, list):
            for i in v:
                if i < 1:
                    raise ValueError("cfg_scale must be greater than 1")
        else:
            if v < 1:
                raise ValueError("cfg_scale must be greater than 1")
        return v

    # based on
    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
    def invoke(self, context: InvocationContext) -> LatentsOutput:
        c, _ = context.services.latents.get(self.positive_conditioning.conditioning_name)
        uc, _ = context.services.latents.get(self.negative_conditioning.conditioning_name)
        graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
        source_node_id = graph_execution_state.prepared_source_mapping[self.id]
        if isinstance(c, torch.Tensor):
            c = c.cpu().numpy()
        if isinstance(uc, torch.Tensor):
            uc = uc.cpu().numpy()
        device = torch.device(choose_torch_device())
        prompt_embeds = np.concatenate([uc, c])

        latents = context.services.latents.get(self.noise.latents_name)
        if isinstance(latents, torch.Tensor):
            latents = latents.cpu().numpy()

        # TODO: better execution device handling
        latents = latents.astype(ORT_TO_NP_TYPE[self.precision])

        # get the initial random noise unless the user supplied it
        do_classifier_free_guidance = True
        # latents_dtype = prompt_embeds.dtype
        # latents_shape = (batch_size * num_images_per_prompt, 4, height // 8, width // 8)
        # if latents.shape != latents_shape:
        #    raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")

        scheduler = get_scheduler(
            context=context,
            scheduler_info=self.unet.scheduler,
            scheduler_name=self.scheduler,
            seed=0,  # TODO: refactor this node
        )

        def torch2numpy(latent: torch.Tensor):
            return latent.cpu().numpy()

        def numpy2torch(latent, device):
            return torch.from_numpy(latent).to(device)

        def dispatch_progress(
            self, context: InvocationContext, source_node_id: str, intermediate_state: PipelineIntermediateState
        ) -> None:
            stable_diffusion_step_callback(
                context=context,
                intermediate_state=intermediate_state,
                node=self.model_dump(),
                source_node_id=source_node_id,
            )

        scheduler.set_timesteps(self.steps)
        latents = latents * np.float64(scheduler.init_noise_sigma)

        extra_step_kwargs = {}
        if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
            extra_step_kwargs.update(
                eta=0.0,
            )

        unet_info = context.services.model_manager.get_model(**self.unet.unet.model_dump())

        with unet_info as unet:  # , ExitStack() as stack:
            # loras = [(stack.enter_context(context.services.model_manager.get_model(**lora.dict(exclude={"weight"}))), lora.weight) for lora in self.unet.loras]
            loras = [
                (
                    context.services.model_manager.get_model(**lora.model_dump(exclude={"weight"})).context.model,
                    lora.weight,
                )
                for lora in self.unet.loras
            ]

            if loras:
                unet.release_session()
            with ONNXModelPatcher.apply_lora_unet(unet, loras):
                # TODO:
                _, _, h, w = latents.shape
                unet.create_session(h, w)

                timestep_dtype = next(
                    (input.type for input in unet.session.get_inputs() if input.name == "timestep"), "tensor(float16)"
                )
                timestep_dtype = ORT_TO_NP_TYPE[timestep_dtype]
                for i in tqdm(range(len(scheduler.timesteps))):
                    t = scheduler.timesteps[i]
                    # expand the latents if we are doing classifier free guidance
                    latent_model_input = np.concatenate([latents] * 2) if do_classifier_free_guidance else latents
                    latent_model_input = scheduler.scale_model_input(numpy2torch(latent_model_input, device), t)
                    latent_model_input = latent_model_input.cpu().numpy()

                    # predict the noise residual
                    timestep = np.array([t], dtype=timestep_dtype)
                    noise_pred = unet(sample=latent_model_input, timestep=timestep, encoder_hidden_states=prompt_embeds)
                    noise_pred = noise_pred[0]

                    # perform guidance
                    if do_classifier_free_guidance:
                        noise_pred_uncond, noise_pred_text = np.split(noise_pred, 2)
                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)

                    # compute the previous noisy sample x_t -> x_t-1
                    scheduler_output = scheduler.step(
                        numpy2torch(noise_pred, device), t, numpy2torch(latents, device), **extra_step_kwargs
                    )
                    latents = torch2numpy(scheduler_output.prev_sample)

                    state = PipelineIntermediateState(
                        run_id="test", step=i, timestep=timestep, latents=scheduler_output.prev_sample
                    )
                    dispatch_progress(self, context=context, source_node_id=source_node_id, intermediate_state=state)

                    # call the callback, if provided
                    # if callback is not None and i % callback_steps == 0:
                    #    callback(i, t, latents)

        torch.cuda.empty_cache()

        name = f"{context.graph_execution_state_id}__{self.id}"
        context.services.latents.save(name, latents)
        return build_latents_output(latents_name=name, latents=torch.from_numpy(latents))


# Latent to image
@invocation(
    "l2i_onnx",
    title="ONNX Latents to Image",
    tags=["latents", "image", "vae", "onnx"],
    category="image",
    version="1.2.0",
)
class ONNXLatentsToImageInvocation(BaseInvocation, WithMetadata):
    """Generates an image from latents."""

    latents: LatentsField = InputField(
        description=FieldDescriptions.denoised_latents,
        input=Input.Connection,
    )
    vae: VaeField = InputField(
        description=FieldDescriptions.vae,
        input=Input.Connection,
    )
    # tiled: bool = InputField(default=False, description="Decode latents by overlaping tiles(less memory consumption)")

    def invoke(self, context: InvocationContext) -> ImageOutput:
        latents = context.services.latents.get(self.latents.latents_name)

        if self.vae.vae.submodel != SubModelType.VaeDecoder:
            raise Exception(f"Expected vae_decoder, found: {self.vae.vae.model_type}")

        vae_info = context.services.model_manager.get_model(
            **self.vae.vae.model_dump(),
        )

        # clear memory as vae decode can request a lot
        torch.cuda.empty_cache()

        with vae_info as vae:
            vae.create_session()

            # copied from
            # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L427
            latents = 1 / 0.18215 * latents
            # image = self.vae_decoder(latent_sample=latents)[0]
            # it seems likes there is a strange result for using half-precision vae decoder if batchsize>1
            image = np.concatenate([vae(latent_sample=latents[i : i + 1])[0] for i in range(latents.shape[0])])

            image = np.clip(image / 2 + 0.5, 0, 1)
            image = image.transpose((0, 2, 3, 1))
            image = VaeImageProcessor.numpy_to_pil(image)[0]

        torch.cuda.empty_cache()

        image_dto = context.services.images.create(
            image=image,
            image_origin=ResourceOrigin.INTERNAL,
            image_category=ImageCategory.GENERAL,
            node_id=self.id,
            session_id=context.graph_execution_state_id,
            is_intermediate=self.is_intermediate,
            metadata=self.metadata,
            workflow=context.workflow,
        )

        return ImageOutput(
            image=ImageField(image_name=image_dto.image_name),
            width=image_dto.width,
            height=image_dto.height,
        )


@invocation_output("model_loader_output_onnx")
class ONNXModelLoaderOutput(BaseInvocationOutput):
    """Model loader output"""

    unet: UNetField = OutputField(default=None, description=FieldDescriptions.unet, title="UNet")
    clip: ClipField = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
    vae_decoder: VaeField = OutputField(default=None, description=FieldDescriptions.vae, title="VAE Decoder")
    vae_encoder: VaeField = OutputField(default=None, description=FieldDescriptions.vae, title="VAE Encoder")


class OnnxModelField(BaseModel):
    """Onnx model field"""

    model_name: str = Field(description="Name of the model")
    base_model: BaseModelType = Field(description="Base model")
    model_type: ModelType = Field(description="Model Type")

    model_config = ConfigDict(protected_namespaces=())


@invocation("onnx_model_loader", title="ONNX Main Model", tags=["onnx", "model"], category="model", version="1.0.0")
class OnnxModelLoaderInvocation(BaseInvocation):
    """Loads a main model, outputting its submodels."""

    model: OnnxModelField = InputField(
        description=FieldDescriptions.onnx_main_model, input=Input.Direct, ui_type=UIType.ONNXModel
    )

    def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
        base_model = self.model.base_model
        model_name = self.model.model_name
        model_type = ModelType.ONNX

        # TODO: not found exceptions
        if not context.services.model_manager.model_exists(
            model_name=model_name,
            base_model=base_model,
            model_type=model_type,
        ):
            raise Exception(f"Unknown {base_model} {model_type} model: {model_name}")

        """
        if not context.services.model_manager.model_exists(
            model_name=self.model_name,
            model_type=SDModelType.Diffusers,
            submodel=SDModelType.Tokenizer,
        ):
            raise Exception(
                f"Failed to find tokenizer submodel in {self.model_name}! Check if model corrupted"
            )

        if not context.services.model_manager.model_exists(
            model_name=self.model_name,
            model_type=SDModelType.Diffusers,
            submodel=SDModelType.TextEncoder,
        ):
            raise Exception(
                f"Failed to find text_encoder submodel in {self.model_name}! Check if model corrupted"
            )

        if not context.services.model_manager.model_exists(
            model_name=self.model_name,
            model_type=SDModelType.Diffusers,
            submodel=SDModelType.UNet,
        ):
            raise Exception(
                f"Failed to find unet submodel from {self.model_name}! Check if model corrupted"
            )
        """

        return ONNXModelLoaderOutput(
            unet=UNetField(
                unet=ModelInfo(
                    model_name=model_name,
                    base_model=base_model,
                    model_type=model_type,
                    submodel=SubModelType.UNet,
                ),
                scheduler=ModelInfo(
                    model_name=model_name,
                    base_model=base_model,
                    model_type=model_type,
                    submodel=SubModelType.Scheduler,
                ),
                loras=[],
            ),
            clip=ClipField(
                tokenizer=ModelInfo(
                    model_name=model_name,
                    base_model=base_model,
                    model_type=model_type,
                    submodel=SubModelType.Tokenizer,
                ),
                text_encoder=ModelInfo(
                    model_name=model_name,
                    base_model=base_model,
                    model_type=model_type,
                    submodel=SubModelType.TextEncoder,
                ),
                loras=[],
                skipped_layers=0,
            ),
            vae_decoder=VaeField(
                vae=ModelInfo(
                    model_name=model_name,
                    base_model=base_model,
                    model_type=model_type,
                    submodel=SubModelType.VaeDecoder,
                ),
            ),
            vae_encoder=VaeField(
                vae=ModelInfo(
                    model_name=model_name,
                    base_model=base_model,
                    model_type=model_type,
                    submodel=SubModelType.VaeEncoder,
                ),
            ),
        )