Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(api): chore: pydantic & fastapi upgrade

Browse Source 
Upgrade pydantic and fastapi to latest.

- pydantic~=2.4.2
- fastapi~=103.2
- fastapi-events~=0.9.1

**Big Changes**

There are a number of logic changes needed to support pydantic v2. Most changes are very simple, like using the new methods to serialized and deserialize models, but there are a few more complex changes.

**Invocations**

The biggest change relates to invocation creation, instantiation and validation.

Because pydantic v2 moves all validation logic into the rust pydantic-core, we may no longer directly stick our fingers into the validation pie.

Previously, we (ab)used models and fields to allow invocation fields to be optional at instantiation, but required when `invoke()` is called. We directly manipulated the fields and invocation models when calling `invoke()`.

With pydantic v2, this is much more involved. Changes to the python wrapper do not propagate down to the rust validation logic - you have to rebuild the model. This causes problem with concurrent access to the invocation classes and is not a free operation.

This logic has been totally refactored and we do not need to change the model any more. The details are in `baseinvocation.py`, in the `InputField` function and `BaseInvocation.invoke_internal()` method.

In the end, this implementation is cleaner.

**Invocation Fields**

In pydantic v2, you can no longer directly add or remove fields from a model.

Previously, we did this to add the `type` field to invocations.

**Invocation Decorators**

With pydantic v2, we instead use the imperative `create_model()` API to create a new model with the additional field. This is done in `baseinvocation.py` in the `invocation()` wrapper.

A similar technique is used for `invocation_output()`.

**Minor Changes**

There are a number of minor changes around the pydantic v2 models API.

**Protected `model_` Namespace**

All models' pydantic-provided methods and attributes are prefixed with `model_` and this is considered a protected namespace. This causes some conflict, because "model" means something to us, and we have a ton of pydantic models with attributes starting with "model_".

Forunately, there are no direct conflicts. However, in any pydantic model where we define an attribute or method that starts with "model_", we must tell set the protected namespaces to an empty tuple.

```py
class IPAdapterModelField(BaseModel):
    model_name: str = Field(description="Name of the IP-Adapter model")
    base_model: BaseModelType = Field(description="Base model")

    model_config = ConfigDict(protected_namespaces=())
```

**Model Serialization**

Pydantic models no longer have `Model.dict()` or `Model.json()`.

Instead, we use `Model.model_dump()` or `Model.model_dump_json()`.

**Model Deserialization**

Pydantic models no longer have `Model.parse_obj()` or `Model.parse_raw()`, and there are no `parse_raw_as()` or `parse_obj_as()` functions.

Instead, you need to create a `TypeAdapter` object to parse python objects or JSON into a model.

```py
adapter_graph = TypeAdapter(Graph)
deserialized_graph_from_json = adapter_graph.validate_json(graph_json)
deserialized_graph_from_dict = adapter_graph.validate_python(graph_dict)
```

**Field Customisation**

Pydantic `Field`s no longer accept arbitrary args.

Now, you must put all additional arbitrary args in a `json_schema_extra` arg on the field.

**Schema Customisation**

FastAPI and pydantic schema generation now follows the OpenAPI version 3.1 spec.

This necessitates two changes:
- Our schema customization logic has been revised
- Schema parsing to build node templates has been revised

The specific aren't important, but this does present additional surface area for bugs.

**Performance Improvements**

Pydantic v2 is a full rewrite with a rust backend. This offers a substantial performance improvement (pydantic claims 5x to 50x depending on the task). We'll notice this the most during serialization and deserialization of sessions/graphs, which happens very very often - a couple times per node.

I haven't done any benchmarks, but anecdotally, graph execution is much faster. Also, very larges graphs - like with massive iterators - are much, much faster.
This commit is contained in:
psychedelicious
2023-09-24 18:11:07 +10:00
parent 19c5435332
commit c238a7f18b
74 changed files with 2788 additions and 3116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

149
invokeai/app/invocations/latent.py
View File

@ -19,7 +19,7 @@ from diffusers.models.attention_processor import (
)
from diffusers.schedulers import DPMSolverSDEScheduler
from diffusers.schedulers import SchedulerMixin as Scheduler
from pydantic import validator
from pydantic import field_validator
from torchvision.transforms.functional import resize as tv_resize
from invokeai.app.invocations.ip_adapter import IPAdapterField
@ -84,12 +84,20 @@ class SchedulerOutput(BaseInvocationOutput):
scheduler: SAMPLER_NAME_VALUES = OutputField(description=FieldDescriptions.scheduler, ui_type=UIType.Scheduler)
@invocation("scheduler", title="Scheduler", tags=["scheduler"], category="latents", version="1.0.0")
@invocation(
"scheduler",
title="Scheduler",
tags=["scheduler"],
category="latents",
version="1.0.0",
)
class SchedulerInvocation(BaseInvocation):
"""Selects a scheduler."""
scheduler: SAMPLER_NAME_VALUES = InputField(
default="euler", description=FieldDescriptions.scheduler, ui_type=UIType.Scheduler
default="euler",
description=FieldDescriptions.scheduler,
ui_type=UIType.Scheduler,
)
def invoke(self, context: InvocationContext) -> SchedulerOutput:
@ -97,7 +105,11 @@ class SchedulerInvocation(BaseInvocation):
@invocation(
"create_denoise_mask", title="Create Denoise Mask", tags=["mask", "denoise"], category="latents", version="1.0.0"
"create_denoise_mask",
title="Create Denoise Mask",
tags=["mask", "denoise"],
category="latents",
version="1.0.0",
)
class CreateDenoiseMaskInvocation(BaseInvocation):
"""Creates mask for denoising model run."""
@ -106,7 +118,11 @@ class CreateDenoiseMaskInvocation(BaseInvocation):
image: Optional[ImageField] = InputField(default=None, description="Image which will be masked", ui_order=1)
mask: ImageField = InputField(description="The mask to use when pasting", ui_order=2)
tiled: bool = InputField(default=False, description=FieldDescriptions.tiled, ui_order=3)
fp32: bool = InputField(default=DEFAULT_PRECISION == "float32", description=FieldDescriptions.fp32, ui_order=4)
fp32: bool = InputField(
default=DEFAULT_PRECISION == "float32",
description=FieldDescriptions.fp32,
ui_order=4,
)
def prep_mask_tensor(self, mask_image):
if mask_image.mode != "L":
@ -134,7 +150,7 @@ class CreateDenoiseMaskInvocation(BaseInvocation):
if image is not None:
vae_info = context.services.model_manager.get_model(
**self.vae.vae.dict(),
**self.vae.vae.model_dump(),
context=context,
)
@ -167,7 +183,7 @@ def get_scheduler(
) -> Scheduler:
scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP["ddim"])
orig_scheduler_info = context.services.model_manager.get_model(
**scheduler_info.dict(),
**scheduler_info.model_dump(),
context=context,
)
with orig_scheduler_info as orig_scheduler:
@ -209,34 +225,64 @@ class DenoiseLatentsInvocation(BaseInvocation):
negative_conditioning: ConditioningField = InputField(
description=FieldDescriptions.negative_cond, input=Input.Connection, ui_order=1
)
noise: Optional[LatentsField] = InputField(description=FieldDescriptions.noise, input=Input.Connection, ui_order=3)
noise: Optional[LatentsField] = InputField(
default=None,
description=FieldDescriptions.noise,
input=Input.Connection,
ui_order=3,
)
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, title="CFG Scale"
)
denoising_start: float = InputField(default=0.0, ge=0, le=1, description=FieldDescriptions.denoising_start)
denoising_start: float = InputField(
default=0.0,
ge=0,
le=1,
description=FieldDescriptions.denoising_start,
)
denoising_end: float = InputField(default=1.0, ge=0, le=1, description=FieldDescriptions.denoising_end)
scheduler: SAMPLER_NAME_VALUES = InputField(
default="euler", description=FieldDescriptions.scheduler, ui_type=UIType.Scheduler
default="euler",
description=FieldDescriptions.scheduler,
ui_type=UIType.Scheduler,
)
unet: UNetField = InputField(description=FieldDescriptions.unet, input=Input.Connection, title="UNet", ui_order=2)
control: Union[ControlField, list[ControlField]] = InputField(
unet: UNetField = InputField(
description=FieldDescriptions.unet,
input=Input.Connection,
title="UNet",
ui_order=2,
)
control: Optional[Union[ControlField, list[ControlField]]] = InputField(
default=None,
input=Input.Connection,
ui_order=5,
)
ip_adapter: Optional[Union[IPAdapterField, list[IPAdapterField]]] = InputField(
description=FieldDescriptions.ip_adapter, title="IP-Adapter", default=None, input=Input.Connection, ui_order=6
description=FieldDescriptions.ip_adapter,
title="IP-Adapter",
default=None,
input=Input.Connection,
ui_order=6,
)
t2i_adapter: Union[T2IAdapterField, list[T2IAdapterField]] = InputField(
description=FieldDescriptions.t2i_adapter, title="T2I-Adapter", default=None, input=Input.Connection, ui_order=7
t2i_adapter: Optional[Union[T2IAdapterField, list[T2IAdapterField]]] = InputField(
description=FieldDescriptions.t2i_adapter,
title="T2I-Adapter",
default=None,
input=Input.Connection,
ui_order=7,
)
latents: Optional[LatentsField] = InputField(
default=None, description=FieldDescriptions.latents, input=Input.Connection
)
latents: Optional[LatentsField] = InputField(description=FieldDescriptions.latents, input=Input.Connection)
denoise_mask: Optional[DenoiseMaskField] = InputField(
default=None, description=FieldDescriptions.mask, input=Input.Connection, ui_order=8
default=None,
description=FieldDescriptions.mask,
input=Input.Connection,
ui_order=8,
)
@validator("cfg_scale")
@field_validator("cfg_scale")
def ge_one(cls, v):
"""validate that all cfg_scale values are >= 1"""
if isinstance(v, list):
@ -259,7 +305,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
stable_diffusion_step_callback(
context=context,
intermediate_state=intermediate_state,
node=self.dict(),
node=self.model_dump(),
source_node_id=source_node_id,
base_model=base_model,
)
@ -451,9 +497,10 @@ class DenoiseLatentsInvocation(BaseInvocation):
# models are needed in memory. This would help to reduce peak memory utilization in low-memory environments.
with image_encoder_model_info as image_encoder_model:
# Get image embeddings from CLIP and ImageProjModel.
image_prompt_embeds, uncond_image_prompt_embeds = ip_adapter_model.get_image_embeds(
input_image, image_encoder_model
)
(
image_prompt_embeds,
uncond_image_prompt_embeds,
) = ip_adapter_model.get_image_embeds(input_image, image_encoder_model)
conditioning_data.ip_adapter_conditioning.append(
IPAdapterConditioningInfo(image_prompt_embeds, uncond_image_prompt_embeds)
)
@ -628,7 +675,10 @@ class DenoiseLatentsInvocation(BaseInvocation):
# TODO(ryand): I have hard-coded `do_classifier_free_guidance=True` to mirror the behaviour of ControlNets,
# below. Investigate whether this is appropriate.
t2i_adapter_data = self.run_t2i_adapters(
context, self.t2i_adapter, latents.shape, do_classifier_free_guidance=True
context,
self.t2i_adapter,
latents.shape,
do_classifier_free_guidance=True,
)
# Get the source node id (we are invoking the prepared node)
@ -641,7 +691,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
def _lora_loader():
for lora in self.unet.loras:
lora_info = context.services.model_manager.get_model(
**lora.dict(exclude={"weight"}),
**lora.model_dump(exclude={"weight"}),
context=context,
)
yield (lora_info.context.model, lora.weight)
@ -649,7 +699,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
return
unet_info = context.services.model_manager.get_model(
**self.unet.unet.dict(),
**self.unet.unet.model_dump(),
context=context,
)
with (
@ -700,7 +750,10 @@ class DenoiseLatentsInvocation(BaseInvocation):
denoising_end=self.denoising_end,
)
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
(
result_latents,
result_attention_map_saver,
) = pipeline.latents_from_embeddings(
latents=latents,
timesteps=timesteps,
init_timestep=init_timestep,
@ -728,7 +781,11 @@ class DenoiseLatentsInvocation(BaseInvocation):
@invocation(
"l2i", title="Latents to Image", tags=["latents", "image", "vae", "l2i"], category="latents", version="1.0.0"
"l2i",
title="Latents to Image",
tags=["latents", "image", "vae", "l2i"],
category="latents",
version="1.0.0",
)
class LatentsToImageInvocation(BaseInvocation):
"""Generates an image from latents."""
@ -743,7 +800,7 @@ class LatentsToImageInvocation(BaseInvocation):
)
tiled: bool = InputField(default=False, description=FieldDescriptions.tiled)
fp32: bool = InputField(default=DEFAULT_PRECISION == "float32", description=FieldDescriptions.fp32)
metadata: CoreMetadata = InputField(
metadata: Optional[CoreMetadata] = InputField(
default=None,
description=FieldDescriptions.core_metadata,
ui_hidden=True,
@ -754,7 +811,7 @@ class LatentsToImageInvocation(BaseInvocation):
latents = context.services.latents.get(self.latents.latents_name)
vae_info = context.services.model_manager.get_model(
**self.vae.vae.dict(),
**self.vae.vae.model_dump(),
context=context,
)
@ -816,7 +873,7 @@ class LatentsToImageInvocation(BaseInvocation):
node_id=self.id,
session_id=context.graph_execution_state_id,
is_intermediate=self.is_intermediate,
metadata=self.metadata.dict() if self.metadata else None,
metadata=self.metadata.model_dump() if self.metadata else None,
workflow=self.workflow,
)
@ -830,7 +887,13 @@ class LatentsToImageInvocation(BaseInvocation):
LATENTS_INTERPOLATION_MODE = Literal["nearest", "linear", "bilinear", "bicubic", "trilinear", "area", "nearest-exact"]
@invocation("lresize", title="Resize Latents", tags=["latents", "resize"], category="latents", version="1.0.0")
@invocation(
"lresize",
title="Resize Latents",
tags=["latents", "resize"],
category="latents",
version="1.0.0",
)
class ResizeLatentsInvocation(BaseInvocation):
"""Resizes latents to explicit width/height (in pixels). Provided dimensions are floor-divided by 8."""
@ -876,7 +939,13 @@ class ResizeLatentsInvocation(BaseInvocation):
return build_latents_output(latents_name=name, latents=resized_latents, seed=self.latents.seed)
@invocation("lscale", title="Scale Latents", tags=["latents", "resize"], category="latents", version="1.0.0")
@invocation(
"lscale",
title="Scale Latents",
tags=["latents", "resize"],
category="latents",
version="1.0.0",
)
class ScaleLatentsInvocation(BaseInvocation):
"""Scales latents by a given factor."""
@ -915,7 +984,11 @@ class ScaleLatentsInvocation(BaseInvocation):
@invocation(
"i2l", title="Image to Latents", tags=["latents", "image", "vae", "i2l"], category="latents", version="1.0.0"
"i2l",
title="Image to Latents",
tags=["latents", "image", "vae", "i2l"],
category="latents",
version="1.0.0",
)
class ImageToLatentsInvocation(BaseInvocation):
"""Encodes an image into latents."""
@ -979,7 +1052,7 @@ class ImageToLatentsInvocation(BaseInvocation):
image = context.services.images.get_pil_image(self.image.image_name)
vae_info = context.services.model_manager.get_model(
**self.vae.vae.dict(),
**self.vae.vae.model_dump(),
context=context,
)
@ -1007,7 +1080,13 @@ class ImageToLatentsInvocation(BaseInvocation):
return vae.encode(image_tensor).latents
@invocation("lblend", title="Blend Latents", tags=["latents", "blend"], category="latents", version="1.0.0")
@invocation(
"lblend",
title="Blend Latents",
tags=["latents", "blend"],
category="latents",
version="1.0.0",
)
class BlendLatentsInvocation(BaseInvocation):
"""Blend two latents using a given alpha. Latents must have same size."""