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

70
invokeai/app/invocations/compel.py
View File

@ -1,6 +1,6 @@
import re
from dataclasses import dataclass
from typing import List, Union
from typing import List, Optional, Union
import torch
from compel import Compel, ReturnedEmbeddingsType
@ -43,7 +43,13 @@ class ConditioningFieldData:
# PerpNeg = "perp_neg"
@invocation("compel", title="Prompt", tags=["prompt", "compel"], category="conditioning", version="1.0.0")
@invocation(
"compel",
title="Prompt",
tags=["prompt", "compel"],
category="conditioning",
version="1.0.0",
)
class CompelInvocation(BaseInvocation):
"""Parse prompt using compel package to conditioning."""
@ -61,17 +67,19 @@ class CompelInvocation(BaseInvocation):
@torch.no_grad()
def invoke(self, context: InvocationContext) -> ConditioningOutput:
tokenizer_info = context.services.model_manager.get_model(
**self.clip.tokenizer.dict(),
**self.clip.tokenizer.model_dump(),
context=context,
)
text_encoder_info = context.services.model_manager.get_model(
**self.clip.text_encoder.dict(),
**self.clip.text_encoder.model_dump(),
context=context,
)
def _lora_loader():
for lora in self.clip.loras:
lora_info = context.services.model_manager.get_model(**lora.dict(exclude={"weight"}), context=context)
lora_info = context.services.model_manager.get_model(
**lora.model_dump(exclude={"weight"}), context=context
)
yield (lora_info.context.model, lora.weight)
del lora_info
return
@ -160,11 +168,11 @@ class SDXLPromptInvocationBase:
zero_on_empty: bool,
):
tokenizer_info = context.services.model_manager.get_model(
**clip_field.tokenizer.dict(),
**clip_field.tokenizer.model_dump(),
context=context,
)
text_encoder_info = context.services.model_manager.get_model(
**clip_field.text_encoder.dict(),
**clip_field.text_encoder.model_dump(),
context=context,
)
@ -172,7 +180,11 @@ class SDXLPromptInvocationBase:
if prompt == "" and zero_on_empty:
cpu_text_encoder = text_encoder_info.context.model
c = torch.zeros(
(1, cpu_text_encoder.config.max_position_embeddings, cpu_text_encoder.config.hidden_size),
(
1,
cpu_text_encoder.config.max_position_embeddings,
cpu_text_encoder.config.hidden_size,
),
dtype=text_encoder_info.context.cache.precision,
)
if get_pooled:
@ -186,7 +198,9 @@ class SDXLPromptInvocationBase:
def _lora_loader():
for lora in clip_field.loras:
lora_info = context.services.model_manager.get_model(**lora.dict(exclude={"weight"}), context=context)
lora_info = context.services.model_manager.get_model(
**lora.model_dump(exclude={"weight"}), context=context
)
yield (lora_info.context.model, lora.weight)
del lora_info
return
@ -273,8 +287,16 @@ class SDXLPromptInvocationBase:
class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
"""Parse prompt using compel package to conditioning."""
prompt: str = InputField(default="", description=FieldDescriptions.compel_prompt, ui_component=UIComponent.Textarea)
style: str = InputField(default="", description=FieldDescriptions.compel_prompt, ui_component=UIComponent.Textarea)
prompt: str = InputField(
default="",
description=FieldDescriptions.compel_prompt,
ui_component=UIComponent.Textarea,
)
style: str = InputField(
default="",
description=FieldDescriptions.compel_prompt,
ui_component=UIComponent.Textarea,
)
original_width: int = InputField(default=1024, description="")
original_height: int = InputField(default=1024, description="")
crop_top: int = InputField(default=0, description="")
@ -310,7 +332,9 @@ class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
[
c1,
torch.zeros(
(c1.shape[0], c2.shape[1] - c1.shape[1], c1.shape[2]), device=c1.device, dtype=c1.dtype
(c1.shape[0], c2.shape[1] - c1.shape[1], c1.shape[2]),
device=c1.device,
dtype=c1.dtype,
),
],
dim=1,
@ -321,7 +345,9 @@ class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
[
c2,
torch.zeros(
(c2.shape[0], c1.shape[1] - c2.shape[1], c2.shape[2]), device=c2.device, dtype=c2.dtype
(c2.shape[0], c1.shape[1] - c2.shape[1], c2.shape[2]),
device=c2.device,
dtype=c2.dtype,
),
],
dim=1,
@ -359,7 +385,9 @@ class SDXLRefinerCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase
"""Parse prompt using compel package to conditioning."""
style: str = InputField(
default="", description=FieldDescriptions.compel_prompt, ui_component=UIComponent.Textarea
default="",
description=FieldDescriptions.compel_prompt,
ui_component=UIComponent.Textarea,
) # TODO: ?
original_width: int = InputField(default=1024, description="")
original_height: int = InputField(default=1024, description="")
@ -403,10 +431,16 @@ class SDXLRefinerCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase
class ClipSkipInvocationOutput(BaseInvocationOutput):
"""Clip skip node output"""
clip: ClipField = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
clip: Optional[ClipField] = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
@invocation("clip_skip", title="CLIP Skip", tags=["clipskip", "clip", "skip"], category="conditioning", version="1.0.0")
@invocation(
"clip_skip",
title="CLIP Skip",
tags=["clipskip", "clip", "skip"],
category="conditioning",
version="1.0.0",
)
class ClipSkipInvocation(BaseInvocation):
"""Skip layers in clip text_encoder model."""
@ -421,7 +455,9 @@ class ClipSkipInvocation(BaseInvocation):
def get_max_token_count(
tokenizer, prompt: Union[FlattenedPrompt, Blend, Conjunction], truncate_if_too_long=False
tokenizer,
prompt: Union[FlattenedPrompt, Blend, Conjunction],
truncate_if_too_long=False,
) -> int:
if type(prompt) is Blend:
blend: Blend = prompt