mirror of
https://github.com/invoke-ai/InvokeAI
synced 2024-08-30 20:32:17 +00:00
c238a7f18b
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.
258 lines
9.5 KiB
Python
258 lines
9.5 KiB
Python
import pytest
|
|
from pydantic import TypeAdapter, ValidationError
|
|
|
|
from invokeai.app.services.session_queue.session_queue_common import (
|
|
Batch,
|
|
BatchDataCollection,
|
|
BatchDatum,
|
|
NodeFieldValue,
|
|
calc_session_count,
|
|
create_session_nfv_tuples,
|
|
populate_graph,
|
|
prepare_values_to_insert,
|
|
)
|
|
from invokeai.app.services.shared.graph import Graph, GraphExecutionState, GraphInvocation
|
|
from tests.nodes.test_nodes import PromptTestInvocation
|
|
|
|
|
|
@pytest.fixture
|
|
def batch_data_collection() -> BatchDataCollection:
|
|
return [
|
|
[
|
|
# zipped
|
|
BatchDatum(node_path="1", field_name="prompt", items=["Banana sushi", "Grape sushi"]),
|
|
BatchDatum(node_path="2", field_name="prompt", items=["Strawberry sushi", "Blueberry sushi"]),
|
|
],
|
|
[
|
|
BatchDatum(node_path="3", field_name="prompt", items=["Orange sushi", "Apple sushi"]),
|
|
],
|
|
]
|
|
|
|
|
|
@pytest.fixture
|
|
def batch_graph() -> Graph:
|
|
g = Graph()
|
|
g.add_node(PromptTestInvocation(id="1", prompt="Chevy"))
|
|
g.add_node(PromptTestInvocation(id="2", prompt="Toyota"))
|
|
g.add_node(PromptTestInvocation(id="3", prompt="Subaru"))
|
|
g.add_node(PromptTestInvocation(id="4", prompt="Nissan"))
|
|
return g
|
|
|
|
|
|
def test_populate_graph_with_subgraph():
|
|
g1 = Graph()
|
|
g1.add_node(PromptTestInvocation(id="1", prompt="Banana sushi"))
|
|
g1.add_node(PromptTestInvocation(id="2", prompt="Banana sushi"))
|
|
n1 = PromptTestInvocation(id="1", prompt="Banana snake")
|
|
subgraph = Graph()
|
|
subgraph.add_node(n1)
|
|
g1.add_node(GraphInvocation(id="3", graph=subgraph))
|
|
|
|
nfvs = [
|
|
NodeFieldValue(node_path="1", field_name="prompt", value="Strawberry sushi"),
|
|
NodeFieldValue(node_path="2", field_name="prompt", value="Strawberry sunday"),
|
|
NodeFieldValue(node_path="3.1", field_name="prompt", value="Strawberry snake"),
|
|
]
|
|
|
|
g2 = populate_graph(g1, nfvs)
|
|
|
|
# do not mutate g1
|
|
assert g1 is not g2
|
|
assert g2.get_node("1").prompt == "Strawberry sushi"
|
|
assert g2.get_node("2").prompt == "Strawberry sunday"
|
|
assert g2.get_node("3.1").prompt == "Strawberry snake"
|
|
|
|
|
|
def test_create_sessions_from_batch_with_runs(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection, runs=2)
|
|
t = list(create_session_nfv_tuples(batch=b, maximum=1000))
|
|
# 2 list[BatchDatum] * length 2 * 2 runs = 8
|
|
assert len(t) == 8
|
|
|
|
assert t[0][0].graph.get_node("1").prompt == "Banana sushi"
|
|
assert t[0][0].graph.get_node("2").prompt == "Strawberry sushi"
|
|
assert t[0][0].graph.get_node("3").prompt == "Orange sushi"
|
|
assert t[0][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
assert t[1][0].graph.get_node("1").prompt == "Banana sushi"
|
|
assert t[1][0].graph.get_node("2").prompt == "Strawberry sushi"
|
|
assert t[1][0].graph.get_node("3").prompt == "Apple sushi"
|
|
assert t[1][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
assert t[2][0].graph.get_node("1").prompt == "Grape sushi"
|
|
assert t[2][0].graph.get_node("2").prompt == "Blueberry sushi"
|
|
assert t[2][0].graph.get_node("3").prompt == "Orange sushi"
|
|
assert t[2][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
assert t[3][0].graph.get_node("1").prompt == "Grape sushi"
|
|
assert t[3][0].graph.get_node("2").prompt == "Blueberry sushi"
|
|
assert t[3][0].graph.get_node("3").prompt == "Apple sushi"
|
|
assert t[3][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
# repeat for second run
|
|
assert t[4][0].graph.get_node("1").prompt == "Banana sushi"
|
|
assert t[4][0].graph.get_node("2").prompt == "Strawberry sushi"
|
|
assert t[4][0].graph.get_node("3").prompt == "Orange sushi"
|
|
assert t[4][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
assert t[5][0].graph.get_node("1").prompt == "Banana sushi"
|
|
assert t[5][0].graph.get_node("2").prompt == "Strawberry sushi"
|
|
assert t[5][0].graph.get_node("3").prompt == "Apple sushi"
|
|
assert t[5][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
assert t[6][0].graph.get_node("1").prompt == "Grape sushi"
|
|
assert t[6][0].graph.get_node("2").prompt == "Blueberry sushi"
|
|
assert t[6][0].graph.get_node("3").prompt == "Orange sushi"
|
|
assert t[6][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
assert t[7][0].graph.get_node("1").prompt == "Grape sushi"
|
|
assert t[7][0].graph.get_node("2").prompt == "Blueberry sushi"
|
|
assert t[7][0].graph.get_node("3").prompt == "Apple sushi"
|
|
assert t[7][0].graph.get_node("4").prompt == "Nissan"
|
|
|
|
|
|
def test_create_sessions_from_batch_without_runs(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection)
|
|
t = list(create_session_nfv_tuples(batch=b, maximum=1000))
|
|
# 2 list[BatchDatum] * length 2 * 1 runs = 8
|
|
assert len(t) == 4
|
|
|
|
|
|
def test_create_sessions_from_batch_without_batch(batch_graph):
|
|
b = Batch(graph=batch_graph, runs=2)
|
|
t = list(create_session_nfv_tuples(batch=b, maximum=1000))
|
|
# 2 runs
|
|
assert len(t) == 2
|
|
|
|
|
|
def test_create_sessions_from_batch_without_batch_or_runs(batch_graph):
|
|
b = Batch(graph=batch_graph)
|
|
t = list(create_session_nfv_tuples(batch=b, maximum=1000))
|
|
# 1 run
|
|
assert len(t) == 1
|
|
|
|
|
|
def test_create_sessions_from_batch_with_runs_and_max(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection, runs=2)
|
|
t = list(create_session_nfv_tuples(batch=b, maximum=5))
|
|
# 2 list[BatchDatum] * length 2 * 2 runs = 8, but max is 5
|
|
assert len(t) == 5
|
|
|
|
|
|
def test_calc_session_count(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection, runs=2)
|
|
# 2 list[BatchDatum] * length 2 * 2 runs = 8
|
|
assert calc_session_count(batch=b) == 8
|
|
|
|
|
|
def test_prepare_values_to_insert(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection, runs=2)
|
|
values = prepare_values_to_insert(queue_id="default", batch=b, priority=0, max_new_queue_items=1000)
|
|
assert len(values) == 8
|
|
|
|
session_adapter = TypeAdapter(GraphExecutionState)
|
|
# graph should be serialized
|
|
ges = session_adapter.validate_json(values[0].session)
|
|
|
|
# graph values should be populated
|
|
assert ges.graph.get_node("1").prompt == "Banana sushi"
|
|
assert ges.graph.get_node("2").prompt == "Strawberry sushi"
|
|
assert ges.graph.get_node("3").prompt == "Orange sushi"
|
|
assert ges.graph.get_node("4").prompt == "Nissan"
|
|
|
|
# session ids should match deserialized graph
|
|
assert [v.session_id for v in values] == [session_adapter.validate_json(v.session).id for v in values]
|
|
|
|
# should unique session ids
|
|
sids = [v.session_id for v in values]
|
|
assert len(sids) == len(set(sids))
|
|
|
|
nfv_list_adapter = TypeAdapter(list[NodeFieldValue])
|
|
# should have 3 node field values
|
|
assert type(values[0].field_values) is str
|
|
assert len(nfv_list_adapter.validate_json(values[0].field_values)) == 3
|
|
|
|
# should have batch id and priority
|
|
assert all(v.batch_id == b.batch_id for v in values)
|
|
assert all(v.priority == 0 for v in values)
|
|
|
|
|
|
def test_prepare_values_to_insert_with_priority(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection, runs=2)
|
|
values = prepare_values_to_insert(queue_id="default", batch=b, priority=1, max_new_queue_items=1000)
|
|
assert all(v.priority == 1 for v in values)
|
|
|
|
|
|
def test_prepare_values_to_insert_with_max(batch_data_collection, batch_graph):
|
|
b = Batch(graph=batch_graph, data=batch_data_collection, runs=2)
|
|
values = prepare_values_to_insert(queue_id="default", batch=b, priority=1, max_new_queue_items=5)
|
|
assert len(values) == 5
|
|
|
|
|
|
def test_cannot_create_bad_batch_items_length(batch_graph):
|
|
with pytest.raises(ValidationError, match="Zipped batch items must all have the same length"):
|
|
Batch(
|
|
graph=batch_graph,
|
|
data=[
|
|
[
|
|
BatchDatum(node_path="1", field_name="prompt", items=["Banana sushi"]), # 1 item
|
|
BatchDatum(node_path="2", field_name="prompt", items=["Toyota", "Nissan"]), # 2 items
|
|
],
|
|
],
|
|
)
|
|
|
|
|
|
def test_cannot_create_bad_batch_items_type(batch_graph):
|
|
with pytest.raises(ValidationError, match="All items in a batch must have the same type"):
|
|
Batch(
|
|
graph=batch_graph,
|
|
data=[
|
|
[
|
|
BatchDatum(node_path="1", field_name="prompt", items=["Banana sushi", 123]),
|
|
]
|
|
],
|
|
)
|
|
|
|
|
|
def test_cannot_create_bad_batch_unique_ids(batch_graph):
|
|
with pytest.raises(ValidationError, match="Each batch data must have unique node_id and field_name"):
|
|
Batch(
|
|
graph=batch_graph,
|
|
data=[
|
|
[
|
|
BatchDatum(node_path="1", field_name="prompt", items=["Banana sushi"]),
|
|
],
|
|
[
|
|
BatchDatum(node_path="1", field_name="prompt", items=["Banana sushi"]),
|
|
],
|
|
],
|
|
)
|
|
|
|
|
|
def test_cannot_create_bad_batch_nodes_exist(
|
|
batch_graph,
|
|
):
|
|
with pytest.raises(ValidationError, match=r"Node .* not found in graph"):
|
|
Batch(
|
|
graph=batch_graph,
|
|
data=[
|
|
[
|
|
BatchDatum(node_path="batman", field_name="prompt", items=["Banana sushi"]),
|
|
],
|
|
],
|
|
)
|
|
|
|
|
|
def test_cannot_create_bad_batch_fields_exist(
|
|
batch_graph,
|
|
):
|
|
with pytest.raises(ValidationError, match=r"Field .* not found in node"):
|
|
Batch(
|
|
graph=batch_graph,
|
|
data=[
|
|
[
|
|
BatchDatum(node_path="1", field_name="batman", items=["Banana sushi"]),
|
|
],
|
|
],
|
|
)
|