[Python] Pydantic v2 + Python 3.13: PEP 695 type-alias of a generic model returns...

I’m experimenting with Python 3.13 and Pydantic v2.11.7. My goal is to avoid writing

GenericTraining[Discriminate[L1LossConfig]]


over and over—I’d like a terse alias such that I can do

TrainingLike[L1LossConfig]


and still have a fully-fledged BaseModel subclass (with Pydantic’s model_rebuild, validators, etc.).

Minimal reproducible example

from typing import Annotated, Literal
from pydantic import BaseModel, Field


class L1LossConfig(BaseModel):
loss_type: Literal["l1"] = "l1"


class GenericTraining[LossT](BaseModel):
loss: LossT


type Discriminate[T] = Annotated[T, Field(discriminator="loss_type")]

# -----------------------------------------------------------------------------
# 1️⃣ VERBOSE version – works
Training1 = GenericTraining[Discriminate[L1LossConfig]]
Training1.model_rebuild() # ✅ OK

# 2️⃣ TERSE version – breaks
type TrainingLike[T] = GenericTraining[Discriminate[T]]

Training2 = TrainingLike[L1LossConfig]
Training2.model_rebuild() # ❌ Attribute "model_rebuild" is unknown


How can I have TrainingLike[L1LossConfig] behave exactly like GenericTraining[Discriminate[L1LossConfig]], i.e. to be a real subclass of BaseModel?

Continue reading...