[Code scan] Package metadata allows incompatible Pydantic 1 installations

[njzjz]

This issue was found by a Codex global repository scan of tracked non-test files at commit 8c93925cb10b401b2b83c738bd9263fd74474468.

Relevant code

LAMBench/pyproject.toml

Lines 8 to 19 in 8c93925

	dependencies = [
	"ase",
	"python-dotenv",
	"numpy",
	"scipy",
	"pydantic",
	"pyyaml",
	"SQLAlchemy[pymysql]",
	"tqdm",
	"dpdata @ git+https://github.com/deepmodeling/dpdata.git@2eadf39a901130ac1e0464e0e24f4fb9545b9e4e#egg=dpdata",
	"pandas",
	"dftd3"

LAMBench/lambench/models/basemodel.py

Lines 1 to 22 in 8c93925

	from pathlib import Path
	from typing import Optional
	from pydantic import BaseModel, ConfigDict
	from enum import Enum
	from abc import abstractmethod
	from datetime import datetime
	class ModelType(str, Enum):
	DP = "DP"
	ASE = "ASE"
	class SkipTaskType(str, Enum):
	DirectPredictTask = "DirectPredictTask"
	PropertyFinetuneTask = "PropertyFinetuneTask"
	CalculatorTask = "CalculatorTask"
	class ModelMetadata(BaseModel):
	model_config = ConfigDict(extra="allow")
	pretty_name: str

LAMBench/lambench/metrics/post_process.py

Lines 180 to 190 in 8c93925

	def main():
	results = {}
	leaderboard_models = get_leaderboard_models()
	for model in leaderboard_models:
	r = results[model.model_metadata.pretty_name] = process_results_for_one_model(
	model
	)
	# PosixPath is not JSON serializable
	r["model"] = json.loads(
	json.dumps(model.model_dump(exclude={"model_path"}), default=str)
	)

Impact

The dependency is declared as bare pydantic, but the code uses Pydantic 2 APIs such as ConfigDict and model_dump(). A resolver can legally install Pydantic 1 when another package constrains it, which would make imports or runtime calls fail.

Suggested fix

Declare the intended major version explicitly, for example pydantic>=2,<3, or remove Pydantic 2-only APIs if Pydantic 1 compatibility is intended.