20 paired (schema, args, expected outcome) fixtures for testing LLM tool-call validators. Each row contains a tool definition, the args an LLM tried to pass, and whether validation should pass or fail (with the expected error if it should fail).
from datasets import load_dataset
ds = load_dataset("mukunda1729/tool-arg-validation-cases", split="train")
for case in ds:
if case["valid"]:
assert validator(case["schema"], case["args"]).valid… See the full description on the dataset page:
https://huggingface.co/datasets/mukunda1729/tool-arg-validation-cases.
