tests/adapters/test_modal_chat_model.py

import pytest


def test_modal_chat_model_raises_without_env(monkeypatch):
    monkeypatch.delenv("MODAL_INFERENCE_URL", raising=False)
    monkeypatch.delenv("MODAL_API_KEY", raising=False)
    from adapters.modal_chat_model import ModalChatModel
    with pytest.raises(KeyError):
        ModalChatModel()


def test_modal_chat_model_init_with_env(monkeypatch):
    monkeypatch.setenv("MODAL_INFERENCE_URL", "https://example.modal.run")
    monkeypatch.setenv("MODAL_API_KEY", "test-key")
    monkeypatch.setenv("MODAL_MODEL_ID", "Qwen/Qwen2.5-32B-Instruct")
    from adapters.modal_chat_model import ModalChatModel
    model = ModalChatModel()
    assert model is not None
    # base_url and model id wired from env
    assert model._llm.openai_api_base == "https://example.modal.run/v1"
    assert model._llm.model_name == "Qwen/Qwen2.5-32B-Instruct"


def test_modal_chat_model_uses_temperature_zero(monkeypatch):
    """Determinism + primary truncation fix: structured calls must sample at temperature 0,
    not the model's 0.6 default (which rambles to max_tokens → truncated JSON)."""
    monkeypatch.setenv("MODAL_INFERENCE_URL", "https://example.modal.run")
    monkeypatch.setenv("MODAL_API_KEY", "test-key")
    monkeypatch.setenv("MODAL_MODEL_ID", "Qwen/Qwen2.5-32B-Instruct")
    from adapters.modal_chat_model import ModalChatModel
    model = ModalChatModel()
    assert model._llm.temperature == 0