mirror of
https://github.com/BerriAI/litellm.git
synced 2026-08-28 05:25:59 +00:00
d9a460277a
1. Cap requires-python to <3.14 — no deps ship 3.14 wheels yet, and uv's cross-version resolver fails on the Python 3.14 split. 2. Change exclude-newer from relative "30 days" to absolute "2026-04-10" so the lockfile stays reproducible. The relative date caused cryptography==46.0.7 (published April 8) to fall outside the window. 3. Parametrize test_eager_loading_env_var_values instead of looping — with xdist the 6 subprocess cases can run in parallel instead of all running sequentially on one worker (~13 min → ~2 min). Also removed redundant case variants (Yes/YES/On/ON) that test the same str_to_bool code path.
107 lines
3.9 KiB
Python
107 lines
3.9 KiB
Python
"""
|
|
Test for LITELLM_DISABLE_LAZY_LOADING environment variable.
|
|
|
|
This test verifies that when LITELLM_DISABLE_LAZY_LOADING is set,
|
|
encoding is loaded at import time (pre-#18070 behavior) instead of lazy loading.
|
|
|
|
This addresses issue #18659: VCR cassette creation broken by lazy loading.
|
|
For now, this only affects encoding as it was the only reported issue.
|
|
|
|
Tests that need to clear sys.modules and re-import litellm run in subprocesses
|
|
to avoid contaminating the test process's module graph (which breaks mock.patch
|
|
for all subsequent tests on the same xdist worker).
|
|
"""
|
|
import subprocess
|
|
import sys
|
|
import textwrap
|
|
|
|
import pytest
|
|
|
|
|
|
def _run_python(script: str, env_override: dict | None = None) -> subprocess.CompletedProcess:
|
|
"""Run a Python script in a subprocess and return the result."""
|
|
import os
|
|
|
|
env = os.environ.copy()
|
|
# Remove the var so each test controls it explicitly
|
|
env.pop("LITELLM_DISABLE_LAZY_LOADING", None)
|
|
env.pop("TIKTOKEN_CACHE_DIR", None)
|
|
if env_override:
|
|
env.update(env_override)
|
|
return subprocess.run(
|
|
[sys.executable, "-c", textwrap.dedent(script)],
|
|
capture_output=True,
|
|
text=True,
|
|
env=env,
|
|
# Importing litellm can cold-load tiktoken/tokenizer assets and is
|
|
# occasionally slow on CI runners; these tests validate behavior, not speed.
|
|
timeout=180,
|
|
)
|
|
|
|
|
|
def test_eager_loading_enabled():
|
|
"""Test that encoding is loaded at import time when env var is set"""
|
|
result = _run_python(
|
|
"""
|
|
import litellm
|
|
assert hasattr(litellm, "encoding"), "Encoding should be available when eager loading is enabled"
|
|
encoding = litellm.encoding
|
|
assert encoding is not None, "Encoding should not be None"
|
|
tokens = encoding.encode("Hello, world!")
|
|
assert len(tokens) > 0, "Encoding should work"
|
|
""",
|
|
env_override={"LITELLM_DISABLE_LAZY_LOADING": "1"},
|
|
)
|
|
assert result.returncode == 0, f"Subprocess failed:\nstdout: {result.stdout}\nstderr: {result.stderr}"
|
|
|
|
|
|
@pytest.mark.parametrize("value", ["1", "true", "True", "TRUE", "yes", "on"])
|
|
def test_eager_loading_env_var_values(value):
|
|
"""Test that various env var values enable eager loading"""
|
|
result = _run_python(
|
|
"""
|
|
import litellm
|
|
assert hasattr(litellm, "encoding"), "Encoding should be available"
|
|
encoding = litellm.encoding
|
|
tokens = encoding.encode("test")
|
|
assert len(tokens) > 0
|
|
""",
|
|
env_override={"LITELLM_DISABLE_LAZY_LOADING": value},
|
|
)
|
|
assert result.returncode == 0, (
|
|
f"Failed for value {value!r}:\nstdout: {result.stdout}\nstderr: {result.stderr}"
|
|
)
|
|
|
|
|
|
def test_lazy_loading_default():
|
|
"""Test that encoding is lazy loaded by default (when env var is not set)"""
|
|
result = _run_python(
|
|
"""
|
|
import litellm
|
|
# Encoding should be accessible via __getattr__ (lazy loading)
|
|
encoding = litellm.encoding
|
|
tokens = encoding.encode("Hello, world!")
|
|
assert len(tokens) > 0, "Encoding should work"
|
|
""",
|
|
)
|
|
assert result.returncode == 0, f"Subprocess failed:\nstdout: {result.stdout}\nstderr: {result.stderr}"
|
|
|
|
|
|
def test_tiktoken_cache_dir_set_on_lazy_load():
|
|
"""Test that TIKTOKEN_CACHE_DIR is set when encoding is lazy loaded.
|
|
|
|
This ensures the local tiktoken cache is used instead of downloading
|
|
from the internet. Regression test for issue #19768.
|
|
"""
|
|
result = _run_python(
|
|
"""
|
|
import os
|
|
import litellm
|
|
# Access encoding (triggers lazy load)
|
|
_ = litellm.encoding
|
|
assert "TIKTOKEN_CACHE_DIR" in os.environ, "TIKTOKEN_CACHE_DIR should be set after lazy loading encoding"
|
|
cache_dir = os.environ["TIKTOKEN_CACHE_DIR"]
|
|
assert "tokenizers" in cache_dir, f"TIKTOKEN_CACHE_DIR should point to tokenizers directory, got: {cache_dir}"
|
|
""",
|
|
)
|
|
assert result.returncode == 0, f"Subprocess failed:\nstdout: {result.stdout}\nstderr: {result.stderr}"
|