Merge branch 'main' into litellm_dev_03_14_2026_p2

litellm/litellm_core_utils/core_helpers.py

@@ -96,6 +96,8 @@ _FINISH_REASON_MAP: dict[str, OpenAIChatCompletionFinishReason] = {
     "tool_calls": "tool_calls",
     "function_call": "function_call",
     "content_filter": "content_filter",
+    # Anthropic Sonnet 4
+    "content_filtered": "content_filter",
 }
 
 

litellm/litellm_core_utils/default_encoding.py

@@ -15,16 +15,19 @@ except (ImportError, AttributeError):
         __name__, "litellm_core_utils/tokenizers"
     )
 
-# Check if the directory is writable. If not, use /tmp as a fallback.
-# This is especially important for non-root Docker environments where the package directory is read-only.
-is_non_root = os.getenv("LITELLM_NON_ROOT", "").lower() == "true"
-if not os.access(filename, os.W_OK) and is_non_root:
-    filename = "/tmp/tiktoken_cache"
-    os.makedirs(filename, exist_ok=True)
+# Always default TIKTOKEN_CACHE_DIR to the bundled tokenizers directory
+# unless the user explicitly overrides it via CUSTOM_TIKTOKEN_CACHE_DIR.
+# This keeps tiktoken fully offline-capable by default (see #1071).
+custom_cache_dir = os.getenv("CUSTOM_TIKTOKEN_CACHE_DIR")
+if custom_cache_dir:
+    # If the user opts into a custom cache dir, ensure it exists.
+    os.makedirs(custom_cache_dir, exist_ok=True)
+    cache_dir = custom_cache_dir
+else:
+    cache_dir = filename
+
+os.environ["TIKTOKEN_CACHE_DIR"] = cache_dir  # use local copy of tiktoken b/c of - https://github.com/BerriAI/litellm/issues/1071
 
-os.environ["TIKTOKEN_CACHE_DIR"] = os.getenv(
-    "CUSTOM_TIKTOKEN_CACHE_DIR", filename
-)  # use local copy of tiktoken b/c of - https://github.com/BerriAI/litellm/issues/1071
 import tiktoken
 import time
 import random
@@ -45,3 +48,4 @@ for attempt in range(_max_retries):
         # Exponential backoff with jitter to reduce collision probability
         delay = _retry_delay * (2**attempt) + random.uniform(0, 0.1)
         time.sleep(delay)
+

tests/test_default_encoding_non_root.py

@@ -1,49 +1,57 @@
+import importlib
 import os
-from unittest.mock import patch
+from unittest.mock import MagicMock, patch
+
+import litellm.litellm_core_utils.default_encoding as default_encoding
 
 
-def test_tiktoken_cache_fallback(monkeypatch):
+def _reload_default_encoding(monkeypatch, **env_overrides):
     """
-    Test that TIKTOKEN_CACHE_DIR falls back to /tmp/tiktoken_cache
-    if the default directory is not writable and LITELLM_NON_ROOT is true.
+    Helper to reload default_encoding with a clean TIKTOKEN_CACHE_DIR and
+    specific environment overrides.
     """
-    # Simulate non-root environment
-    monkeypatch.setenv("LITELLM_NON_ROOT", "true")
+    monkeypatch.delenv("TIKTOKEN_CACHE_DIR", raising=False)
     monkeypatch.delenv("CUSTOM_TIKTOKEN_CACHE_DIR", raising=False)
-
-    # Mock os.access to return False (not writable)
-    # and mock os.makedirs to avoid actually creating /tmp/tiktoken_cache on local machine
-    with patch("os.access", return_value=False), patch("os.makedirs"):
-        # We need to reload or re-run the logic in default_encoding.py
-        # But since it's already executed, we'll just test the logic directly
-        # mirroring what we wrote in the file.
-
-        filename = (
-            "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/tokenizers"
-        )
-        is_non_root = os.getenv("LITELLM_NON_ROOT", "").lower() == "true"
-
-        if not os.access(filename, os.W_OK) and is_non_root:
-            filename = "/tmp/tiktoken_cache"
-            # mock_makedirs(filename, exist_ok=True)
-
-        assert filename == "/tmp/tiktoken_cache"
+    for key, value in env_overrides.items():
+        monkeypatch.setenv(key, value)
+    importlib.reload(default_encoding)
 
 
-def test_tiktoken_cache_no_fallback_if_writable(monkeypatch):
+def test_default_encoding_uses_bundled_tokenizers_by_default(monkeypatch):
     """
-    Test that TIKTOKEN_CACHE_DIR does NOT fall back if writable
+    TIKTOKEN_CACHE_DIR should point at the bundled tokenizers directory
+    when no CUSTOM_TIKTOKEN_CACHE_DIR is set, even in non-root environments.
     """
-    monkeypatch.setenv("LITELLM_NON_ROOT", "true")
+    _reload_default_encoding(monkeypatch, LITELLM_NON_ROOT="true")
 
-    filename = "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/tokenizers"
+    assert "TIKTOKEN_CACHE_DIR" in os.environ
+    cache_dir = os.environ["TIKTOKEN_CACHE_DIR"]
+    assert "tokenizers" in cache_dir
 
-    with patch("os.access", return_value=True):
-        is_non_root = os.getenv("LITELLM_NON_ROOT", "").lower() == "true"
-        if not os.access(filename, os.W_OK) and is_non_root:
-            filename = "/tmp/tiktoken_cache"
 
-        assert (
-            filename
-            == "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/tokenizers"
+def test_custom_tiktoken_cache_dir_override(monkeypatch, tmp_path):
+    """
+    CUSTOM_TIKTOKEN_CACHE_DIR must override the default bundled directory
+    and the directory should be created if it does not exist.
+    Reload with an empty custom dir would otherwise trigger tiktoken to
+    download the vocab; we patch get_encoding so the test is offline-safe
+    and does not depend on tiktoken's in-memory cache state.
+    """
+    custom_dir = tmp_path / "tiktoken_cache"
+    with patch(
+        "litellm.litellm_core_utils.default_encoding.tiktoken.get_encoding",
+        return_value=MagicMock(),
+    ):
+        _reload_default_encoding(
+            monkeypatch, CUSTOM_TIKTOKEN_CACHE_DIR=str(custom_dir)
         )
+
+    cache_dir = os.environ.get("TIKTOKEN_CACHE_DIR")
+    assert cache_dir == str(custom_dir)
+    assert os.path.isdir(cache_dir)
+
+    # Restore module to a clean state so default_encoding.encoding is a real
+    # tiktoken Encoding, not the MagicMock, for any test that runs after this.
+    monkeypatch.delenv("TIKTOKEN_CACHE_DIR", raising=False)
+    monkeypatch.delenv("CUSTOM_TIKTOKEN_CACHE_DIR", raising=False)
+    importlib.reload(default_encoding)

tests/test_litellm/litellm_core_utils/test_core_helpers.py

@@ -59,20 +59,19 @@ VALID_OPENAI_FINISH_REASONS = {"stop", "length", "tool_calls", "function_call",
 
 
 class TestMapFinishReasonAnthropic:
-    def test_stop_sequence(self):
-        assert map_finish_reason("stop_sequence") == "stop"
-
-    def test_end_turn(self):
-        assert map_finish_reason("end_turn") == "stop"
-
-    def test_max_tokens(self):
-        assert map_finish_reason("max_tokens") == "length"
-
-    def test_tool_use(self):
-        assert map_finish_reason("tool_use") == "tool_calls"
-
-    def test_compaction(self):
-        assert map_finish_reason("compaction") == "length"
+    @pytest.mark.parametrize(
+        "provider_reason,expected",
+        [
+            ("stop_sequence", "stop"),
+            ("end_turn", "stop"),
+            ("max_tokens", "length"),
+            ("tool_use", "tool_calls"),
+            ("compaction", "length"),
+            ("content_filtered", "content_filter"),
+        ],
+    )
+    def test_anthropic_finish_reasons(self, provider_reason: str, expected: str) -> None:
+        assert map_finish_reason(provider_reason) == expected
 
 
 class TestMapFinishReasonGemini: