What causes "File not found. file_path={file_path}" in BerriAI/litellm?

Create the file at the configured path with the LlamaGuard category definitions (one category token per line, e.g. S1..S13 subset).

BerriAI/litellm · error · Exception

File not found. file_path={file_path}

Error message

File not found. file_path={file_path}

What it means

Exception raised in _ENTERPRISE_LlamaGuard.__init__ when litellm.llamaguard_unsafe_content_categories points to a file path that does not exist. This optional file defines which LlamaGuard safety categories to enforce (S1–S13 etc.); when configured, the hook must be able to read it at init time.

Source

Thrown at enterprise/litellm_enterprise/enterprise_callbacks/llama_guard.py:44

from litellm.types.utils import CallTypesLiteral, Choices, ModelResponse


class _ENTERPRISE_LlamaGuard(CustomLogger):
    # Class variables or attributes
    def __init__(self, model_name: Optional[str] = None):
        _model = model_name or litellm.llamaguard_model_name
        if _model is None:
            raise ValueError("model_name not set for LlamaGuard")
        self.model = _model
        file_path = litellm.llamaguard_unsafe_content_categories
        data = None

        if file_path is not None:
            try:
                with open(file_path, "r") as file:
                    data = file.read()
            except FileNotFoundError:
                raise Exception(f"File not found. file_path={file_path}")
            except Exception as e:
                raise Exception(f"An error occurred: {str(e)}, file_path={file_path}")

        self.unsafe_content_categories = data

        verbose_proxy_logger.debug(
            f"self.unsafe_content_categories: {self.unsafe_content_categories}"
        )

    def print_verbose(self, print_statement):
        try:
            verbose_proxy_logger.debug(print_statement)
            if litellm.set_verbose:
                print(print_statement)  # noqa
        except Exception:
            pass

    def set_custom_prompt_template(self, messages: list):

View on GitHub (pinned to 6c2dcb801b)

Solutions

Create the file at the configured path with the LlamaGuard category definitions (one category token per line, e.g. S1..S13 subset).
Use an absolute path to avoid CWD-dependent resolution.
In Docker, COPY/mount the categories file and reference the mounted path.
If you want all categories enforced by default, unset llamaguard_unsafe_content_categories entirely (the hook skips file loading when it is None).

Example fix

# before
litellm_settings:
  llamaguard_unsafe_content_categories: ./unsafe_categories.txt  # missing

# after
litellm_settings:
  llamaguard_unsafe_content_categories: /etc/litellm/unsafe_categories.txt
# or remove the setting to use defaults

Defensive patterns

Strategy: validation

Validate before calling

import os

cats = os.environ.get("LLAMAGUARD_UNSAFE_CONTENT_CATEGORIES_FILE")
if cats and not os.path.isfile(cats):
    raise SystemExit(f"unsafe categories file missing: {cats}")

Try / catch

try:
    guard = _ENTERPRISE_LlamaGuard(model_name=name)
except Exception as e:
    if "File not found" in str(e):
        logger.error("create the categories file or unset llamaguard_unsafe_content_categories")
    raise

Prevention

Use an absolute path for llamaguard_unsafe_content_categories.
Ship the categories file inside the image (COPY) rather than assuming CWD.
Unset the setting entirely to use LlamaGuard defaults when a custom subset is unnecessary.
Validate the file exists in a startup probe.

When it happens

Trigger: Setting llamaguard_unsafe_content_categories: "./unsafe_categories.txt" in litellm settings where the file is missing, the relative path resolves against a different working directory, or the container image lacks the file.

Common situations: Following LlamaGuard docs that reference a categories file without creating/mounting it; moving the proxy's working directory so relative paths break; forgetting to COPY the file into a Docker image.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/d061a1bc0e4905c9. Report an issue: GitHub.