BerriAI/litellm · error · ValueError

start_time is required, got=

Error message

start_time is required, got={start_time} of type {type(start_time)}

What it means

StandardLoggingPayloadSetup.cleanup_timestamps() normalizes a request's start/end/completion timestamps into epoch-second floats for the standard logging payload. It accepts only datetime.datetime instances or Python floats; every other type - None, str, int (time.time_ns(), int(time.time())), pandas.Timestamp, Decimal - raises this ValueError echoing the offending value and type. Note that Python ints fail even though they look numeric.

Solutions

Pass epoch seconds as float: float(time.time()) instead of int or time.time_ns()
Or pass datetime.datetime objects and let litellm call .timestamp() itself
Or omit start_time entirely - litellm stamps it internally per request
Coerce at the boundary: start_time = start_time.timestamp() if isinstance(start_time, datetime.datetime) else float(start_time)

Example fix

# before
import time
litellm.completion(model="gpt-4o", messages=m, start_time=time.time_ns())  # int -> ValueError

# after
import time
litellm.completion(model="gpt-4o", messages=m, start_time=float(time.time()))
# or: from datetime import datetime, timezone; start_time=datetime.now(timezone.utc)

Defensive patterns

Strategy: type-guard

Validate before calling

from datetime import datetime

def to_epoch_seconds(v):
    if isinstance(v, datetime):
        return v.timestamp()
    if isinstance(v, float):
        return v
    if isinstance(v, int) and not isinstance(v, bool):
        return float(v)
    raise TypeError(f"unsupported timestamp: {v!r}")

litellm.completion(model="gpt-4o", messages=m, start_time=to_epoch_seconds(start))

Type guard

from datetime import datetime

def is_valid_timestamp(v: object) -> bool:
    return (isinstance(v, datetime) or isinstance(v, float)) and not isinstance(v, bool)

Try / catch

try:
    litellm.completion(..., start_time=start)
except ValueError as e:
    if "start_time is required" in str(e):
        start = float(start)  # or datetime.now(timezone.utc)
        litellm.completion(..., start_time=start)

Prevention

Always pass datetime objects or float(time.time()); never ints or time.time_ns()
Centralize timestamp creation in one helper so all call sites share the coercion
Omit start_time when you have no explicit timing need - litellm stamps it itself

When it happens

Trigger: Passing start_time=time.time_ns() or int(time.time()) to litellm.completion/embedding, or a custom logger invoking StandardLoggingPayloadSetup.cleanup_timestamps with an ISO string, None, or other non-datetime/non-float value.

Common situations: Latency-test harnesses seeding nanosecond ints; timestamps loaded from JSON/DB as ISO-8601 strings and passed through; None leaking from conditionally computed timer values; pandas Timestamps from dataframe pipelines.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/7d014fd507e4c89f. Report an issue: GitHub.

Appendix: source

Thrown at litellm/litellm_core_utils/litellm_logging.py:4726

    ) -> tuple[float, float, float]:
        """
        Convert datetime objects to floats

        Args:
            start_time: Union[dt_object, float]
            end_time: Union[dt_object, float]
            completion_start_time: Union[dt_object, float]

        Returns:
            Tuple[float, float, float]: A tuple containing the start time, end time, and completion start time as floats.
        """

        if isinstance(start_time, datetime.datetime):
            start_time_float = start_time.timestamp()
        elif isinstance(start_time, float):
            start_time_float = start_time
        else:
            raise ValueError(f"start_time is required, got={start_time} of type {type(start_time)}")

        if isinstance(end_time, datetime.datetime):
            end_time_float = end_time.timestamp()
        elif isinstance(end_time, float):
            end_time_float = end_time
        else:
            raise ValueError(f"end_time is required, got={end_time} of type {type(end_time)}")

        if isinstance(completion_start_time, datetime.datetime):
            completion_start_time_float = completion_start_time.timestamp()
        elif isinstance(completion_start_time, float):
            completion_start_time_float = completion_start_time
        else:
            completion_start_time_float = end_time_float

        return start_time_float, end_time_float, completion_start_time_float

    @staticmethod

View on GitHub (pinned to 77b7c6c40c)