BerriAI/litellm · warning · ValueError

Error counting field '{field_name}': {e}

Error message

Error counting field '{field_name}': {e}

What it means

A wrapper around token counting of a named field in an Anthropic content block (e.g. 'text' or 'input'): the underlying count call threw (often a nested image/object inside tool_use input, or a non-serializable value). If default_token_count was supplied, that fallback is returned instead; otherwise the error is re-raised wrapped with the field name.

Source

Thrown at litellm/litellm_core_utils/token_counter.py:681

        field_value = content.get(field_name)
        if field_value is None:
            continue
        try:
            if isinstance(field_value, str):
                tokens += count_function(field_value)
            elif isinstance(field_value, list):
                tokens += _count_content_list(
                    count_function,
                    field_value,
                    use_default_image_token_count,
                    default_token_count,
                )
            elif isinstance(field_value, dict):
                tokens += count_function(str(field_value))
        except Exception as e:
            if default_token_count is not None:
                return default_token_count
            raise ValueError(f"Error counting field '{field_name}': {e}")
    return tokens


def _count_content_list(
    count_function: TokenCounterFunction,
    content_list: OpenAIMessageContent,
    use_default_image_token_count: bool,
    default_token_count: int | None,
) -> int:
    """
    Recursively count tokens from a list of content blocks.
    """
    try:
        num_tokens = 0
        for c in content_list:
            if isinstance(c, str):
                num_tokens += count_function(c)
            elif c["type"] == "text":

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Inspect the inner exception embedded in the message - it names the field and root cause; fix or normalize that field's value.
2. Pass default_token_count (e.g. 0 or an estimate) when counting untrusted or complex messages so counting degrades gracefully.

Example fix

# before
n = litellm.token_counter(model="claude-3-5-sonnet", messages=msgs)  # raises on odd field

# after
n = litellm.token_counter(model="claude-3-5-sonnet", messages=msgs, default_token_count=0)

n = litellm.token_counter(model=model, messages=msgs, default_token_count=4)  # degrades instead of raising

try:
    n = litellm.token_counter(model=m, messages=msgs)
except ValueError as e:
    if "Error counting field" in str(e):
        n = litellm.token_counter(model=m, messages=msgs, default_token_count=0)
    else:
        raise

When it happens

Trigger: A tool_use block whose 'input' contains nested structures the counter chokes on (e.g. image data inside input), or a field value of an unexpected type - and token_counter was called without default_token_count.

Common situations: Complex nested tool inputs (base64 images inside tool arguments); malformed base64 in tool_result content; cost-accounting code that does not pass default_token_count.

Related errors

• Error getting number of tokens from content list: {e}, defau
• Anthropic content missing required field: 'type'
• Unknown Anthropic content type: '{content_type}'
• Missing required fields in {content_type} block: {', '.join(
• WebSearchInterception: missing follow-up messages

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/fca3eea2737b5c1a. Report an issue: GitHub.