2noise/ChatTTS · error · ValueError

The lengths of prompts and prompt_token_ids must be the same

Error message

The lengths of prompts and prompt_token_ids must be the same.

What it means

When both prompts and prompt_token_ids are supplied, generate() uses them in parallel (prompts[i] is only used for display, token ids are taken from prompt_token_ids[i]). Their lengths must match or the request pairing is ambiguous, so the engine raises before submitting anything.

Source

Thrown at ChatTTS/model/velocity/llm.py:157

            prompt_token_ids: A list of token IDs for the prompts. If None, we
                use the tokenizer to convert the prompts to token IDs.
            use_tqdm: Whether to use tqdm to display the progress bar.

        Returns:
            A list of `RequestOutput` objects containing the generated
            completions in the same order as the input prompts.
        """
        if prompts is None and prompt_token_ids is None:
            raise ValueError("Either prompts or prompt_token_ids must be " "provided.")
        if isinstance(prompts, str):
            # Convert a single prompt to a list.
            prompts = [prompts]
        if (
            prompts is not None
            and prompt_token_ids is not None
            and len(prompts) != len(prompt_token_ids)
        ):
            raise ValueError(
                "The lengths of prompts and prompt_token_ids " "must be the same."
            )
        if sampling_params is None:
            # Use default sampling params.
            sampling_params = SamplingParams()

        # Add requests to the engine.
        num_requests = len(prompts) if prompts is not None else len(prompt_token_ids)
        for i in range(num_requests):
            prompt = prompts[i] if prompts is not None else None
            token_ids = None if prompt_token_ids is None else prompt_token_ids[i]
            self._add_request(prompt, sampling_params, token_ids)

        rtns = self._run_engine(use_tqdm)
        for i, rtn in enumerate(rtns):
            token_ids = rtn.outputs[0].token_ids
            for j, token_id in enumerate(token_ids):
                if len(token_id) == 1:

View on GitHub (pinned to 77b89ee281)

Solutions

1. Make both lists the same length: len(prompts) == len(prompt_token_ids).
2. If you deduplicated prompts, deduplicate token ids with the same filter so indices stay aligned.
3. Simpler: pass only prompt_token_ids (prompts=None) if you don't need display strings - then no pairing exists to break.

Example fix

# before
prompts = [p for p in prompts if p]          # length shrinks
out = llm.generate(prompts, prompt_token_ids)  # token_ids unchanged -> mismatch

# after
keep = [i for i, p in enumerate(prompts) if p]
out = llm.generate([prompts[i] for i in keep], [prompt_token_ids[i] for i in keep])

def aligned_prompts(prompts, prompt_token_ids):
    if prompts is not None and prompt_token_ids is not None:
        assert len(prompts) == len(prompt_token_ids), (
            f'{len(prompts)} prompts vs {len(prompt_token_ids)} token-id lists')
    return prompts, prompt_token_ids

try:
    out = llm.generate(prompts, prompt_token_ids)
except ValueError as e:
    if 'must be the same' in str(e):
        out = llm.generate(prompt_token_ids=prompt_token_ids)  # drop unaligned prompts
    else:
        raise

When it happens

Trigger: llm.generate(prompts=['a','b','c'], prompt_token_ids=[[1],[2]]) - lists of different lengths, typically after zipping/misaligning two separately built lists.

Common situations: Deduplicating prompts but not token ids (or vice versa); one list truncated by a batching bug; appending to one list in a loop but not the other.

Related errors

• Either prompts or prompt_token_ids must be provided.

AI-assisted analysis of 2noise/ChatTTS@77b89ee281 (2026-08-26). Data as JSON: /api/errors/c1246a98b6b1317f. Report an issue: GitHub.