run-llama/llama_index · error · ValueError

Chunk size {chunk_size} is not positive.

Error message

Chunk size {chunk_size} is not positive.

What it means

After PromptHelper computes the available chunk size (available context minus padding and room for num_chunks), it validates the result is positive before building a TokenTextSplitter. A non-positive chunk size means there is no leftover space in the context window to place even one text chunk.

Source

Thrown at llama-index-core/llama_index/core/indices/prompt_helper.py:248

        return result

    def get_text_splitter_given_prompt(
        self,
        prompt: BasePromptTemplate,
        num_chunks: int = 1,
        padding: int = DEFAULT_PADDING,
        llm: Optional[LLM] = None,
        tools: Optional[List["BaseTool"]] = None,
    ) -> TokenTextSplitter:
        """
        Get text splitter configured to maximally pack available context window,
        taking into account of given prompt, and desired number of chunks.
        """
        chunk_size = self._get_available_chunk_size(
            prompt, num_chunks, padding=padding, llm=llm, tools=tools
        )
        if chunk_size <= 0:
            raise ValueError(f"Chunk size {chunk_size} is not positive.")
        chunk_overlap = int(self.chunk_overlap_ratio * chunk_size)
        return TokenTextSplitter(
            separator=self.separator,
            chunk_size=chunk_size,
            chunk_overlap=chunk_overlap,
            tokenizer=self._token_counter.tokenizer,
        )

    def truncate(
        self,
        prompt: BasePromptTemplate,
        text_chunks: Sequence[str],
        padding: int = DEFAULT_PADDING,
        llm: Optional[LLM] = None,
        tools: Optional[List["BaseTool"]] = None,
    ) -> List[str]:
        """Truncate text chunks to fit available context window."""
        if not text_chunks:

View on GitHub (pinned to afd0fef371)

Solutions

1. Fix the underlying budget: raise context_window or reduce num_output so available context is positive (see error 200)
2. Reduce num_chunks so each chunk gets more tokens
3. Reduce the padding argument passed to get_text_splitter
4. Shorten the prompt template so fewer tokens are pre-consumed

Example fix

# before
splitter = prompt_helper.get_text_splitter(prompt, num_chunks=10, padding=DEFAULT_PADDING)

# after
splitter = prompt_helper.get_text_splitter(prompt, num_chunks=2, padding=5)

available = helper.context_window - helper._token_counter(prompt) - helper.num_output
if available <= 0 or (available // num_chunks) - padding <= 0:
    num_chunks = 1
    padding = min(padding, max(available - 1, 0))

try:
    splitter = prompt_helper.get_text_splitter(prompt, num_chunks, padding=padding)
except ValueError as e:
    if 'not positive' in str(e):
        splitter = prompt_helper.get_text_splitter(prompt, num_chunks=1, padding=1)
    else:
        raise

When it happens

Trigger: Calling get_text_splitter(prompt, num_chunks, padding=...) when the prompt + num_output already consume the whole window (chained from the same arithmetic as error 200), or when padding/num_chunks eat the remaining budget (chunk_size = available/num_chunks - padding drops to <= 0).

Common situations: Large num_output or long prompts against a small context window; requesting many chunks (high num_chunks) so per-chunk budget collapses; big padding values.

Related errors

• Calculated available context size {context_size_tokens} was
• Configured node parser does not have chunk size.
• Configured node parser does not have chunk overlap.
• First argument to Readability constructor should be a docume
• Aborting parsing document; {numTags} elements found

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/701deb03eade6b98. Report an issue: GitHub.