sgl-project/sglang · error · RuntimeError

Completion template {completion_template_arg} is not a built

Error message

Completion template {completion_template_arg} is not a built-in template name or a valid completion template file path.

What it means

Same pattern as the chat-template case but for --completion-template: load_completion_template accepts either a registered builtin completion template name or a file path; when the name isn't builtin and the path doesn't exist, it raises RuntimeError.

Source

Thrown at python/sglang/srt/parser/template_manager.py:230

            model_path: Path to the model
        """
        template_name = get_conv_template_by_model_path(model_path)
        if template_name is not None:
            logger.info(f"Inferred chat template from model path: {template_name}")
            self._chat_template_name = template_name

    def load_completion_template(self, completion_template_arg: str) -> None:
        """
        Load completion template for code completion.

        Args:
            completion_template_arg: Template name or file path
        """
        logger.info(f"Loading completion template: {completion_template_arg}")

        if not completion_template_exists(completion_template_arg):
            if not os.path.exists(completion_template_arg):
                raise RuntimeError(
                    f"Completion template {completion_template_arg} is not a built-in template name "
                    "or a valid completion template file path."
                )

            self._load_json_completion_template(completion_template_arg)
        else:
            self._completion_template_name = completion_template_arg

        set_completion_template(self._completion_template_name)

    def initialize_templates(
        self,
        tokenizer_manager: TokenizerManager,
        model_path: str,
        chat_template: Optional[str] = None,
        completion_template: Optional[str] = None,
    ) -> None:
        """

View on GitHub (pinned to 0132848349)

Solutions

1. Check builtin completion template names and use one exactly
2. Otherwise pass an absolute path to your JSON/jinja file and confirm it exists (ls)
3. Mount the file into the container or bake it into the image

Example fix

# before
python -m sglang.launch_server --completion-template tmpl.json
# after
python -m sglang.launch_server --completion-template /app/templates/tmpl.json

import os
ok = completion_template_exists(arg) or os.path.isfile(os.path.abspath(arg))
if not ok:
    raise FileNotFoundError(arg)

def is_loadable_completion_template(arg: str) -> TypeGuard[str]:
    import os
    from sglang.srt.parser.template_manager import completion_template_exists
    return completion_template_exists(arg) or os.path.isfile(arg)

try:
    tm.load_completion_template(completion_template_arg)
except RuntimeError as e:
    if "Completion template" in str(e):
        sys.exit(f"bad --completion-template: {completion_template_arg}")
    raise

When it happens

Trigger: Passing --completion-template base (not builtin) with no file named base in the working directory, or a JSON completion-template path that's mistyped or not mounted.

Common situations: Custom JSON completion template files not shipped in the Docker image, wrong relative path, or a builtin-name typo.

Related errors

• Chat template {chat_template_arg} is not a built-in template
• metal shader source not found: {metal_src}
• MiniMax-H3 on MPS requires synchronous layerwise offload for
• MiniMax-H3 MPS execution does not support torch.compile; pas
• MiniMax-H3 ring parallelism requires the FlashAttention back

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/9c0a19a1513c881b. Report an issue: GitHub.