sgl-project/sglang · error · EntryNotFoundError

{exc}

Error message

{exc}

What it means

hf_hub_download wraps a ModelScope NotExistError into EntryNotFoundError to keep the Hugging Face-compatible exception contract: the requested file does not exist in the given repository. This mirrors huggingface_hub.EntryNotFoundError so callers can handle missing files uniformly across hubs.

Source

Thrown at python/sglang/multimodal_gen/runtime/utils/hf_diffusers_utils.py:1161

    local_dir: Optional[Union[str, Path]] = None,
    **kwargs,
) -> str:
    """Unified hf_hub_download that supports both Hugging Face Hub and ModelScope."""
    if envs.SGLANG_USE_MODELSCOPE.get():
        from modelscope import model_file_download
        from modelscope.hub.errors import NotExistError

        try:
            return model_file_download(
                model_id=repo_id,
                file_path=filename,
                local_dir=str(local_dir) if local_dir is not None else None,
                **kwargs,
            )
        except NotExistError as exc:
            # Keep the Hugging Face-compatible exception contract used by
            # maybe_download_model_index for repositories without model_index.json.
            raise EntryNotFoundError(str(exc)) from exc
    else:
        from huggingface_hub import hf_hub_download as _hf_hub_download

        return _hf_hub_download(
            repo_id=repo_id,
            filename=filename,
            local_dir=local_dir,
            **kwargs,
        )


def snapshot_download(
    repo_id: str,
    local_dir: Optional[Union[str, Path]] = None,
    ignore_patterns: Optional[Union[list[str], str]] = None,
    allow_patterns: Optional[Union[list[str], str]] = None,
    local_files_only: bool = False,
    max_workers: int = 8,

View on GitHub (pinned to 0132848349)

Solutions

1. Verify the exact filename (and subfolder) exists in the repo with huggingface-cli ls-files or the web UI
2. If the file is optional, catch EntryNotFoundError and fall back to the next candidate
3. Use the correct repo variant (e.g. the diffusion repo rather than the base LLM repo)
4. If the file is sharded, resolve the shard list from the index instead of a single filename

Example fix

// before
f = hf_hub_download(repo_id, "model_index.json")
// after
try:
    f = hf_hub_download(repo_id, "model_index.json")
except EntryNotFoundError:
    f = None  # not a diffusers-style repo

from huggingface_hub import HfApi
files = set(HfApi().list_repo_files(repo_id))
if filename not in files:
    # choose fallback before calling

try:
    path = hf_hub_download(repo_id, filename)
except EntryNotFoundError:
    path = fallback_resolution(repo_id)

When it happens

Trigger: Calling hf_hub_download (directly or via get_is_diffusion_model, download_and_cache_hf_file, resolve_transformer_safetensors_to_load) with a repo_id/filename combination where the file is absent — e.g. looking for model_index.json in a non-diffusers repo, or a wrong filename/subfolder.

Common situations: Probing repos for optional files (model_index.json, config.json) with try/except, wrong subfolder paths for sharded weights, renamed files upstream, repos that are pure transformers vs diffusers layouts.

Related errors

• No cached files for {repo_id} match {allow_patterns or '**/*
• Weight file {source.filename!r} was not found in {source.rep
• No model weights found in {path} (expected model.safetensors
• metal shader source not found: {metal_src}
• Prompt text file not found: {path}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/32e283532c29f11e. Report an issue: GitHub.