invoke-ai/InvokeAI · warning · NotAMatchError

expected a .safetensors file, got {mod.path.suffix or '(no s

Error message

expected a .safetensors file, got {mod.path.suffix or '(no suffix)'}

What it means

`NotAMatchError` from the single-file branch of `QwenVLTextEncoderConfig.from_model_on_disk`: the candidate file is not a `.safetensors` file, so the loader rejects it before reading any keys. Only safetensors checkpoints are supported for single-file Qwen VL encoders; other extensions (`.bin`, `.pt`, `.ckpt`, no suffix) cannot be inspected cheaply or safely.

Source

Thrown at invokeai/backend/model_manager/configs/qwen_vl_encoder.py:142

    The matching tokenizer + processor are pulled from HuggingFace
    (`Qwen/Qwen2.5-VL-7B-Instruct`) on first use and cached for offline use.
    """

    base: Literal[BaseModelType.Any] = Field(default=BaseModelType.Any)
    type: Literal[ModelType.QwenVLEncoder] = Field(default=ModelType.QwenVLEncoder)
    format: Literal[ModelFormat.Checkpoint] = Field(default=ModelFormat.Checkpoint)

    @classmethod
    def from_model_on_disk(cls, mod: ModelOnDisk, override_fields: dict[str, Any]) -> Self:
        raise_if_not_file(mod)

        raise_for_override_fields(cls, override_fields)

        # Only safetensors checkpoints are supported as single-file Qwen VL encoders.
        # Reject other extensions cheaply before attempting to read keys.
        if mod.path.suffix != ".safetensors":
            raise NotAMatchError(f"expected a .safetensors file, got {mod.path.suffix or '(no suffix)'}")

        # Read only the key index — a 7GB fp8 encoder weighs ~7GB on disk, but we
        # only need the key names to classify it, not the tensor data.
        try:
            keys = _read_safetensors_keys(mod.path)
        except Exception as e:
            raise NotAMatchError(f"could not read safetensors header: {e}") from e

        if not _has_qwen_vl_keys(keys):
            raise NotAMatchError("state dict does not look like a Qwen2.5-VL/Qwen2-VL checkpoint")

        return cls(**override_fields)

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Convert the checkpoint to safetensors (e.g. via `safetensors.torch.save_file` or a conversion script) and install the `.safetensors` file
2. Download the safetensors variant from the model repo instead of the `.bin` variant
3. Ensure the file keeps its `.safetensors` extension (don't strip it when renaming)
4. If the path is actually a directory-based model, let the folder-branch of the config handle it instead of pointing at a loose file

Example fix

// before
mv model.bin qwen_encoder.safetensors   # wrong: just renaming
// after
python -c "from safetensors.torch import load_file,save_file; ..."  # actually convert, keep .safetensors

from pathlib import Path

def is_safetensors_file(path: Path) -> bool:
    return path.is_file() and path.suffix == ".safetensors"

def is_single_file_safetensors(mod_path: Path) -> bool:
    return mod_path.is_file() and mod_path.suffix == ".safetensors"

try:
    cfg = QwenVLTextEncoderConfig.from_model_on_disk(mod, override_fields)
except NotAMatchError as e:
    if "expected a .safetensors file" in str(e):
        print("Convert or re-download the checkpoint in safetensors format")
    raise

When it happens

Trigger: Installing a single-file Qwen VL encoder whose `ModelOnDisk.path` suffix is not `.safetensors` — e.g. a PyTorch `.bin`/`.pt` checkpoint, an archive, or a directory-less file with no extension.

Common situations: Old-style torch pickle checkpoints, config files placed at the model root and probed as weights, renamed downloads that lost the extension, GGUF or fp8 files with custom extensions.

Related errors

• state dict does not look like a Qwen2.5-VL/Qwen2-VL checkpoi
• The {model_name} model must be a Diffusers format model. The
• The {model_name} model must be a Diffusers-style FLUX.2 pipe
• Unsupported model format: {config.format}
• To extract the VAE and Qwen3-VL encoder, the {model_name} mo

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/4d57f328cb34325e. Report an issue: GitHub.