invoke-ai/InvokeAI · error · NotAMatchError

directory looks like a full diffusers pipeline, not a standa

Error message

directory looks like a full diffusers pipeline, not a standalone Gemma2 encoder

What it means

After config.json exists, the prober rejects directories that also contain model_index.json at the root, since that marks a full diffusers pipeline (UNet+VAE+encoders together) rather than the standalone Gemma2 encoder InvokeAI requires for PiD caption projection. Probing such a pipeline raises NotAMatchError.

Source

Thrown at invokeai/backend/model_manager/configs/gemma2_encoder.py:67

    """

    base: Literal[BaseModelType.Any] = Field(default=BaseModelType.Any)
    type: Literal[ModelType.Gemma2Encoder] = Field(default=ModelType.Gemma2Encoder)
    format: Literal[ModelFormat.Gemma2Encoder] = Field(default=ModelFormat.Gemma2Encoder)
    cpu_only: bool | None = Field(default=None, description="Whether this model should run on CPU only")

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

        config_path = mod.path / "config.json"
        if not config_path.exists():
            raise NotAMatchError(f"missing config.json at {config_path}")

        # Reject full diffusers pipelines (they have model_index.json at root).
        if (mod.path / "model_index.json").exists():
            raise NotAMatchError("directory looks like a full diffusers pipeline, not a standalone Gemma2 encoder")

        # Architecture marker is the canonical signal.
        raise_for_class_name(config_path, {"Gemma2ForCausalLM"})

        # Only Gemma-2-2b (2304-dim hidden state) is compatible with PiD's fixed caption projection.
        # Reject 9B/27B variants here so they are not offered as compatible encoders and then fail with
        # a matrix-shape error deep inside PiD inference.
        hidden_size = get_config_dict_or_raise(config_path).get("hidden_size")
        if hidden_size != _PID_GEMMA_HIDDEN_SIZE:
            raise NotAMatchError(
                f"Gemma2 hidden_size {hidden_size} is incompatible with PiD, which requires "
                f"{_PID_GEMMA_HIDDEN_SIZE} (Gemma-2-2b); 9B/27B variants are not supported."
            )

        # Sanity check that tokenizer files live alongside the model (PiD calls
        # AutoTokenizer.from_pretrained on the same directory).
        if not any((mod.path / f).exists() for f in ("tokenizer.json", "tokenizer.model")):
            raise NotAMatchError("directory does not contain Gemma2 tokenizer files (tokenizer.json/tokenizer.model)")

View on GitHub (pinned to 0b6a024f2f)

Solutions

Import only the text_encoder/ subdirectory of the pipeline, not the repo root
Remove model_index.json only if you truly isolated the encoder files into their own directory (otherwise keep the pipeline elsewhere)
Download the standalone Gemma2 model and import its top-level folder directly
Keep full pipelines outside InvokeAI's models directory to avoid component mis-probing

Example fix

// before
install_model("/models/gemma-2-2b-it")            # full pipeline: has model_index.json
// after
install_model("/models/gemma-2-2b-it/text_encoder")  # standalone encoder dir

Defensive patterns

Strategy: validation

Validate before calling

import pathlib

def is_full_diffusers_pipeline(path: str) -> bool:
    p = pathlib.Path(path)
    return (p / "model_index.json").exists()

def encoder_import_path(path: str) -> str:
    p = pathlib.Path(path)
    return str(p / "text_encoder") if is_full_diffusers_pipeline(str(p)) else str(p)

Try / catch

try:
    config = probe(mod)
except NotAMatchError as e:
    if "full diffusers pipeline" in str(e):
        encoder = pathlib.Path(mod.path) / "text_encoder"
        config = probe(encoder)
    else:
        raise

Prevention

Never import full pipeline snapshots as single models
Point at the text_encoder subdirectory of pipelines
Keep full pipelines outside the InvokeAI models directory
Check for model_index.json before import to detect pipelines

When it happens

Trigger: from_model_on_disk pointed at a downloaded diffusers repo root (e.g. a full pipeline snapshot with model_index.json) instead of the text_encoder subfolder; or copying the whole HF snapshot into the models directory.

Common situations: Downloading gemma-2-2b-it as a full pipeline snapshot and importing the whole folder; pointing InvokeAI at the repo root rather than the encoder component directory.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/d618259877a5b16e. Report an issue: GitHub.