invoke-ai/InvokeAI · error · ValueError

Only MistralEncoder_Diffusers_Config models are supported he

Error message

Only MistralEncoder_Diffusers_Config models are supported here.

What it means

MistralEncoderDiffusersLoader._load_model asserts that the config record it was handed is a MistralEncoder_Diffusers_Config before touching config.path. This loader is registered for ModelType.MistralEncoder with format MistralEncoder, so the registry should only route Diffusers-style Mistral encoder records here; a ValueError is raised when some other config subclass (e.g. checkpoint or GGUF config) is passed programmatically or the registry dispatch is bypassed. It is an internal invariant/argument-validation error, not a data-corruption issue.

Source

Thrown at invokeai/backend/model_manager/load/model_loaders/mistral_encoder.py:836

@ModelLoaderRegistry.register(
    base=BaseModelType.Any,
    type=ModelType.MistralEncoder,
    format=ModelFormat.MistralEncoder,
)
class MistralEncoderDiffusersLoader(ModelLoader):
    """Load a Mistral text encoder from a HuggingFace folder layout.

    Handles both the full FLUX.2-dev pipeline layout (with sibling ``tokenizer/``)
    and a standalone download where ``text_encoder/`` files live at the root.
    """

    def _load_model(
        self,
        config: AnyModelConfig,
        submodel_type: Optional[SubModelType] = None,
    ) -> AnyModel:
        if not isinstance(config, MistralEncoder_Diffusers_Config):
            raise ValueError("Only MistralEncoder_Diffusers_Config models are supported here.")

        model_path = Path(config.path)
        text_encoder_path = model_path / "text_encoder"

        # Standalone download: text_encoder files at the root.
        if not text_encoder_path.exists() and (model_path / "config.json").exists():
            text_encoder_path = model_path

        target_device = TorchDevice.choose_torch_device()
        model_dtype = TorchDevice.choose_bfloat16_safe_dtype(target_device)

        match submodel_type:
            case SubModelType.Tokenizer:
                logger = InvokeAILogger.get_logger("MistralEncoderProcessor")
                # Let the multi-strategy loader own the full ladder: embedded Tekken,
                # sibling tokenizer/, root-level processor files, then the HF fallback.
                return _load_tokenizer_for_model(model_path, logger)
            case SubModelType.TextEncoder:

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Pass a config record created for a Diffusers-format Mistral encoder (MistralEncoder_Diffusers_Config, i.e. a folder layout with text_encoder/ or root config.json), not a single-file checkpoint or GGUF.
2. If loading a single .safetensors checkpoint or GGUF file, let the model manager route to MistralEncoderCheckpointLoader / MistralEncoderGGUFLoader instead of calling this loader.
3. Check how the config record was created/imported — re-scan or re-import the model so InvokeAI derives the correct config class from the actual file layout.
4. If you must call the loader directly, wrap the call in isinstance(config, MistralEncoder_Diffusers_Config) before invoking.

Example fix

// before
loader = MistralEncoderDiffusersLoader(...)
model = loader._load_model(checkpoint_cfg, SubModelType.TextEncoder)  # ValueError
// after
from invokeai.backend.model_manager.configs.mistral import MistralEncoder_Diffusers_Config
assert isinstance(cfg, MistralEncoder_Diffusers_Config), "use the checkpoint/GGUF loader for this file"
model = loader._load_model(cfg, SubModelType.TextEncoder)

from invokeai.backend.model_manager.configs.factory import AnyModelConfig
from invokeai.backend.model_manager.configs.mistral import MistralEncoder_Diffusers_Config

def can_load_with_diffusers_loader(cfg: AnyModelConfig) -> bool:
    return isinstance(cfg, MistralEncoder_Diffusers_Config)

def is_mistral_diffusers_config(cfg: AnyModelConfig) -> TypeGuard[MistralEncoder_Diffusers_Config]:
    return isinstance(cfg, MistralEncoder_Diffusers_Config)

try:
    model = loader._load_model(cfg, SubModelType.TextEncoder)
except ValueError as e:
    if "Only MistralEncoder_Diffusers_Config" in str(e):
        model = pick_loader_for_config(cfg)._load_model(cfg, SubModelType.TextEncoder)
    else:
        raise

When it happens

Trigger: Calling MistralEncoderDiffusersLoader._load_model directly with a config that is not MistralEncoder_Diffusers_Config (e.g. a MistralEncoder_Checkpoint_Config or MistralEncoder_GGUF_Config record), or custom code that constructs/injects model configs whose 'type/format' fields resolve to this loader while the config class differs.

Common situations: Custom scripts or plugins that build model config records by hand and load them outside the normal ModelManager install/load flow; a merged or edited models.yaml/DB row whose config class no longer matches its declared format; testing the loader with a mocked AnyModelConfig.

Related errors

• Only MistralEncoder_Checkpoint_Config models are supported h
• Only MistralEncoder_GGUF_Config models are supported here.
• Only Tokenizer and TextEncoder submodels are supported. Rece
• A submodel type must be provided when loading onnx pipelines
• Unrecognised PiD decoder checkpoint extension: {suffix!r}

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/bc017f1f738faa44. Report an issue: GitHub.