invoke-ai/InvokeAI · error · TypeError

Expected PreTrainedModel for Gemma encoder, got {type(gemma_

Error message

Expected PreTrainedModel for Gemma encoder, got {type(gemma_encoder).__name__}.

What it means

The Gemma2 text encoder used during FLUX.2 PiD decode must be a transformers PreTrainedModel. The check runs right after model_on_device() context entry; a different class means the loaded encoder file is corrupted, incomplete, or not a Gemma model.

Source

Thrown at invokeai/app/invocations/flux2_pid_decode.py:173

            with vae_info.model_on_device() as (_, vae):
                config = getattr(vae, "config", None)
                if config is not None and hasattr(config, "scaling_factor"):
                    scaling_factor = float(config.scaling_factor)
                    shift_factor = float(getattr(config, "shift_factor", None) or 0.0)
                else:
                    scaling_factor = float(getattr(vae, "scale_factor", scaling_factor))
                    shift_factor = float(getattr(vae, "shift_factor", shift_factor))
            del vae_info
            TorchDevice.empty_cache()

        # 3) Encode caption with Gemma-2.
        gemma_text_encoder_info = context.models.load(self.gemma2_encoder.text_encoder)
        gemma_tokenizer_info = context.models.load(self.gemma2_encoder.tokenizer)
        with ExitStack() as stack:
            (_, gemma_encoder) = stack.enter_context(gemma_text_encoder_info.model_on_device())
            (_, gemma_tokenizer) = stack.enter_context(gemma_tokenizer_info.model_on_device())
            if not isinstance(gemma_encoder, PreTrainedModel):
                raise TypeError(f"Expected PreTrainedModel for Gemma encoder, got {type(gemma_encoder).__name__}.")
            if not isinstance(gemma_tokenizer, PreTrainedTokenizerBase):
                raise TypeError(
                    f"Expected PreTrainedTokenizerBase for Gemma tokenizer, got {type(gemma_tokenizer).__name__}."
                )

            # Encode on the encoder's intended compute device. compute_device honours cpu_only and is
            # stable under partial loading — the first parameter may be offloaded to CPU while later
            # modules load on CUDA, so inferring the device from the first parameter could place caption
            # inputs on the wrong device.
            device = gemma_text_encoder_info.compute_device
            encode_dtype = TorchDevice.choose_bfloat16_safe_dtype(device)
            context.util.signal_progress("Encoding caption with Gemma-2")
            caption_embs, caption_mask = encode_caption_for_pid(
                [self.prompt],
                tokenizer=gemma_tokenizer,
                encoder=gemma_encoder,
                device=device,
                dtype=encode_dtype,

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Re-download the Gemma2 text encoder via the model manager
2. Verify gemma2_encoder.text_encoder points to the correct Gemma model record
3. Update transformers to a compatible version
4. Re-scan/repair model records if hashes are stale

Example fix

// before: text_encoder -> generic llama ckpt
ModelConfig(type='main', path='/models/llama-ckpt/')
// after
ModelConfig(type='main', path='/models/google/gemma-2-2b/', name='Gemma2 encoder')

info = context.models.load(gemma2_encoder.text_encoder)
if not isinstance(info.model, PreTrainedModel):
    raise TypeError(f'Gemma encoder invalid: {type(info.model).__name__}')

from transformers import PreTrainedModel

def is_gemma_encoder(obj) -> bool:
    return isinstance(obj, PreTrainedModel)

try:
    result = pid_decode.invoke(context)
except TypeError as e:
    if 'Gemma encoder' in str(e):
        reimport_model_manager_entry(gemma2_encoder.text_encoder)
    raise

When it happens

Trigger: gemma_text_encoder_info.model_on_device() yields an object that is not PreTrainedModel in the invoke ExitStack; the gemma2_encoder.text_encoder field references a wrong or damaged model record.

Common situations: Interrupted downloads of Gemma encoder weights; model config pointing at a non-Gemma directory; transformers version changes changing the wrapper class; converted/quantized models loaded as custom classes.

Related errors

• Expected PreTrainedModel for text encoder, got {type(text_en
• Expected LlavaOnevisionForConditionalGeneration, got {type(m
• Expected PreTrainedTokenizerBase for tokenizer, got {type(to
• Expected ModelPatchRaw for LoRA '{lora.lora.key}', got {type
• Expected PreTrainedTokenizerBase for Gemma tokenizer, got {t

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/4a19960a45539e31. Report an issue: GitHub.