invoke-ai/InvokeAI · error · TypeError

Expected LlavaOnevisionProcessor, got {type(processor).__nam

Error message

Expected LlavaOnevisionProcessor, got {type(processor).__name__}

What it means

After the model passes its type check, AutoProcessor.from_pretrained is called on the model path and the result is asserted to be a LlavaOnevisionProcessor. AutoProcessor returns whatever class the checkpoint's processor_config/preprocessor_config declares; if the checkpoint ships a generic CLIP/OPT processor or an incompatible config, the isinstance check fails and this TypeError is raised before any inference runs.

Source

Thrown at invokeai/app/api/routers/utilities.py:271

    if task_id is not None:
        events.emit_llm_task_progress(task_id=task_id, user_id=user_id, phase="loading_model", message="Loading model")

    with _model_load_lock:
        loaded_model = model_manager.load.load_model(model_config, user_id=user_id)

    # Load the image from InvokeAI's image store
    image = ApiDependencies.invoker.services.images.get_pil_image(image_name)
    image = image.convert("RGB")

    with torch.no_grad(), loaded_model.model_on_device() as (_, model):
        if not isinstance(model, LlavaOnevisionForConditionalGeneration):
            raise TypeError(f"Expected LlavaOnevisionForConditionalGeneration, got {type(model).__name__}")

        model_abs_path = _resolve_model_path(model_config.path)
        processor = AutoProcessor.from_pretrained(model_abs_path, local_files_only=True)
        if not isinstance(processor, LlavaOnevisionProcessor):
            raise TypeError(f"Expected LlavaOnevisionProcessor, got {type(processor).__name__}")

        pipeline = LlavaOnevisionPipeline(model, processor)
        model_device = next(model.parameters()).device

        progress_callback = _make_progress_callback(events, task_id, user_id)

        output = pipeline.run(
            prompt=instruction,
            images=[image],
            device=model_device,
            dtype=TorchDevice.choose_torch_dtype(),
            progress_callback=progress_callback,
        )

    return output


@utilities_router.post(

View on GitHub (pinned to 0b6a024f2f)

Solutions

Re-download or re-save the model so its directory contains a valid LlavaOnevision processor config (AutoProcessor.save_pretrained from a correct transformers version)
Verify the files at _resolve_model_path(model_config.path) — check preprocessor_config.json / processor_config.json exist and declare llava_onevision
Confirm the installed transformers version matches the version used to create the checkpoint
Clear the partially-downloaded model cache entry and re-fetch

Example fix

# before
processor = AutoProcessor.from_pretrained(model_abs_path, local_files_only=True)
pipeline = LlavaOnevisionPipeline(model, processor)
# after
processor = AutoProcessor.from_pretrained(model_abs_path, local_files_only=True)
if not isinstance(processor, LlavaOnevisionProcessor):
    raise TypeError(f"Expected LlavaOnevisionProcessor, got {type(processor).__name__}")
pipeline = LlavaOnevisionPipeline(model, processor)

Defensive patterns

Strategy: type-guard

Validate before calling

from transformers import AutoProcessor
p = AutoProcessor.from_pretrained(model_path, local_files_only=True)
assert p.__class__.__name__ == "LlavaOnevisionProcessor", f"got {p.__class__.__name__}"

Type guard

def is_llava_onevision_processor(processor) -> bool:
    from transformers import LlavaOnevisionProcessor
    return isinstance(processor, LlavaOnevisionProcessor)

Try / catch

try:
    resp = requests.post(f"{base}/utilities/image_to_prompt", json={...})
    resp.raise_for_status()
except requests.HTTPError as e:
    if "Expected LlavaOnevisionProcessor" in e.response.text:
        re_download_model(model_key)

Prevention

Verify checkpoint directory contains preprocessor_config.json before use
Re-save processors with the same transformers version used at inference
Avoid mixing processors from llava-1.5 conversions into onevision checkpoints

When it happens

Trigger: POST /utilities/image_to_prompt where the LLaVA checkpoint directory lacks a LLaVA OneVision preprocessor_config.json or contains a processor saved by an older/incompatible transformers release, so AutoProcessor instantiates e.g. CLIPImageProcessor+LlamaTokenizer instead of LlavaOnevisionProcessor.

Common situations: Incomplete downloads (missing processor config files), checkpoints converted from llava-1.5 checkpoints without re-saving the processor, local_files_only=True hiding a partially-populated directory, mismatched transformers version saving/loading different processor classes.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/b91c086e8506b3c4. Report an issue: GitHub.