invoke-ai/InvokeAI · warning · HTTPException

str(e)

Error message

str(e)

What it means

This is a 422 Unprocessable Entity raised by the PATCH/PUT runtime-config endpoint when the requested generation devices (e.g. torch device strings like 'cuda:99') cannot be resolved by TorchDevice.get_generation_devices. InvokeAI pre-validates device changes with the same resolution logic as the startup path so a bad config can never be persisted and break the next startup. The detail is str(e) from the underlying ValueError, typically naming the invalid device.

Source

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

@app_router.patch(
    "/runtime_config",
    operation_id="update_runtime_config",
    status_code=200,
    response_model=InvokeAIAppConfigWithSetFields,
)
def update_runtime_config(
    _: AdminUserOrDefault,
    changes: UpdateAppGenerationSettingsRequest = Body(description="Writable runtime configuration changes"),
) -> InvokeAIAppConfigWithSetFields:
    # The request model validates the *shape* of generation_devices; also verify the devices exist
    # on this machine before persisting, so we can't write a config that fails on the next startup
    # (e.g. 'cuda:99' on a 2-GPU box). Same resolution the startup path uses.
    if changes.generation_devices is not None:
        try:
            TorchDevice.get_generation_devices(changes.generation_devices)
        except ValueError as e:
            raise HTTPException(status_code=422, detail=str(e))
    with _EXTERNAL_PROVIDER_CONFIG_LOCK:
        config = get_config()
        update_dict = changes.model_dump(exclude_unset=True)
        config.update_config(update_dict)

        if config.config_file_path.exists():
            persisted_config = load_and_migrate_config(config.config_file_path)
        else:
            persisted_config = DefaultInvokeAIAppConfig()

        persisted_config.update_config(update_dict)
        persisted_config.write_file(config.config_file_path)
        return InvokeAIAppConfigWithSetFields(set_fields=config.model_fields_set, config=_redact_config_secrets(config))


@app_router.get(
    "/external_providers/status",
    operation_id="get_external_provider_statuses",

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Read the detail message to identify the offending device string
2. Use a valid torch device available on this machine (e.g. 'cuda:0', 'mps', 'cpu'); verify with python -c "import torch; print(torch.cuda.device_count())"
3. Send only the fields you intend to change (exclude_unset) instead of echoing devices copied from another machine
4. Remove the generation_devices key from the request if you don't mean to change devices

Example fix

// before
{ "generation_devices": { "denoising": "cuda:99", "clip": "cuda:99" } }
// after
{ "generation_devices": { "denoising": "cuda:0", "clip": "cuda:0" } }

import torch
from invokeai.backend.util.devices import TorchDevice

def validate_generation_devices(devices: dict) -> None:
    # mirror the server's pre-check before sending the config
    TorchDevice.get_generation_devices(devices)  # raises ValueError if invalid

def is_valid_device(name: str) -> bool:
    import torch
    if name == 'auto':
        return True
    base = name.split(':')[0]
    if base == 'cuda':
        idx = int(name.split(':')[1]) if ':' in name else 0
        return torch.cuda.is_available() and idx < torch.cuda.device_count()
    return base in ('mps', 'cpu')

try:
    resp = requests.patch(f'{base}/app/config', json=payload)
    resp.raise_for_status()
except requests.HTTPError as e:
    if e.response.status_code == 422:
        print('Invalid device:', e.response.json()['detail'])

When it happens

Trigger: PATCHing runtime config with generation_devices containing an invalid/unavailable torch device string (typo, nonexistent CUDA index like 'cuda:99' on a 2-GPU box, CPU-only machine given 'cuda:0').

Common situations: Copying config from a multi-GPU machine to a smaller box; typo in device name ('cuda' vs 'cuda:0' vs 'gpu'); Docker container without GPU passthrough; driver/CUDA mismatch making a device invisible to torch.

Related errors

• Multiuser mode is disabled. Authentication is not required i
• Multiuser mode is disabled. Admin setup is not required in s
• Invalid regex: {e}
• Invalid generation_devices value '{v}'. Use 'auto' or a list
• generation_devices cannot be an empty list. Use 'auto' or a

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/17ff695dbfc01021. Report an issue: GitHub.