mudler/LocalAI · error · ValueError

Invalid scheduler '{'k_' if is_karras else ''}{name}'

Error message

Invalid scheduler '{'k_' if is_karras else ''}{name}'

What it means

Raised by the diffusers backend's scheduler factory when the requested scheduler name does not match any branch of the if/elif chain mapping DiffusionScheduler enum values to diffusers scheduler classes. The message echoes the name with a 'k_' prefix when K-Diffusion Karras scheduling was requested.

Source

Thrown at backend/python/diffusers/backend.py:211

        # Equivalent to DPM2 in K-Diffusion
        sched_class = KDPM2DiscreteScheduler
    elif name == DiffusionScheduler.dpm_2_a:
        # Equivalent to `DPM2 a`` in K-Diffusion
        sched_class = KDPM2AncestralDiscreteScheduler
    elif name == DiffusionScheduler.dpmpp_2m:
        # Equivalent to `DPM++ 2M` in K-Diffusion
        sched_class = DPMSolverMultistepScheduler
        config["algorithm_type"] = "dpmsolver++"
        config["solver_order"] = 2
    elif name == DiffusionScheduler.dpmpp_sde:
        # Equivalent to `DPM++ SDE` in K-Diffusion
        sched_class = DPMSolverSinglestepScheduler
    elif name == DiffusionScheduler.dpmpp_2m_sde:
        # Equivalent to `DPM++ 2M SDE` in K-Diffusion
        sched_class = DPMSolverMultistepScheduler
        config["algorithm_type"] = "sde-dpmsolver++"
    else:
        raise ValueError(f"Invalid scheduler '{'k_' if is_karras else ''}{name}'")

    return sched_class.from_config(config)


# Implement the BackendServicer class with the service methods
class BackendServicer(backend_pb2_grpc.BackendServicer):

    def _load_pipeline(self, request, model_ref, from_single_file, local_only, torchType, variant, device_map=None):
        """
        Load a diffusers pipeline dynamically using the dynamic loader.

        This method uses load_diffusers_pipeline() for most pipelines, falling back
        to explicit handling only for pipelines requiring custom initialization
        (e.g., quantization, special VAE handling).

        Args:
            request: The gRPC request containing pipeline configuration
            model_ref: Repository ID or local model file/directory

View on GitHub (pinned to 44413a9d06)

Solutions

1. Use one of the supported scheduler names visible in the if/elif chain (e.g. dpmpp_2m, dpmpp_sde, dpmpp_2m_sde and the other branches above line 211).
2. Update the backend image/repo so newly mapped schedulers are available.
3. Check the exact enum value being sent — log request.Scheduler before the mapping; compare against the DiffusionScheduler enum definition.
4. If a Karras variant was intended, verify the correct base name is passed and the is_karras flag is set rather than baking 'k_' into the name.

Example fix

# before
request.Scheduler = "dpm++_2m_karras"  # typo / raw KD name

# after
request.Scheduler = "dpmpp_2m"  # matches DiffusionScheduler.dpmpp_2m branch

valid_schedulers = {s.value for s in DiffusionScheduler}
assert request.Scheduler in valid_schedulers, f"unsupported scheduler {request.Scheduler!r}"

def is_supported_scheduler(name: str) -> bool:
    return name in {s.value for s in DiffusionScheduler}

try:
    scheduler = build_scheduler(name, is_karras)
except ValueError as e:
    return error_reply(f"{e}; supported: {sorted(s.value for s in DiffusionScheduler)}")

When it happens

Trigger: Requesting a scheduler value that the backend's mapping table does not cover — e.g. a newly added or experimental DiffusionScheduler member, or a raw string scheduler name passed through from the request that is not in the enum branches.

Common situations: Version mismatch: the request (or a newer LocalAI frontend) uses a scheduler name this backend build predates; a typo in the scheduler field of the request; or a K-Diffusion scheduler name (e.g. 'k_dpmpp_2m') passed where the diffusers backend expects the non-Karras enum spelling.

Related errors

• Unknown pipeline class '{class_name}'. Available pipelines:
• Unknown task '{task}'. Available tasks: {', '.join(sorted(al
• Failed to load pipeline '{effective_pipeline_type}': {e}\nAv
• Unknown pipeline: {class_name}
• Invalid scheduler, using default: %s\n

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/4b2723378757f191. Report an issue: GitHub.