mudler/LocalAI · error · Exception

Model not loaded. Call LoadModel first.

Error message

Model not loaded. Call LoadModel first.

What it means

Raised inside AudioTranscription of the moonshine backend when a transcription request arrives before any successful LoadModel call has set self.transcriber. It is a plain Exception (not a typed error) raised deliberately to be caught by the surrounding try block and returned as a failed gRPC Result.

Source

Thrown at backend/python/moonshine/backend.py:88

                model_arch = self.options["model_arch"]
            
            # Get the model path and architecture
            model_path, model_arch = get_model_for_language(language, model_arch)
            print(f"Loading model: {model_path} with architecture: {model_arch} for language: {language}", file=sys.stderr)
            
            # Initialize the transcriber
            self.transcriber = Transcriber(model_path=model_path, model_arch=model_arch)
            print("Model loaded successfully", file=sys.stderr)
        except Exception as err:
            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
        return backend_pb2.Result(message="Model loaded successfully", success=True)

    def AudioTranscription(self, request, context):
        resultSegments = []
        text = ""
        try:
            if self.transcriber is None:
                raise Exception("Model not loaded. Call LoadModel first.")
            
            # Load the audio file
            audio_data, sample_rate = load_wav_file(request.dst)
            print(f"Loaded audio file: {request.dst} with sample rate: {sample_rate}", file=sys.stderr)
            
            # Transcribe without streaming
            transcript = self.transcriber.transcribe_without_streaming(
                audio_data, sample_rate=sample_rate, flags=0
            )
            
            # Process transcript lines
            full_text_parts = []
            for idx, line in enumerate(transcript.lines):
                line_text = line.text.strip()
                full_text_parts.append(line_text)
                
                # Create segment with timing information
                start_ms = int(line.start_time * 1000)

View on GitHub (pinned to 44413a9d06)

Solutions

1. Call LoadModel with a valid model_path/model_arch and confirm success=true before transcribing
2. Check the LoadModel Result message — a prior failure like a missing model file leaves transcriber None
3. Verify the model directory exists and the moonshine model arch string is correct, then retry the load

Example fix

# before
stub.AudioTranscription(req)  # 'Model not loaded. Call LoadModel first.'

# after
res = stub.LoadModel(load_req)
assert res.success, res.message
stub.AudioTranscription(req)

# gRPC clients: ensure LoadModel succeeded first
load_resp = stub.LoadModel(backend_pb2.ModelOptions(model=model_path))
if not load_resp.success:
    raise RuntimeError(f'LoadModel failed: {load_resp.message}')
# only now transcribe

def backend_ready(servicer) -> bool:
    return servicer.transcriber is not None

try:
    result = stub.AudioTranscription(req)
except grpc.RpcError as err:
    if 'Model not loaded' in (err.details() or ''):
        load_and_retry()  # explicit recovery, not silent fallback
    else:
        raise

When it happens

Trigger: Calling the AudioTranscription RPC on a fresh backend instance that never loaded a model, or after a LoadModel that failed (e.g. bad model path/arch) leaving self.transcriber as None.

Common situations: Startup ordering issues where the client sends transcription before load completes, failed loads being ignored by orchestration code, or health checks not distinguishing loaded vs unloaded state.

Related errors

• audio is required for LongCat-Video-Avatar-1.5
• audio input is not a readable staged file
• Model not loaded
• dataset_source is required (path to a preprocessed dataset)
• start_image is not a readable staged file

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/c9a7a0d437466f6f. Report an issue: GitHub.