hiyouga/LlamaFactory · error · ValueError

Unsupported model type: {getattr(config, 'model_type')}.

Error message

Unsupported model type: {getattr(config, 'model_type')}.

What it means

get_train_dataloader in the HyperParallel trainer (trainer.py:311) mirrors the upstream transformers behavior: if self.train_dataset is None it cannot build a dataloader and raises ValueError('Trainer: training requires a train_dataset.'). In LlamaFactory this means the YAML stage/dataset wiring produced no dataset before the trainer was constructed.

Source

Thrown at scripts/qwen_omni_merge.py:117

    Args:
        model_path (str): Directory path of the original model.
        thinker_path (str): Path to the saved thinker weights.
        save_path (str): Directory where the merged model and configurations will be saved.
        extra_file (str): Name of the extra file to be copied (default: "spk_dict.pt").
    """
    # 1. Load the saved thinker module and the original model
    config = AutoConfig.from_pretrained(model_path)
    if getattr(config, "model_type") == "qwen2_5_omni":
        from transformers.models.qwen2_5_omni import Qwen2_5OmniThinkerForConditionalGeneration  # type: ignore

        ThinkerClass = Qwen2_5OmniThinkerForConditionalGeneration
    elif getattr(config, "model_type") == "qwen3_omni_moe":
        from transformers.models.qwen3_omni_moe import Qwen3OmniMoeThinkerForConditionalGeneration  # type: ignore

        ThinkerClass = Qwen3OmniMoeThinkerForConditionalGeneration
    else:
        raise ValueError(f"Unsupported model type: {getattr(config, 'model_type')}.")

    thinker = ThinkerClass.from_pretrained(thinker_path, torch_dtype="auto", device_map="cpu")
    base_model = AutoModelForTextToWaveform.from_pretrained(model_path, torch_dtype="auto", device_map="cpu")
    base_model.thinker = thinker
    processor = AutoProcessor.from_pretrained(thinker_path)
    print("Successfully loaded model weights and processor.")

    # 2. Save the complete model along with its tokenizer and processor configuration
    base_model.save_pretrained(save_path)
    processor.save_pretrained(save_path)
    print(f"Merged model and processor saved to {save_path}.")

    # 3. Copy the extra file from the base model directory to the save_path
    try:
        source_file = cached_file(path_or_repo_id=model_path, filename=extra_file)
        shutil.copy(source_file, os.path.join(save_path, extra_file))
        print(f"File '{extra_file}' copied from {model_path} to {save_path}.")
    except Exception:

View on GitHub (pinned to f28afaf635)

Solutions

1. Add a valid `dataset: <name>` entry (present in data/dataset_info.json) to the training YAML
2. Check for typos in the dataset name and that dataset_dir points at the right dataset_info.json
3. If you meant preprocessing-only or no data, use a different workflow — the HP trainer cannot run without a train dataset

Example fix

# before (YAML)
### dataset
# dataset: (missing)

# after (YAML)
dataset: alpaca_gpt4_zh  # must exist in data/dataset_info.json

assert data_args.dataset, 'No dataset configured: add a `dataset:` entry that exists in data/dataset_info.json'
assert train_dataset is not None, 'train_dataset resolved to None; check dataset names and dataset_dir'

def has_train_dataset(data_args) -> bool:
    return bool(getattr(data_args, 'dataset', None))

try:
    trainer.train()
except ValueError as e:
    if 'requires a train_dataset' in str(e):
        raise SystemExit('Add dataset: <name> (from data/dataset_info.json) to the YAML') from e
    raise

When it happens

Trigger: Training YAML omits `dataset:` / `dataset_dir:` so DataArguments carries no datasets; or the stage (e.g. pt/sft) resolved to an empty dataset list and the trainer was instantiated with train_dataset=None.

Common situations: Reusing a chat/eval YAML for a HyperParallel train run and forgetting the dataset key; dataset name typo causing zero matching datasets; intentionally testing trainer construction without data.

Related errors

• Cannot specify `val_size` if `eval_dataset` is not None.
• The model does not have a submodule named '{submodule_name}'
• Template is required for MultiModalDataCollator.
• Dataset converter {name} not found.
• Unknown mixing strategy: {data_args.mix_strategy}.

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/d70aed2fe81a050d. Report an issue: GitHub.