hiyouga/LlamaFactory · error · ValueError

Please specify dataset for training.

Error message

Please specify dataset for training.

What it means

Raised in parser.py:478 when training_args.do_train is true but data_args.dataset is None. LlamaFactory trains purely from dataset names registered in data/dataset_info.json, so an empty dataset list leaves the trainer with nothing to consume.

Source

Thrown at src/llamafactory/hparams/parser.py:478

        if finetuning_args.reward_model_type == "lora" and model_args.use_unsloth:
            raise ValueError("Unsloth does not support lora reward model.")

        if training_args.report_to and any(
            logger not in ("wandb", "tensorboard", "trackio", "none") for logger in training_args.report_to
        ):
            raise ValueError("PPO only accepts wandb, tensorboard, or trackio logger.")

    if not model_args.use_kt and training_args.parallel_mode == ParallelMode.NOT_DISTRIBUTED:
        raise ValueError("Please launch distributed training with `llamafactory-cli` or `torchrun`.")

    if training_args.deepspeed and training_args.parallel_mode != ParallelMode.DISTRIBUTED:
        raise ValueError("Please use `FORCE_TORCHRUN=1` to launch DeepSpeed training.")

    if training_args.max_steps == -1 and data_args.streaming:
        raise ValueError("Please specify `max_steps` in streaming mode.")

    if training_args.do_train and data_args.dataset is None:
        raise ValueError("Please specify dataset for training.")

    if (training_args.do_eval or training_args.do_predict or training_args.predict_with_generate) and (
        data_args.eval_dataset is None and data_args.val_size < 1e-6
    ):
        raise ValueError("Please make sure eval_dataset be provided or val_size >1e-6")

    if training_args.predict_with_generate:
        if is_deepspeed_zero3_enabled():
            raise ValueError("`predict_with_generate` is incompatible with DeepSpeed ZeRO-3.")

        if finetuning_args.compute_accuracy:
            raise ValueError("Cannot use `predict_with_generate` and `compute_accuracy` together.")

    if training_args.do_train and model_args.quantization_device_map == "auto":
        raise ValueError("Cannot use device map for quantized models in training.")

    if finetuning_args.pissa_init and is_deepspeed_zero3_enabled():
        raise ValueError("Please use scripts/pissa_init.py to initialize PiSSA in DeepSpeed ZeRO-3.")

View on GitHub (pinned to f28afaf635)

Solutions

1. Add `dataset: <name>` matching an entry in data/dataset_info.json to the config
2. Check spelling of the key — it must be exactly `dataset` (list or string)
3. If the dataset is not registered yet, add its definition to dataset_info.json first, then reference it

Example fix

# before (YAML)
### dataset
# do_train: true, no dataset key

# after
do_train: true
dataset: alpaca_gpt4_zh

if config.get("do_train") and not config.get("dataset"):
    raise SystemExit("do_train: true requires dataset: <name from dataset_info.json>")

When it happens

Trigger: `llamafactory-cli train` with a config missing the `dataset:` key (or dataset: null) while do_train is set; also calling get_train_args on a dict without "dataset" then running run_sft.

Common situations: Typos like `datasets:` or `dataset_name:` instead of `dataset:`; a WebUI-generated YAML where the dataset field was left blank; template variables (e.g. ${DATASET}) expanding to empty in CI.

Related errors

• Unsupported model type: {getattr(config, 'model_type')}.
• Cannot specify `val_size` if `eval_dataset` is not None.
• YAML config must be a dictionary mapping tokens to descripti
• LLaMA-Factory `kt_config` must be a flat mapping.
• Please specify `max_steps` in streaming mode.

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/2680050f76fdf2cd. Report an issue: GitHub.