TypeError in formatter.py when value is None

[nmhjklnm]

Reminder

• I have read the README and searched the existing issues.

Reproduction

accelerate launch --config_file config.yaml src/train_bash.py
--stage sft
--do_train
--model_name_or_path ${MODEL_NAME}
--dataset alpaca_zh
--dataloader_persistent_workers
--dataloader_num_workers 8
--template default
--use_unsloth
--finetuning_type ${FINETUNING_TYPE}
--lora_target q_proj,v_proj
--lora_rank 8
--lora_dropout 0.0
--output_dir ${SAVE_DIR}
--overwrite_cache
--per_device_train_batch_size 20
--gradient_accumulation_steps 4
--lr_scheduler_type cosine
--logging_steps 100
--save_steps 3000
--learning_rate 5e-5
--num_train_epochs 3.0
--plot_loss
--fp16

Expected behavior

The Formatter.apply method should handle None values gracefully, either by skipping the replacement or substituting a default value.

System Info

Traceback (most recent call last):
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/train_bash.py", line 14, in
main()
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/train_bash.py", line 5, in main
run_exp()
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/train/tuner.py", line 31, in run_exp
run_sft(model_args, data_args, training_args, finetuning_args, generating_args, callbacks)
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/train/sft/workflow.py", line 32, in run_sft
dataset = get_dataset(tokenizer, model_args, data_args, training_args, stage="sft")
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/data/loader.py", line 180, in get_dataset
dataset = dataset.map(preprocess_func, batched=True, remove_columns=column_names, **kwargs)
File "/nvme/share/share_data/yangyihe/envs/llama_factory/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 592, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/nvme/share/share_data/yangyihe/envs/llama_factory/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 557, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/nvme/share/share_data/yangyihe/envs/llama_factory/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3093, in map
for rank, done, content in Dataset._map_single(**dataset_kwargs):
File "/nvme/share/share_data/yangyihe/envs/llama_factory/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3470, in _map_single
batch = apply_function_on_filtered_inputs(
File "/nvme/share/share_data/yangyihe/envs/llama_factory/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3349, in apply_function_on_filtered_inputs
processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/data/preprocess.py", line 60, in preprocess_supervised_dataset
template.encode_multiturn(
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/data/template.py", line 65, in encode_multiturn
return self._encode(tokenizer, messages, system, tools, cutoff_len, reserved_label_len)
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/data/template.py", line 94, in _encode
elements += self.format_assistant.apply(content=message["content"])
File "/home/zhuwenhui.p/project/LLaMA-Factory/src/llmtuner/data/formatter.py", line 99, in apply
slot = slot.replace("{{" + name + "}}", value, 1)
TypeError: replace() argument 2 must be str, not None

Others

Possible Fix

Modify the Formatter.apply method to check for None and replace it with a default string. Here is a suggested change:

           for slot in self.slots:
            if isinstance(slot, str):
                for name, value in kwargs.items():
                    value = value if value is not None else ""
                    slot = slot.replace("{{" + name + "}}", value, 1)
                elements.append(slot)

[hiyouga]

fixed