ValueError: Unknown split "330k_train". Should be one of ['train', 'test'].

[Chase-Xuu]

Hello, I have faced a problem and have not seen it in the issues.

(llm_unlearn) root@5e0bce17bb64:/home/chixu/ELEC631/llm_unlearn# python unlearn_harm.py --model_name=facebook/opt-1.3b --model_save_dir=models/opt1.3b_unlearned --log_file=logs/opt-1.3b-unlearn.log
Traceback (most recent call last):
File "/home/chixu/ELEC631/llm_unlearn/unlearn_harm.py", line 228, in
main(args)
File "/home/chixu/ELEC631/llm_unlearn/unlearn_harm.py", line 56, in main
train_dataset = load_dataset("PKU-Alignment/PKU-SafeRLHF", split="330k_train")
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/load.py", line 2166, in load_dataset
ds = builder_instance.as_dataset(split=split, verification_mode=verification_mode, in_memory=keep_in_memory)
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/builder.py", line 1190, in as_dataset
datasets = map_nested(
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/utils/py_utils.py", line 456, in map_nested
return function(data_struct)
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/builder.py", line 1220, in _build_single_dataset
ds = self._as_dataset(
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/builder.py", line 1294, in _as_dataset
dataset_kwargs = ArrowReader(cache_dir, self.info).read(
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/arrow_reader.py", line 240, in read
files = self.get_file_instructions(name, instructions, split_infos)
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/arrow_reader.py", line 213, in get_file_instructions
file_instructions = make_file_instructions(
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/arrow_reader.py", line 130, in make_file_instructions
absolute_instructions = instruction.to_absolute(name2len)
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/arrow_reader.py", line 653, in to_absolute
return [_rel_to_abs_instr(rel_instr, name2len) for rel_instr in self._relative_instructions]
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/arrow_reader.py", line 653, in
return [_rel_to_abs_instr(rel_instr, name2len) for rel_instr in self._relative_instructions]
File "/home/chixu/anaconda3/envs/llm_unlearn/lib/python3.10/site-packages/datasets/arrow_reader.py", line 465, in _rel_to_abs_instr
raise ValueError(f'Unknown split "{split}". Should be one of {list(name2len)}.')
ValueError: Unknown split "330k_train". Should be one of ['train', 'test'].

[uyenlk]

I had the same problem. Go to the unlearn_harm.py file and change the split from '330k_train' to 'train'.

[moro0v0]

BRO，I checked and tested the results myself, and found that firstly, bad_rass kept jumping between 2 and 3 during the training process, without any significant increase. In addition, I also found that using its template itself does not output content for the same issue. Excuse me, are you still researching this issue? Can you provide contact information for communication? thank you!

[Zhang-byte-plus]

I had the same problem. Go to the unlearn_harm.py file and change the split from '330k_train' to 'train'.

Hey bro! I followed your suggestion to change the split from '330k_train' to 'train' in the unlearn_harm.py file, then retried the run—but it failed again. Are you still looking into this issue? If you fixed it, could you let me know how you did it?

[moro0v0]

this project is shit cant work at all.thx

…

---Original--- From: ***@***.***> Date: Sun, Jan 25, 2026 17:52 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [kevinyaobytedance/llm_unlearn] ValueError: Unknown split"330k_train". Should be one of ['train', 'test']. (Issue #7) Zhang-byte-plus left a comment (kevinyaobytedance/llm_unlearn#7) I had the same problem. Go to the unlearn_harm.py file and change the split from '330k_train' to 'train'. Hey bro! I followed your suggestion to change the split from '330k_train' to 'train' in the unlearn_harm.py file, then retried the run—but it failed again. Are you still looking into this issue? If you fixed it, could you let me know how you did it? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>