Unrecognized arguments base_dir and use_gpt2_tokenizer about gpt-neo HOT 5 CLOSED

Same here. Modified the command like this:
!python data/create_tfrecords.py --mode documents --input_dir /content/GPTNeo/$dataset_path --name $dataset_name --output_dir $out_name --write_dataset_config

The command fails with this error:
Traceback (most recent call last): File "data/create_tfrecords.py", line 205, in <module> results = create_tfrecords_mp(files, args) File "data/create_tfrecords.py", line 186, in create_tfrecords_mp files = split_list(files, len(files) // args.processes) File "data/create_tfrecords.py", line 67, in split_list return [l[i:i+n] for i in range(0, len(l), n)] ValueError: range() arg 3 must not be zero

This happens when the number of documents in the dataset is smaller than the amount of CPUs you have. Set the argument --processes 1 to address this

from gpt-neo.

Same here. Modified the command like this:
!python data/create_tfrecords.py --mode documents --input_dir /content/GPTNeo/$dataset_path --name $dataset_name --output_dir $out_name --write_dataset_config

The command fails with this error:
Traceback (most recent call last): File "data/create_tfrecords.py", line 205, in <module> results = create_tfrecords_mp(files, args) File "data/create_tfrecords.py", line 186, in create_tfrecords_mp files = split_list(files, len(files) // args.processes) File "data/create_tfrecords.py", line 67, in split_list return [l[i:i+n] for i in range(0, len(l), n)] ValueError: range() arg 3 must not be zero

As explained by @322997am this is an unrelated error and is caused by the fact that you have more CPUs trying to create TFRecords than you have documents. You should either increase the number of documents or only use some of your CPUs to create TFRecords. The flag --processes n allows you to limit the number of parallel processes being employed.

from gpt-neo.

yeah it maked me also some problems, but worked at last. but I couldn't train..there is some bug over there.

from gpt-neo.

Same here. Modified the command like this:
!python data/create_tfrecords.py --mode documents --input_dir /content/GPTNeo/$dataset_path --name $dataset_name --output_dir $out_name --write_dataset_config

The command fails with this error:
Traceback (most recent call last): File "data/create_tfrecords.py", line 205, in <module> results = create_tfrecords_mp(files, args) File "data/create_tfrecords.py", line 186, in create_tfrecords_mp files = split_list(files, len(files) // args.processes) File "data/create_tfrecords.py", line 67, in split_list return [l[i:i+n] for i in range(0, len(l), n)] ValueError: range() arg 3 must not be zero

from gpt-neo.

create_tfrecords.py: error: unrecognized arguments: --base_dir /content/GPTNeo/openwebtext --use_gpt2_tokenizer

I even tried using input_dir instead of base_dir and without gpt2_tokenizer. I think that worked, but when I got to copying data to the storage bucket it did this instead: No URLs matched: /content/GPTNeo/openwebtext_tokenized
I tried to make folders in the bucket to match the path but nothing worked.

The project documentation is a little out of date. using input_dir and dropping use_gpt2_tokenizer is the correct way to use the code. Can you elaborate on what goes wrong when you make those changes?

from gpt-neo.