Lab 3 - Fine-tune FLAN-T5 with reinforcement learning to generate more-positive summaries

Course Q&A Generative AI with Large Language Models GenAI with LLMs Resources
1

When working on the lab 3, the following cell results in a error

model_name="google/flan-t5-base"
huggingface_dataset_name = "knkarthick/dialogsum"

dataset_original = load_dataset(huggingface_dataset_name)

dataset_original

ValueError Traceback (most recent call last)
Cell In[11], line 4
1 model_name=“google/flan-t5-base”
2 huggingface_dataset_name = “knkarthick/dialogsum”
----> 4 dataset_original = load_dataset(huggingface_dataset_name)
6 dataset_original

File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1767, in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, ignore_verifications, keep_in_memory, save_infos, revision, use_auth_token, task, streaming, num_proc, storage_options, **config_kwargs)
1762 verification_mode = VerificationMode(
1763 (verification_mode or VerificationMode.BASIC_CHECKS) if not save_infos else VerificationMode.ALL_CHECKS
1764 )
1766 # Create a dataset builder
→ 1767 builder_instance = load_dataset_builder(
1768 path=path,
1769 name=name,
1770 data_dir=data_dir,
1771 data_files=data_files,
1772 cache_dir=cache_dir,
1773 features=features,
1774 download_config=download_config,
1775 download_mode=download_mode,
1776 revision=revision,
1777 use_auth_token=use_auth_token,
1778 storage_options=storage_options,
1779 **config_kwargs,
1780 )
1782 # Return iterable dataset in case of streaming
1783 if streaming:

File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1498, in load_dataset_builder(path, name, data_dir, data_files, cache_dir, features, download_config, download_mode, revision, use_auth_token, storage_options, **config_kwargs)
1496 download_config = download_config.copy() if download_config else DownloadConfig()
1497 download_config.use_auth_token = use_auth_token
→ 1498 dataset_module = dataset_module_factory(
1499 path,
1500 revision=revision,
1501 download_config=download_config,
1502 download_mode=download_mode,
1503 data_dir=data_dir,
1504 data_files=data_files,
1505 )
1507 # Get dataset builder class from the processing script
1508 builder_cls = import_main_class(dataset_module.module_path)

File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1215, in dataset_module_factory(path, revision, download_config, download_mode, dynamic_modules_path, data_dir, data_files, **download_kwargs)
1210 if isinstance(e1, FileNotFoundError):
1211 raise FileNotFoundError(
1212 f"Couldn’t find a dataset script at {relative_to_absolute_path(combined_path)} or any data file in the same directory. "
1213 f"Couldn’t find ‘{path}’ on the Hugging Face Hub either: {type(e1).name}: {e1}"
1214 ) from None
→ 1215 raise e1 from None
1216 else:
1217 raise FileNotFoundError(
1218 f"Couldn’t find a dataset script at {relative_to_absolute_path(combined_path)} or any data file in the same directory."
1219 )

File /opt/conda/lib/python3.10/site-packages/datasets/load.py:1199, in dataset_module_factory(path, revision, download_config, download_mode, dynamic_modules_path, data_dir, data_files, **download_kwargs)
1184 return HubDatasetModuleFactoryWithScript(
1185 path,
1186 revision=revision,
(…)
1189 dynamic_modules_path=dynamic_modules_path,
1190 ).get_module()
1191 else:
1192 return HubDatasetModuleFactoryWithoutScript(
1193 path,
1194 revision=revision,
1195 data_dir=data_dir,
1196 data_files=data_files,
1197 download_config=download_config,
1198 download_mode=download_mode,
→ 1199 ).get_module()
1200 except (
1201 Exception
1202 ) as e1: # noqa: all the attempts failed, before raising the error we should check if the module is already cached.
1203 try:

File /opt/conda/lib/python3.10/site-packages/datasets/load.py:765, in HubDatasetModuleFactoryWithoutScript.get_module(self)
755 def get_module(self) → DatasetModule:
756 hfh_dataset_info = HfApi(config.HF_ENDPOINT).dataset_info(
757 self.name,
758 revision=self.revision,
759 token=self.download_config.use_auth_token,
760 timeout=100.0,
761 )
762 patterns = (
763 sanitize_patterns(self.data_files)
764 if self.data_files is not None
→ 765 else get_data_patterns_in_dataset_repository(hfh_dataset_info, self.data_dir)
766 )
767 data_files = DataFilesDict.from_hf_repo(
768 patterns,
769 dataset_info=hfh_dataset_info,
770 base_path=self.data_dir,
771 allowed_extensions=ALL_ALLOWED_EXTENSIONS,
772 )
773 module_names = {
774 key: infer_module_for_data_files(data_files_list, use_auth_token=self.download_config.use_auth_token)
775 for key, data_files_list in data_files.items()
776 }

File /opt/conda/lib/python3.10/site-packages/datasets/data_files.py:675, in get_data_patterns_in_dataset_repository(dataset_info, base_path)
673 resolver = partial(_resolve_single_pattern_in_dataset_repository, dataset_info, base_path=base_path)
674 try:
→ 675 return _get_data_files_patterns(resolver)
676 except FileNotFoundError:
677 raise EmptyDatasetError(
678 f"The dataset repository at ‘{dataset_info.id}’ doesn’t contain any data files"
679 ) from None

File /opt/conda/lib/python3.10/site-packages/datasets/data_files.py:236, in _get_data_files_patterns(pattern_resolver)
234 try:
235 for pattern in patterns:
→ 236 data_files = pattern_resolver(pattern)
237 if len(data_files) > 0:
238 non_empty_splits.append(split)

File /opt/conda/lib/python3.10/site-packages/datasets/data_files.py:486, in _resolve_single_pattern_in_dataset_repository(dataset_info, pattern, base_path, allowed_extensions)
484 else:
485 base_path = “/”
→ 486 glob_iter = [PurePath(filepath) for filepath in fs.glob(PurePath(pattern).as_posix()) if fs.isfile(filepath)]
487 matched_paths = [
488 filepath
489 for filepath in glob_iter
(…)
496 )
497 ] # ignore .ipynb and pycache, but keep /…/
498 if allowed_extensions is not None:

File /opt/conda/lib/python3.10/site-packages/fsspec/spec.py:606, in AbstractFileSystem.glob(self, path, maxdepth, **kwargs)
602 depth = None
604 allpaths = self.find(root, maxdepth=depth, withdirs=True, detail=True, **kwargs)
→ 606 pattern = glob_translate(path + (“/” if ends_with_sep else “”))
607 pattern = re.compile(pattern)
609 out = {
610 p: info
611 for p, info in sorted(allpaths.items())
(…)
618 )
619 }

File /opt/conda/lib/python3.10/site-packages/fsspec/utils.py:734, in glob_translate(pat)
732 continue
733 elif “" in part:
→ 734 raise ValueError(
735 "Invalid pattern: '’ can only be an entire path component”
736 )
737 if part:
738 results.extend(_translate(part, f"{not_sep}*", not_sep))

ValueError: Invalid pattern: ‘**’ can only be an entire path component

How to solve it? What happen when I can submit my assignment due an error?

Regards,

Emerson

2

Facing Same issue

3

Reading other threads, I found a recommendation to use %pip install -U datasets, so this the cell I used to install the libraries, also after that I restarted the kernel, on the menu Kernel → restart kernel.

%pip install --upgrade pip
%pip install --disable-pip-version-check \
    torch==1.13.1 \
    torchdata==0.5.1 --quiet

%pip install \
    transformers==4.27.2 \
    #datasets==2.11.0 \
    evaluate==0.4.0 \
    rouge_score==0.1.2 \
    peft==0.3.0 --quiet

%pip install -U datasets 

# Installing the Reinforcement Learning library directly from github.
%pip install git+https://github.com/lvwerra/trl.git@25fa1bd

Lab3 is working again.

4

USE datasets ==2.15.0 as it is the latest version…it worked for me

5

Thank you. I did that and restarted the Kernel. How did you find/know the latest version?

6

Anyone who runs into this problem make sure to restart kernel after upgrading datasets!

7

Hi,

The pip install command with the --upgrade or -U will update/install the latest version of the package. %pip install -U datasets

Or you can visit https://pypi.org/ look for the package version and use the version you need.

Thanks

Emerson

8

Hi,

Could you please share some insights on this?

I ran the code and restarted the kernal. But when I moved on and ran the code below (with all the imports), I got an error message:

RuntimeError: Failed to import transformers.pipelines because of the following error (look up to see its traceback):
No module named ‘torch.distributed’

Thank you very much!

Zhengsi