cache_dir cannot be the same as model name

[arunasank]

from transformers import GPTNeoXForCausalLM, AutoTokenizer

model = GPTNeoXForCausalLM.from_pretrained(
    "EleutherAI/pythia-12b",
    revision="step143000",
    cache_dir="./pythia-12b/step143000"
)

tokenizer = AutoTokenizer.from_pretrained(
    "EleutherAI/pythia-12b",
    cache_dir="./pythia-12b/step143000"
)

inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])

The above piece of code works for me. However, if I change the cache dir to be "EleutherAI/pythia-12b", the bin files are not downloaded, and the code errors. Shouldn't it work irrespective of what the cache_dir is?

Error Trace:

``` --------------------------------------------------------------------------- OSError Traceback (most recent call last) Cell In[30], line 3 1 from transformers import GPTNeoXForCausalLM, AutoTokenizer ----> 3 model = GPTNeoXForCausalLM.from_pretrained( 4 "EleutherAI[/pythia-12b](http://localhost:8001/pythia-12b)", 5 revision="step143000", 6 cache_dir="EleutherAI[/pythia-12b](http://localhost:8001/pythia-12b)" 7 ) 9 tokenizer = AutoTokenizer.from_pretrained( 10 "EleutherAI[/pythia-12b](http://localhost:8001/pythia-12b)", 11 cache_dir="EleutherAI[/pythia-12b](http://localhost:8001/pythia-12b)" 12 ) 14 inputs = tokenizer("Hello, I am", return_tensors="pt")

File [/mnt/align4_drive/arunas/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/modeling_utils.py:3075](http://localhost:8001/lab/tree/rm-interp/rm-playground/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/modeling_utils.py#line=3074), in PreTrainedModel.from_pretrained(cls, pretrained_model_name_or_path, config, cache_dir, ignore_mismatched_sizes, force_download, local_files_only, token, revision, use_safetensors, *model_args, **kwargs)
   3073 if not isinstance(config, PretrainedConfig):
   3074     config_path = config if config is not None else pretrained_model_name_or_path
-> 3075     config, model_kwargs = cls.config_class.from_pretrained(
   3076         config_path,
   3077         cache_dir=cache_dir,
   3078         return_unused_kwargs=True,
   3079         force_download=force_download,
   3080         resume_download=resume_download,
   3081         proxies=proxies,
   3082         local_files_only=local_files_only,
   3083         token=token,
   3084         revision=revision,
   3085         subfolder=subfolder,
   3086         _from_auto=from_auto_class,
   3087         _from_pipeline=from_pipeline,
   3088         **kwargs,
   3089     )
   3090 else:
   3091     # In case one passes a config to `from_pretrained` + "attn_implementation"
   3092     # override the `_attn_implementation` attribute to `attn_implementation` of the kwargs
   (...)
   3096     # we pop attn_implementation from the kwargs but this handles the case where users
   3097     # passes manually the config to `from_pretrained`.
   3098     config = copy.deepcopy(config)

File [/mnt/align4_drive/arunas/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/configuration_utils.py:605](http://localhost:8001/lab/tree/rm-interp/rm-playground/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/configuration_utils.py#line=604), in PretrainedConfig.from_pretrained(cls, pretrained_model_name_or_path, cache_dir, force_download, local_files_only, token, revision, **kwargs)
    601 kwargs["revision"] = revision
    603 cls._set_token_in_kwargs(kwargs, token)
--> 605 config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
    606 if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
    607     logger.warning(
    608         f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
    609         f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
    610     )

File [/mnt/align4_drive/arunas/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/configuration_utils.py:634](http://localhost:8001/lab/tree/rm-interp/rm-playground/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/configuration_utils.py#line=633), in PretrainedConfig.get_config_dict(cls, pretrained_model_name_or_path, **kwargs)
    632 original_kwargs = copy.deepcopy(kwargs)
    633 # Get config dict associated with the base config file
--> 634 config_dict, kwargs = cls._get_config_dict(pretrained_model_name_or_path, **kwargs)
    635 if "_commit_hash" in config_dict:
    636     original_kwargs["_commit_hash"] = config_dict["_commit_hash"]

File [/mnt/align4_drive/arunas/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/configuration_utils.py:689](http://localhost:8001/lab/tree/rm-interp/rm-playground/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/configuration_utils.py#line=688), in PretrainedConfig._get_config_dict(cls, pretrained_model_name_or_path, **kwargs)
    685 configuration_file = kwargs.pop("_configuration_file", CONFIG_NAME)
    687 try:
    688     # Load from local folder or from cache or download from model Hub and cache
--> 689     resolved_config_file = cached_file(
    690         pretrained_model_name_or_path,
    691         configuration_file,
    692         cache_dir=cache_dir,
    693         force_download=force_download,
    694         proxies=proxies,
    695         resume_download=resume_download,
    696         local_files_only=local_files_only,
    697         token=token,
    698         user_agent=user_agent,
    699         revision=revision,
    700         subfolder=subfolder,
    701         _commit_hash=commit_hash,
    702     )
    703     commit_hash = extract_commit_hash(resolved_config_file, commit_hash)
    704 except EnvironmentError:
    705     # Raise any environment error raise by `cached_file`. It will have a helpful error message adapted to
    706     # the original exception.

File [/mnt/align4_drive/arunas/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/utils/hub.py:356](http://localhost:8001/lab/tree/rm-interp/rm-playground/miniconda3/envs/rm-score/lib/python3.11/site-packages/transformers/utils/hub.py#line=355), in cached_file(path_or_repo_id, filename, cache_dir, force_download, resume_download, proxies, token, revision, local_files_only, subfolder, repo_type, user_agent, _raise_exceptions_for_missing_entries, _raise_exceptions_for_connection_errors, _commit_hash, **deprecated_kwargs)
    354 if not os.path.isfile(resolved_file):
    355     if _raise_exceptions_for_missing_entries:
--> 356         raise EnvironmentError(
    357             f"{path_or_repo_id} does not appear to have a file named {full_filename}. Checkout "
    358             f"'https://huggingface.co/{path_or_repo_id}/{[revision](https://huggingface.co/%7Bpath_or_repo_id%7D/%7Brevision)}' for available files."
    359         )
    360     else:
    361         return None

OSError: EleutherAI[/pythia-12b](http://localhost:8001/pythia-12b) does not appear to have a file named config.json. Checkout 'https://huggingface.co/EleutherAI/pythia-12b/step143000' for available files.
```