search

exasol / transformers-extension

An Exasol extension for using state-of-the-art pretrained machine learning models via the Hugging Face Transformers API.
MIT License
2 stars 2 forks source link

Directory Structure that Model Upload UDF creates is different from what PredictionUDFs expect #211

Closed tkilias closed 2 months ago

tkilias commented 2 months ago

Reproduction:

CREATE OR REPLACE CONNECTION BUCKETFS_CONNECTION
    TO 'http://localhost:6583/default/models;bfsdefault'
    USER 'w'
    IDENTIFIED BY 'write';

SELECT TE_MODEL_DOWNLOADER_UDF(
    'sschet/biomedical-ner-all',
    'token_classification',
    'BUCKETFS_CONNECTION',
    NULL
);

SELECT TE_TOKEN_CLASSIFICATION_UDF(
    NULL,
    'BUCKETFS_CONNECTION',
    'token_classification',
    'sschet/biomedical-ner-all',
    'Influenza, commonly known as "the flu" or just "flu", is an infectious disease caused by influenza viruses. Symptoms range from mild to severe and often include fever, runny nose, sore throat, muscle pain, headache, coughing, and fatigue.',
    NULL
);

Error Message of TE_TOKEN_CLASSIFICATION_UDF

Traceback (most recent call last):   File "/usr/local/lib/python3.8/dist-packages/transformers/utils/hub.py", line 398, in cached_file     resolved_file = hf_hub_download(   File "/usr/local/lib/pytho
n3.8/dist-packages/huggingface_hub/utils/_validators.py", line 110, in _inner_fn     validate_repo_id(arg_value)   File "/usr/local/lib/python3.8/dist-packages/huggingface_hub/utils/_validators.py", l
ine 158, in validate_repo_id     raise HFValidationError( huggingface_hub.utils._validators.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/buckets/bfsdefault/de
fault/models/token_classification/sschet/biomedical-ner-all/pretrained/sschet/biomedical-ner-all'. Use `repo_type` argument if needed.  The above exception was the direct cause of the following except
ion:  Traceback (most recent call last):   File "/usr/local/lib/python3.8/dist-packages/exasol_transformers_extension/udfs/models/base_model_udf.py", line 95, in get_predictions_from_batch     self.ch
eck_cache(model_df)   File "/usr/local/lib/python3.8/dist-packages/exasol_transformers_extension/udfs/models/base_model_udf.py", line 192, in check_cache     self.last_created_pipeline = self.model_lo
ader.load_models(self.cache_dir,   File "/usr/local/lib/python3.8/dist-packages/exasol_transformers_extension/utils/load_local_model.py", line 49, in load_models     loaded_model = self._base_model_fa
ctory.from_pretrained(str(model_path))   File "/usr/local/lib/python3.8/dist-packages/transformers/models/auto/auto_factory.py", line 482, in from_pretrained     resolved_config_file = cached_file(   
File "/usr/local/lib/python3.8/dist-packages/transformers/utils/hub.py", line 462, in cached_file     raise EnvironmentError( OSError: Incorrect path_or_model_id: '/buckets/bfsdefault/default/models/t
oken_classification/sschet/biomedical-ner-all/pretrained/sschet/biomedical-ner-all'. Please provide either the path to a local folder or the repo_id of a model on the Hub.

Investigation

Directory Structure created by TE_MODEL_DOWNLOADER_UDF

[root@n11 /]# ls /exa/data/bucketfs/bfsdefault/.dest/default/models/token_classification/sschet/biomedical-ner-all/
config.json  model.safetensors  special_tokens_map.json  tokenizer.json  tokenizer_config.json  vocab.txt

[root@n11 /]# ls /exa/data/bucketfs/bfsdefault/.dest/default/models/token_classification/sschet/biomedical-ner-all/

Directory structured expected by PredictionUDFs

/buckets/bfsdefault/default/models/token_classification/sschet/biomedical-ner-all/pretrained/sschet/biomedical-ner-all

Acceptance Criteria