run_ner.txt

Named Entity Recognition (NER), and explicitly using a specific model different from the defaults used previously.

NER models identify and categorize key entities in text, such as names of people, organizations, locations, etc.

Prerequisites:

Ensure you have the necessary libraries. Like the zero-shot example, many tokenizers used by models like BERT require sentencepiece.

Bash

pip install transformers sentencepiece torch
# Or: pip install transformers sentencepiece tensorflow
(Ensure your existing virtual environment is active).

Explicit Model Choice:

We will explicitly use dbmdz/bert-large-cased-finetuned-conll03-english. This is a BERT-large model fine-tuned on the CoNLL-2003 NER dataset, known for good performance in identifying standard entity types (Person, Location, Organization, Misc).

How to Run:

Make sure you've run pip install transformers sentencepiece torch (or the tensorflow equivalent) in your activated virtual environment.
Save the code above into a file named run_ner.py.
Open your Ubuntu terminal.
Make sure your virtual environment is activated (source .venv/bin/activate).
Run the script:
Bash

python run_ner.py
What to Expect:

First Run: It will download the dbmdz/bert-large-cased-finetuned-conll03-english model files, which are quite large (BERT-large is over 1GB), and cache them.
NER Execution: It will process the input text sentence by sentence.
Output: It will print a list of the entities found in the text. Thanks to aggregation_strategy="simple", it should group consecutive parts of an entity (like "Eleanor Vance"). You should expect it to identify:
Eleanor Vance as PER (Person)
Perth as LOC (Location)
Globex Corporation as ORG (Organization)
Yagan Square as LOC (Location)
David Chen as PER (Person)
Zenith Solutions as ORG (Organization)
Art Gallery of Western Australia likely as ORG (sometimes complex names are tagged ORG) or maybe MISC.
Western Australia might be identified separately as LOC.
This example explicitly uses a different, specified model (dbmdz/bert...) for a distinct task (NER), showcasing more direct control over model choice while still leveraging the convenience of the pipeline framework.