nlp_portfolio

Replication of the "Evaluating Parameter-Efficient Finetuning Approaches for Pre-trained Models on the Financial Domain"

Peft methods

Recent findings show that LoRa and Adapter fine-tuning methods show the similar performance as the full fine-tuning while saving time, computational resources and memory.

Steps done

We repeated the experiment and got nearly the same performance and confirm that paper's results are reliable

The tasks in Finance tested here are

• Question Answering
• Named Entity Recognition
• News Headline Classification
• Sentiment analysis

Link to the Notebook: is here -> Notebook