1. Data for Paragraph2Action

The data/few_example_data of Paragraph2Action dataset is a subset of https://github.com/rxn4chemistry/paragraph2actions/tree/main/test_data

The whole paragraph2action dataset data/hand_annotated is available upon request (with pistachio license).

The processed dataset is in data/processed_data.

2. Methods for Paragraph2Action

Prompt Engineering ChatGPT (GPT-4, GPT-3.5-Turbo)

See in prompt_chatgpt_for_paragraph2action.ipynb

Fine-tuning ChatGPT (GPT-3.5-Turbo)

See in finetune_chatgpt_for_paragraph2action.ipynb

Full Parameter Fine-tuning Open-source Large Language Models (Mistral, Llama3, Llama2)

Training Code in finetune_llms_full_for_paragraph2action.py

Inferencing Code in vllm_inference_full_finetuned_llms.ipynb

Parameter Efficient Fine-tuning (PEFT) Open-source Large Language Models (Mistral, Llama3, Llama2)

Training Code in finetune_llms_peft_for_paragraph2action.py

Inferencing Code in vllm_inference_peft_finetuned_llms.ipynb

Fine-tuning Language Models (T5, Bart)

Training Code in finetune_bart_or_t5_for_paragraph2action.py

3. Evaluating the results of Paragraph2Action

All predictions will be saved in results/predictions

Evaluating codes are in evaluate_for_paragraph2action.ipynb