ADL_FINAL

Chemical Informatics and Alternative Substance Chatbot Service

Environment Setup

0. Install Ollama

First, download and install the appropriate version of Ollama from the official website: Ollama Official Website.

After installation, launch Ollama and ensure that the model runs locally:

ollama run kenneth85/llama-3-taiwan

1. Create a Conda Environment

conda create -n FINAL python=3.11

2. Activate the Conda Environment

conda activate FINAL

3. Install Required Packages

pip install -r requirements.txt

Execute RAG

Launch Command

Click the URL displayed in the output to start using the chatbot service:

streamlit run chatbot.py --server.port xxxx

Example

streamlit run chatbot.py --server.port 2024

Data Processing

Convert chemical substance data in a .txt file into a vector database. Configure the file path, chunk size, and overlap size, and use Chroma to save it as a vector database:

python txt_to_vector.py

Model Loading

Set up the LLM and Embedding models. Choose the appropriate models based on your needs:

python init_model.py

Prompt and Chatbot Usage Rules Configuration

• Modify RAG Character Prompt Content: Adjust retriever_chain.py.
• Modify Nemoguardrails Constraints: Edit ./config/config.yml.

Evalutation

We store the chatbot's responses to our various questions in a JSON file and split them into smaller parts to avoid OpenAI API timeouts. We then execute evaluation.py on each part to evaluate our RAG model. Finally, we compile the results into a chart and include it in the report.

python evaluation/evaluation.py

The following are the four indicators that will be generated after the RAGAS assessment: