2024-09-12
LLM Exercises for Conversational AI in Global Health at the MBA IHM Swiss TPH
HIV Exercises — Prompts
When should a person with a potential exposure to HIV get tested?
Should I disclose my HIV status?
Can someone with HIV get married?
…
Basic Prompting
Exercise 1 – Prompt a small open model on HIV
Mistral-7B-Instruct-v0.3 is an advanced AI language model developed by Mistral AI, a company based in France.
It is designed to follow instructions and generate human-like text, and it contains 7 billion parameters, making it relatively compact compared to other LLMs.
HIV - Prompt - Mistral-7B-Instruct-v0.3: https://andreasmartin-flowise.hf.space/chatbot/8a8dc96e-d1aa-4cc8-97b6-ed3c635e0371
Exercise 2 – Prompt a mid-size open model on HIV
Meta-Llama-3.1-70B-Instruct is a highly advanced AI language model developed by Meta, designed for generating human-like text and following complex instructions.
It is part of the Llama 3.1 family of models, featuring 70 billion parameters, which makes it significantly larger and more powerful compared to smaller models like Mistral-7B.
HIV - Prompt - Meta-Llama-3.1-70B-Instruct: https://andreasmartin-flowise.hf.space/chatbot/c88c607f-6cc4-45dc-8162-59599191c1c3
Retrieval Augmented Generation (RAG)
In the following, we make use of our information source and instruct an LLM to provide an answer on the retrieved information.
Exercise 3 — RAG using a mid-size open model
In this exercise, we prompt Meta-Llama-3.1-70B-Instruct in RAG-fashion.
HIV - RAG - Meta-Llama-3.1-70B-Instruct: https://andreasmartin-flowise.hf.space/chatbot/edf2d55e-542e-48e0-bca6-f5e0962c1bf9
Exercise 4 — RAG using a small open model
In this exercise, we prompt Mistral-7B-Instruct-v0.3 in RAG-fashion.
HIV - RAG - Mistral-7B-Instruct-v0.3: https://andreasmartin-flowise.hf.space/chatbot/18f88b14-9e73-4c1a-a287-ed9a81e78a31
Gender Bias Mitigation — Prompt
The line manager gives the secretary a task.
Basic Prompting
Exercise 5 — Gender Bias Prompt
Gender Bias Mitigation - Prompt: https://andreasmartin-flowise.hf.space/chatbot/bd88111e-2e28-4fe7-a7c6-11b8b466505f
Exercise 6 — Gender Bias Guardrails
Gender Bias Mitigation - Prompt and Guardrails: https://andreasmartin-flowise.hf.space/chatbot/36e03965-9ef7-4afc-8c2c-25992a1ccd22
Fine-Tuning & Alignment
Exercise 7 — Supervised Fine-Tuning
Gender Bias Mitigation - Prompt SFT: https://andreasmartin-flowise.hf.space/chatbot/8e1bccef-623c-4a24-ac1b-83e6dbb72e1a
Exercise 8 — Preference Alignment
Gender Bias Mitigation - Prompt DPO: https://andreasmartin-flowise.hf.space/chatbot/5dd53b85-c5d1-46b6-b494-95dacb6cb15f
Last updated