HoT: Highlighted Chain of Thought for Referencing Supporting Facts from Inputs

An Achilles heel of Large Language Models (LLMs) is their tendency to hallucinate non-factual statements. A response mixed of factual and non-factual statements poses a challenge for humans to verify and accurately base their decisions on. To combat this problem, we propose Highlighted Chain-of-Thought Prompting (HoT), a technique for prompting LLMs to generate responses with XML-tags that ground facts to those provided in the query. That is, given an input question, LLMs would first re-format the question to add XML tags highlighting key facts, and then, generate a response with highlights over the facts referenced from the input. Interestingly, in few-shot settings, HoT **outperforms** the vanilla chain of thoughts (CoT) on a wide range of 17 tasks from arithmetic, reading comprehension to logical reasoning.

1. Requirements

python=3.10.15
google-generativeai==0.8.3
openai==1.58.1

2. How to Run

Run the following command to execute the script:

python main.py --save_answer --llm_model "$llm_model" --dataset "$dataset" --answer_mode "$run_mode" --data_mode "$data_mode"

Parameters:

• --llm_model: Defines the LLM model to use. Choices include:
  • gemini-1.5-pro-002, gemini-1.5-flash-002,
  • gpt-4o-2024-08-06
  • llama_8b, llama_70b, llama_sambanova_405b
  • qwen25_coder_32b, qwq_32b, deepseek_r1
• --dataset: Specifies the dataset to evaluate, such as:
  • GSM8K, AQUA, DROP
• --answer_mode": Determines the answering strategy:
  • cot: Chain-of-Thought prompting
  • hot: Highlight Chain-of-Thought prompting
• --data_mode:
  • random: Runs the model on 200 randomly selected samples.
  • longest: Runs the model on 200 longest samples.
  • shortest: Runs the model on 200 shortest samples.
  • full: Runs the model on the whole dataset.

Example Usage

python main.py --save_answer --llm_model "gpt-4o-2024-08-06" --dataset "GSM8K" --answer_mode "cot" --data_mode random

3. How to evaluate the result

Run the following command to evaluate the results:

python evaluate.py --llm_model "$llm_model" --dataset "$dataset" --answer_mode "$answer_mode" --data_mode "$data_mode"

Example Usage

python evaluate.py --llm_model "gpt-4o-2024-08-06" --dataset "GSM8K" --answer_mode "cot" --data_mode longest

4. How to visualize the result

Run the following command to render the result on html pages:

python visualize.py --llm_model "$llm_model" --dataset "$dataset" --answer_mode "$answer_mode" --save_html

Example Usage

python visualize.py --llm_model "gpt-4o-2024-08-06" --dataset "GSM8K" --answer_mode "cot" --data_mode --save_html

5. License

MIT