Question Answering: We directly feed the question as input prompt
Text Summarization: We add TL;DR after the input text to instruct the LLMs to generate summarization.
Machine Translation:We used the prompt template from the GPT-3 paper:
Translate French to English:
loutre de mer => sea otter #example
[Input French Transcript] =>
The example is drawn randomly from WMT-14 training data.
Code Generation: We directly feed the requirements in the dataset (HumanEval, MBPP) to the LLM.
We added the following system prompt to the fine-tuned Llama-2-7b-chat model.
[INST] <<SYS>>
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information
\n<</SYS>>\n\n
<TASK PROMPT>
[/INST]
We added "<| file ext=.py |>\n" as the prefix to the incoder model.
We added "# Import libraries.\n import numpy as np" as the prefix to the codegen model
Below are the results for Llama-2 without prompt template (system prompt)
(Supplementary results of RQ2)