Recently, I trace many codes about LLaMA LoRA training (almost two months ago…), and when I reading the codes, I found if they want to configure the prompt for LLMs, if they need to change the prompt content by different training, the format_map() function is the best choice.
So, today I want to record something about format_map() function!
Sample Code
Simply put, format_map() can fill the original string if they are keep the null variables, and become to the new string.
For example, if we set two variables like {Name} and {Age}, and we have a lot of training data, we can use for-loop to get the new combined string.
# coding: utf-8
def main() -> None:
data = [
{"Name": "Clay", "Age": 29},
{"Name": "Atlas", "Age": 27},
{"Name": "Wendy", "Age": 20},
]
original_text = "My name is {Name} and I am {Age} years old."
for item in data:
print(original_text.format_map(item))
if __name__ == "__main__":
main()
Output:
My name is Clay and I am 29 years old. My name is Atlas and I am 27 years old. My name is Wendy and I am 20 years old.
If you do not know the syntax, we also can do it like:
for item in data:
print("My name is {} and I am {} years old.".format(item["Name"], item["Age"]))for item in data:
print(f"My name is {item['Name']} and I am {item['Age']} years old.")for item in data:
print(f"My name is " + item["Name"] + " and I am " + str(item["Age"]) + " years old.")The first method can look very messy when there are many variables.
The second method is fairly good and has good readability; it’s just that when it comes to conditionally switching different variables for filling, it’s not as convenient as format_map(), and might require modifying the original data.
The third method has the worst readability.
Above are some Python tips for concatenating strings.