GenAI – Types Of Prompting Techniques.
Table Of Contents:
- Zero Short Prompting.
- Few Short Prompting.
- Chain Of Thoughts Prompting.
- Role Prompting.
- System Prompting.
- Instruction Tuning.
- Self Consistency Prompting.
(1) Zero Shot Prompting
Example – 1
Example – 2
(2) Few Shot Prompting
Example – 1
Example – 2
Example – 3
Example – 4
Example – 5
Example – 6
(3) Chain Of Thoughts Prompting
Example – 1
Example – 2
Example – 3
Example – 4
Example – 5
How You Will Pass The Questins To The LLM In Chain Of Thoughts ?
(4) Role Prompting
You are a [role].
[Give task/instruction here].
Example – 1:
Example – 2:
Example – 3:
Example – 4:
Example – 5:
(5) System Prompting
Example – 1:
Example – 2:
Example – 3:
Example – 4:
Example – 5:
Where Do You Mention The System Lebel Prompt ?
(1) In OpenAI API (chat/completions endpoint)
user_question = input("Please enter your legal question: ")
messages = [
{"role": "system", "content": "You are a helpful legal advisor. Only answer legal questions. Always be formal."},
{"role": "user", "content": user_question}
]
response = openai.ChatCompletion.create(
model="gpt-4",
messages=messages
)
(2) In LangChain : LangChain has special SystemMessagePromptTemplate.
from langchain.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate
system_prompt = SystemMessagePromptTemplate.from_template(
"You are a helpful customer support chatbot. Only answer questions about orders and returns."
)
human_prompt = HumanMessagePromptTemplate.from_template(
"{user_question}"
)
chat_prompt = ChatPromptTemplate.from_messages([system_prompt, human_prompt])
(6) Instruction Tuning
Who Will Do Instruction Tuning?
Prompt Engineering Vs Instruction Tuning
Steps To Do Instruction Tuning?
(7) Self-Consistency Prompting
Example : Math Problem
Python Example:
from collections import Counter
import openai
answers = []
for _ in range(10): # 10 reasoning paths
response = openai.ChatCompletion.create(
model="gpt-4",
messages=[
{"role": "system", "content": "You are a helpful math tutor."},
{"role": "user", "content": "Q: A car moves 40 km/h for 3 hours. How far? Let's think step by step."}
],
temperature=0.8 # randomness to vary reasoning paths
)
output = response['choices'][0]['message']['content']
final_answer = output.strip().split()[-2] # example: extract "120" from "Answer: 120 km"
answers.append(final_answer)
# Choose most common answer
most_common = Counter(answers).most_common(1)[0][0]
print("Most consistent answer:", most_common)
