Requirements
1
2
3
4
5
6
7
8
9
10
11
12
| apt update
apt install python3-pip -y
mkdir llm-prompts
cd llm-prompts
cat >requirements.txt <<EOF
transformers==4.48.3
torch==2.6.0
accelerate==1.8.1
einops==0.8.1
jinja2==3.1.6
EOF
pip install -r requirements.txt
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
| cat >llm-prompts.py <<EOF
from transformers import AutoModelForCausalLM, AutoTokenizer
revision_id = "0a67737cc96d2554230f90338b163bc6380a2a85"
model = AutoModelForCausalLM.from_pretrained(
"microsoft/Phi-3-mini-4k-instruct",
revision=revision_id,
device_map="auto",
torch_dtype="auto",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct", revision=revision_id)
from transformers import pipeline
generator = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
return_full_text=False,
max_new_tokens=500,
do_sample=False
)
while True:
print("-" *50) # Horizontal line
print("What do you want?")
user_input = input("\033[92mType something, or X to exit: \033[0m") # Ask the user for input in green color
if user_input in ['X', 'x']: # If user types X or x, exit the program
print("Exiting.")
break
else:
messages = [{"role":"user", "content": user_input}]
response = generator(messages)
print(response[0]["generated_text"])
EOF
python3 llm-prompts.py
|
In our llm-prompts.py, here’s how we are passing a user prompt to the model.
1
2
3
| messages = [
{"role":"user", "content": user_input}
]
|
Well, we are actually doing all that in one line. But in the example above, we have split the single line code into multiple lines for better readability.
In that particular piece of code, how can we pass a system prompt to the model?
We can do that by adding a system prompt to the messages list.
1
2
3
4
| messages = [
{"role":"system", "content": system_prompt},
{"role":"user", "content": user_input}
]
|
What Can Our First System Prompt Be?
Let’s experiment a bit. Wouldn’t it be nice if we created a system prompt that will help with cooking recipes, and it responded like the great poet, William Shakespeare?
1
| system_prompt = "You are a helpful assistant that helps with cooking recipes. You are also a great poet, and you can write poems about anything. When you respond, make sure to include a small 4 line poem like William Shakespeare would."
|
If we add the following lines we can ask the prompt to behave in a different way:
1
2
3
4
5
6
7
8
9
10
11
12
| while True:
print("-" *50) # Horizontal line
print("What do you want?")
user_input = input("\033[92mType something, or X to exit: \033[0m") # Ask the user for input in green color
system_prompt = "You are a helpful assistant that helps with cooking recipes. You are also a great poet, and you can write poems about anything. When you respond, make sure to include a small 4 line poem like William Shakespeare would."
if user_input in ['X', 'x']: # If user types X or x, exit the program
print("Exiting.")
break
else:
messages = [{"role":"user", "content": user_input},{"role":"system", "content": system_prompt}]
response = generator(messages)
print(response[0]["generated_text"])
|
Changing System Prompts On the Fly
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
| cat >llm-prompts.py <<EOF
from transformers import AutoModelForCausalLM, AutoTokenizer
revision_id = "0a67737cc96d2554230f90338b163bc6380a2a85"
model = AutoModelForCausalLM.from_pretrained(
"microsoft/Phi-3-mini-4k-instruct",
revision=revision_id,
device_map="auto",
torch_dtype="auto",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct", revision=revision_id)
from transformers import pipeline
generator = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
return_full_text=False,
max_new_tokens=500,
do_sample=False
)
# Main interaction loop
print("🤖 What do you want? 🤖")
print("\033[92mType your message. Use '/system' to modify system prompt. Type 'X' or 'x' to exit.\033[0m")
# Default system prompt configuration
system_prompt = """You are a helpful, respectful, and honest AI assistant.
Always prioritize safety and provide accurate information.
Respond clearly and concisely."""
while True:
print("-" * 50)
print("\033[95mCurrent System Prompt: \033[0m" + system_prompt)
print("-" * 50)
user_input = input("\033[92mYour message: \033[0m")
if user_input in ['X', 'x']: # If user types X or x, exit the program
print("Exiting.")
break
# System prompt modification
elif user_input.startswith('/system'):
print("\033[94mEnter new system prompt:\033[0m")
system_prompt = input()
print("\033[93mSystem prompt updated.\033[0m")
continue
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_input}
]
response = generator(messages)
print(response[0]["generated_text"])
EOF
|
Changing the System Prompt
To change the system prompt, you can input the following command:
Then enter:
1
| You will not talk about capital of any country. You will no answer any questions regarding geographies or distances.
|
After updating the system prompt, you can ask the chatbot, a few questions.
1
2
3
| What is the capital of Britain
What is the capital of Greenland
|
