Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import pipeline
|
3 |
+
import os
|
4 |
+
|
5 |
+
# Cache pipelines so we don’t reload every call
|
6 |
+
pipeline_cache = {}
|
7 |
+
|
8 |
+
def get_generator(model_name):
|
9 |
+
if model_name not in pipeline_cache:
|
10 |
+
pipeline_cache[model_name] = pipeline(
|
11 |
+
"text-generation",
|
12 |
+
model=model_name,
|
13 |
+
device=0 if os.environ.get("CUDA_VISIBLE_DEVICES") else -1
|
14 |
+
)
|
15 |
+
return pipeline_cache[model_name]
|
16 |
+
|
17 |
+
def build_prompt(system, examples, query, style):
|
18 |
+
# Prefix with system role
|
19 |
+
prompt = f"SYSTEM: {system.strip()}\n\n"
|
20 |
+
if style == "Zero-Shot":
|
21 |
+
prompt += f"USER: {query.strip()}\nASSISTANT:"
|
22 |
+
else: # Few-Shot
|
23 |
+
# Insert up to three examples
|
24 |
+
for ex in examples:
|
25 |
+
ex = ex.strip()
|
26 |
+
if ex:
|
27 |
+
prompt += f"EXAMPLE: {ex}\n"
|
28 |
+
prompt += f"USER: {query.strip()}\nASSISTANT:"
|
29 |
+
return prompt
|
30 |
+
|
31 |
+
def generate(system, ex1, ex2, ex3, query, model_name, style):
|
32 |
+
# Build prompt text
|
33 |
+
examples = [ex1, ex2, ex3]
|
34 |
+
prompt = build_prompt(system, examples, query, style)
|
35 |
+
gen = get_generator(model_name)
|
36 |
+
# Generate up to 100 tokens beyond prompt
|
37 |
+
out = gen(prompt, max_length=len(prompt.split())+100, do_sample=True)[0]["generated_text"]
|
38 |
+
# Strip prompt prefix
|
39 |
+
return out[len(prompt):].strip()
|
40 |
+
|
41 |
+
# Available small/medium models on HF Hub
|
42 |
+
AVAILABLE_MODELS = [
|
43 |
+
"gpt2",
|
44 |
+
"EleutherAI/gpt-neo-125M",
|
45 |
+
"distilgpt2"
|
46 |
+
]
|
47 |
+
|
48 |
+
with gr.Blocks(theme=gr.themes.Default()) as demo:
|
49 |
+
gr.Markdown("## 🎮 Prompt Engineering Playground")
|
50 |
+
gr.Markdown(
|
51 |
+
"Experiment with **Zero-Shot** vs **Few-Shot** prompting across different open models."
|
52 |
+
)
|
53 |
+
|
54 |
+
with gr.Row():
|
55 |
+
with gr.Column(scale=2):
|
56 |
+
system_box = gr.Textbox(
|
57 |
+
label="🖥️ System Prompt",
|
58 |
+
placeholder="You are a helpful assistant."
|
59 |
+
)
|
60 |
+
ex1 = gr.Textbox(label="📖 Example 1 (few-shot only)", placeholder="…")
|
61 |
+
ex2 = gr.Textbox(label="📖 Example 2 (few-shot only)")
|
62 |
+
ex3 = gr.Textbox(label="📖 Example 3 (few-shot only)")
|
63 |
+
query_box = gr.Textbox(label="🗣️ User Prompt", placeholder="Type your query here…")
|
64 |
+
|
65 |
+
with gr.Row():
|
66 |
+
model_choice = gr.Dropdown(
|
67 |
+
choices=AVAILABLE_MODELS,
|
68 |
+
value=AVAILABLE_MODELS[0],
|
69 |
+
label="🤖 Model"
|
70 |
+
)
|
71 |
+
style_choice = gr.Radio(
|
72 |
+
choices=["Zero-Shot", "Few-Shot"],
|
73 |
+
value="Zero-Shot",
|
74 |
+
label="🎨 Prompt Style"
|
75 |
+
)
|
76 |
+
|
77 |
+
run_btn = gr.Button("🚀 Generate")
|
78 |
+
|
79 |
+
with gr.Column(scale=3):
|
80 |
+
output_box = gr.Textbox(
|
81 |
+
label="💬 Model Output",
|
82 |
+
lines=15
|
83 |
+
)
|
84 |
+
|
85 |
+
run_btn.click(
|
86 |
+
fn=generate,
|
87 |
+
inputs=[system_box, ex1, ex2, ex3, query_box, model_choice, style_choice],
|
88 |
+
outputs=output_box
|
89 |
+
)
|
90 |
+
|
91 |
+
demo.launch()
|