Upload folder using huggingface_hub
Browse files
demo.py
ADDED
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
"""
|
3 |
+
T5 Prompt Enhancer V0.3 Demo Script
|
4 |
+
Quick test of all four instruction types
|
5 |
+
"""
|
6 |
+
|
7 |
+
import torch
|
8 |
+
from transformers import T5Tokenizer, T5ForConditionalGeneration
|
9 |
+
|
10 |
+
def load_model():
|
11 |
+
"""Load the T5 V0.3 model"""
|
12 |
+
print("🤖 Loading T5 Prompt Enhancer V0.3...")
|
13 |
+
|
14 |
+
tokenizer = T5Tokenizer.from_pretrained(".")
|
15 |
+
model = T5ForConditionalGeneration.from_pretrained(".")
|
16 |
+
|
17 |
+
if torch.cuda.is_available():
|
18 |
+
model = model.cuda()
|
19 |
+
print("✅ Model loaded on GPU")
|
20 |
+
else:
|
21 |
+
print("✅ Model loaded on CPU")
|
22 |
+
|
23 |
+
return model, tokenizer
|
24 |
+
|
25 |
+
def enhance_prompt(model, tokenizer, text, style="clean"):
|
26 |
+
"""Generate enhanced prompt with style control"""
|
27 |
+
|
28 |
+
style_prompts = {
|
29 |
+
"clean": f"Enhance this prompt (no lora): {text}",
|
30 |
+
"technical": f"Enhance this prompt (with lora): {text}",
|
31 |
+
"simplify": f"Simplify this prompt: {text}",
|
32 |
+
"standard": f"Enhance this prompt: {text}"
|
33 |
+
}
|
34 |
+
|
35 |
+
prompt = style_prompts[style]
|
36 |
+
inputs = tokenizer(prompt, return_tensors="pt", max_length=256, truncation=True)
|
37 |
+
|
38 |
+
if torch.cuda.is_available():
|
39 |
+
inputs = {k: v.cuda() for k, v in inputs.items()}
|
40 |
+
|
41 |
+
with torch.no_grad():
|
42 |
+
outputs = model.generate(
|
43 |
+
**inputs,
|
44 |
+
max_length=80,
|
45 |
+
num_beams=2,
|
46 |
+
repetition_penalty=2.0,
|
47 |
+
no_repeat_ngram_size=3,
|
48 |
+
pad_token_id=tokenizer.pad_token_id
|
49 |
+
)
|
50 |
+
|
51 |
+
return tokenizer.decode(outputs[0], skip_special_tokens=True)
|
52 |
+
|
53 |
+
def main():
|
54 |
+
"""Demo all four instruction types"""
|
55 |
+
|
56 |
+
# Load model
|
57 |
+
model, tokenizer = load_model()
|
58 |
+
|
59 |
+
# Test prompts
|
60 |
+
test_prompts = [
|
61 |
+
"woman in red dress",
|
62 |
+
"cat on chair",
|
63 |
+
"cyberpunk cityscape",
|
64 |
+
"masterpiece, best quality, ultra-detailed render of a fantasy dragon with golden scales"
|
65 |
+
]
|
66 |
+
|
67 |
+
styles = ["standard", "clean", "technical", "simplify"]
|
68 |
+
|
69 |
+
print("\n🎨 T5 Prompt Enhancer V0.3 Demo")
|
70 |
+
print("="*60)
|
71 |
+
|
72 |
+
for prompt in test_prompts:
|
73 |
+
print(f"\n📝 Input: '{prompt}'")
|
74 |
+
print("-" * 40)
|
75 |
+
|
76 |
+
for style in styles:
|
77 |
+
try:
|
78 |
+
result = enhance_prompt(model, tokenizer, prompt, style)
|
79 |
+
print(f"{style:>10}: {result}")
|
80 |
+
except Exception as e:
|
81 |
+
print(f"{style:>10}: ERROR - {e}")
|
82 |
+
|
83 |
+
print()
|
84 |
+
|
85 |
+
if __name__ == "__main__":
|
86 |
+
main()
|