Spaces:
Build error
Build error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,235 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
3 |
+
import torch # Import torch for device management
|
4 |
+
import os # For file operations
|
5 |
+
|
6 |
+
# --- Configuration and Model Loading ---
|
7 |
+
# You can choose a different model here if you have access to more powerful ones.
|
8 |
+
# For larger models, ensure you have sufficient VRAM (GPU memory).
|
9 |
+
# For CPU, smaller models might be necessary or use quantization.
|
10 |
+
MODEL_NAME = "google/flan-t5-large" # Changed to 'large' for slightly better performance than 'base' and still manageable.
|
11 |
+
# If you have a powerful GPU, consider "google/flan-t5-xl" or even "google/flan-t5-xxl"
|
12 |
+
# For even larger models, consider using model.to(torch.bfloat16) or bitsandbytes for 4-bit loading if available.
|
13 |
+
|
14 |
+
try:
|
15 |
+
# Determine the device to use (GPU if available, else CPU)
|
16 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
17 |
+
print(f"Loading model on device: {device}")
|
18 |
+
|
19 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
20 |
+
# Load model with half-precision (float16) to save VRAM if on GPU
|
21 |
+
# Or load in 8-bit/4-bit if using libraries like bitsandbytes (requires installation)
|
22 |
+
if device == "cuda":
|
23 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME, torch_dtype=torch.float16).to(device)
|
24 |
+
else:
|
25 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME).to(device)
|
26 |
+
|
27 |
+
model.eval() # Set model to evaluation mode
|
28 |
+
print(f"Model '{MODEL_NAME}' loaded successfully.")
|
29 |
+
|
30 |
+
except Exception as e:
|
31 |
+
print(f"Error loading model: {e}")
|
32 |
+
print("Please check your internet connection, model name, and available resources (RAM/VRAM).")
|
33 |
+
# Exit or handle gracefully if model loading fails
|
34 |
+
tokenizer, model = None, None
|
35 |
+
|
36 |
+
# --- Prompt Engineering Functions (more structured) ---
|
37 |
+
|
38 |
+
def create_arabic_prompt(topic, style):
|
39 |
+
if style == "Blog Post (Descriptive)":
|
40 |
+
return f"اكتب مقالاً احترافياً بأسلوب شخصي عن: {topic}. ركز على التفاصيل، الوصف الجذاب، قدم نصائح عملية. اجعل النص منسقاً بفقرات وعناوين فرعية."
|
41 |
+
elif style == "Social Media Post (Short & Catchy)":
|
42 |
+
return f"اكتب منشوراً قصيراً وجذاباً ومثيراً للتفاعل عن: {topic}. أضف 2-3 إيموجي مناسبة واقترح 4 هاشتاغات شائعة. ابدأ بسؤال أو جملة جذابة."
|
43 |
+
else: # Video Script (Storytelling)
|
44 |
+
return f"اكتب سيناريو فيديو احترافي ومقنع عن: {topic}. اجعل الأسلوب قصصي وسردي، مقسماً إلى مشاهد رئيسية، مع اقتراح لقطات بصرية (B-roll) وأصوات (SFX) لكل مشهد. ركز على إثارة المشاعر."
|
45 |
+
|
46 |
+
def create_english_prompt(topic, style):
|
47 |
+
if style == "Blog Post (Descriptive)":
|
48 |
+
return f"Write a detailed and professional blog post about: {topic}. Focus on personal insights, vivid descriptions, and practical advice. Structure it with clear paragraphs and subheadings."
|
49 |
+
elif style == "Social Media Post (Short & Catchy)":
|
50 |
+
return f"Write a short, catchy, and engaging social media post about: {topic}. Include 2-3 relevant emojis and suggest 4 trending hashtags. Start with a hook question or statement."
|
51 |
+
else: # Video Script (Storytelling)
|
52 |
+
return f"Write a professional, compelling video script about: {topic}. Make it emotionally engaging and story-driven, divided into key scenes, with suggested visual shots (B-roll) and sound effects (SFX) for each scene."
|
53 |
+
|
54 |
+
# --- Content Generation Function ---
|
55 |
+
|
56 |
+
@torch.no_grad() # Disable gradient calculations for inference to save memory
|
57 |
+
def generate_content(topic, style_choice, lang_choice, length_choice, creativity, detail_level, diversity_penalty):
|
58 |
+
if tokenizer is None or model is None:
|
59 |
+
return "⚠️ Error: Model not loaded. Please check the console for details."
|
60 |
+
|
61 |
+
if not topic:
|
62 |
+
return "⚠️ Please enter a topic to generate content."
|
63 |
+
|
64 |
+
# Max length based on desired length and model's context window
|
65 |
+
# Flan-T5 has a context window of 512, so max_length should be within this.
|
66 |
+
if length_choice == "Short":
|
67 |
+
max_new_tokens = 150
|
68 |
+
min_new_tokens = 50
|
69 |
+
elif length_choice == "Medium":
|
70 |
+
max_new_tokens = 300
|
71 |
+
min_new_tokens = 100
|
72 |
+
else: # Long
|
73 |
+
max_new_tokens = 450 # Max for Flan-T5 effectively
|
74 |
+
min_new_tokens = 150
|
75 |
+
|
76 |
+
# Adjust generation parameters based on user input
|
77 |
+
temperature = creativity # Direct mapping
|
78 |
+
top_p = detail_level # Direct mapping, higher means more detail/diversity
|
79 |
+
no_repeat_ngram_size = diversity_penalty # Higher means less repetition
|
80 |
+
|
81 |
+
# Build the prompt
|
82 |
+
if lang_choice == "Arabic":
|
83 |
+
prompt = create_arabic_prompt(topic, style_choice)
|
84 |
+
else: # English
|
85 |
+
prompt = create_english_prompt(topic, style_choice)
|
86 |
+
|
87 |
+
# Add detail level instruction to prompt if high
|
88 |
+
if detail_level > 0.7: # Only if user explicitly wants high detail
|
89 |
+
prompt += " Ensure comprehensive coverage and rich descriptions."
|
90 |
+
if creativity > 0.8:
|
91 |
+
prompt += " Be highly creative and imaginative in your writing."
|
92 |
+
|
93 |
+
try:
|
94 |
+
inputs = tokenizer(prompt, return_tensors="pt", max_length=512, truncation=True).to(device)
|
95 |
+
|
96 |
+
outputs = model.generate(
|
97 |
+
**inputs,
|
98 |
+
max_new_tokens=max_new_tokens,
|
99 |
+
min_new_tokens=min_new_tokens,
|
100 |
+
num_beams=5, # Beam search for better quality
|
101 |
+
do_sample=True, # Enable sampling for creativity
|
102 |
+
temperature=temperature,
|
103 |
+
top_p=top_p,
|
104 |
+
top_k=50, # Consider top 50 words
|
105 |
+
no_repeat_ngram_size=no_repeat_ngram_size,
|
106 |
+
length_penalty=1.0, # Adjust to control output length
|
107 |
+
early_stopping=True
|
108 |
+
)
|
109 |
+
content = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
110 |
+
|
111 |
+
return content
|
112 |
+
except RuntimeError as e:
|
113 |
+
if "out of memory" in str(e):
|
114 |
+
return "⚠️ Generation failed: Out of memory. Try a shorter length, a less complex model, or restart the application if on GPU."
|
115 |
+
return f"⚠️ Generation failed due as runtime error: {str(e)}"
|
116 |
+
except Exception as e:
|
117 |
+
return f"⚠️ An unexpected error occurred during generation: {str(e)}"
|
118 |
+
|
119 |
+
# --- Gradio Interface ---
|
120 |
+
|
121 |
+
# Custom CSS for a more polished look
|
122 |
+
custom_css = """
|
123 |
+
h1, h2, h3 { color: #4B0082; } /* Dark Purple */
|
124 |
+
.gradio-container {
|
125 |
+
background-color: #F8F0FF; /* Light Lavender */
|
126 |
+
font-family: 'Segoe UI', sans-serif;
|
127 |
+
}
|
128 |
+
.gr-button {
|
129 |
+
background-color: #8A2BE2; /* Blue Violet */
|
130 |
+
color: white;
|
131 |
+
border-radius: 10px;
|
132 |
+
padding: 10px 20px;
|
133 |
+
font-size: 1.1em;
|
134 |
+
}
|
135 |
+
.gr-button:hover {
|
136 |
+
background-color: #9370DB; /* Medium Purple */
|
137 |
+
}
|
138 |
+
.gr-text-input, .gr-textarea {
|
139 |
+
border: 1px solid #DDA0DD; /* Plum */
|
140 |
+
border-radius: 8px;
|
141 |
+
padding: 10px;
|
142 |
+
}
|
143 |
+
.gradio-radio input:checked + label {
|
144 |
+
background-color: #DA70D6 !important; /* Orchid */
|
145 |
+
color: white !important;
|
146 |
+
}
|
147 |
+
.gradio-radio label {
|
148 |
+
border: 1px solid #DDA0DD;
|
149 |
+
border-radius: 8px;
|
150 |
+
padding: 8px 15px;
|
151 |
+
}
|
152 |
+
"""
|
153 |
+
|
154 |
+
with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as iface:
|
155 |
+
gr.Markdown("# ✨ AI Content Creation Studio")
|
156 |
+
gr.Markdown("## Generate professional blogs, social media posts, or video scripts in seconds!")
|
157 |
+
|
158 |
+
with gr.Row():
|
159 |
+
with gr.Column(scale=2):
|
160 |
+
topic = gr.Textbox(
|
161 |
+
label="Topic / الموضوع",
|
162 |
+
placeholder="e.g., The Future of AI in Healthcare / مثال: مستقبل الذكاء الاصطناعي في الرعاية الصحية",
|
163 |
+
lines=2
|
164 |
+
)
|
165 |
+
|
166 |
+
with gr.Accordion("Advanced Settings", open=False):
|
167 |
+
with gr.Row():
|
168 |
+
creativity = gr.Slider(
|
169 |
+
minimum=0.1, maximum=1.0, value=0.7, step=0.1,
|
170 |
+
label="Creativity (Temperature)",
|
171 |
+
info="Higher values lead to more creative, less predictable text. Lower values are more focused."
|
172 |
+
)
|
173 |
+
detail_level = gr.Slider(
|
174 |
+
minimum=0.1, maximum=1.0, value=0.9, step=0.1,
|
175 |
+
label="Detail Level (Top-p Sampling)",
|
176 |
+
info="Higher values allow for more diverse and detailed vocabulary. Lower values prune less likely words."
|
177 |
+
)
|
178 |
+
with gr.Row():
|
179 |
+
diversity_penalty = gr.Slider(
|
180 |
+
minimum=1, maximum=5, value=2, step=1,
|
181 |
+
label="Repetition Penalty (N-gram)",
|
182 |
+
info="Higher values reduce the chance of repeating the same phrases or words. Set to 1 for no penalty."
|
183 |
+
)
|
184 |
+
|
185 |
+
with gr.Column(scale=1):
|
186 |
+
with gr.Group():
|
187 |
+
style_choice = gr.Radio(
|
188 |
+
["Blog Post (Descriptive)", "Social Media Post (Short & Catchy)", "Video Script (Storytelling)"],
|
189 |
+
label="Content Style / نوع المحتوى",
|
190 |
+
value="Blog Post (Descriptive)",
|
191 |
+
interactive=True
|
192 |
+
)
|
193 |
+
with gr.Group():
|
194 |
+
lang_choice = gr.Radio(
|
195 |
+
["English", "Arabic"],
|
196 |
+
label="Language / اللغة",
|
197 |
+
value="English",
|
198 |
+
interactive=True
|
199 |
+
)
|
200 |
+
with gr.Group():
|
201 |
+
length_choice = gr.Radio(
|
202 |
+
["Short", "Medium", "Long"],
|
203 |
+
label="Content Length / طول النص",
|
204 |
+
value="Medium",
|
205 |
+
interactive=True
|
206 |
+
)
|
207 |
+
gr.Markdown("*(Note: 'Long' is relative to model capabilities, max ~450 words)*")
|
208 |
+
|
209 |
+
btn = gr.Button("🚀 Generate Content", variant="primary")
|
210 |
+
|
211 |
+
output = gr.Textbox(label="Generated Content", lines=20, interactive=True)
|
212 |
+
|
213 |
+
# Download button logic
|
214 |
+
def download_file(content):
|
215 |
+
if content and not content.startswith("⚠️"): # Only provide file if content is valid
|
216 |
+
file_path = "generated_content.txt"
|
217 |
+
with open(file_path, "w", encoding="utf-8") as f:
|
218 |
+
f.write(content)
|
219 |
+
return file_path
|
220 |
+
return None # Return None if no valid content to download
|
221 |
+
|
222 |
+
download_button = gr.DownloadButton("⬇️ Download Content", file_path=None, interactive=False)
|
223 |
+
|
224 |
+
# Event handlers
|
225 |
+
btn.click(
|
226 |
+
fn=generate_content,
|
227 |
+
inputs=[topic, style_choice, lang_choice, length_choice, creativity, detail_level, diversity_penalty],
|
228 |
+
outputs=output
|
229 |
+
)
|
230 |
+
|
231 |
+
# Enable download button only when there's valid content
|
232 |
+
output.change(fn=download_file, inputs=[output], outputs=[download_button])
|
233 |
+
|
234 |
+
if __name__ == "__main__":
|
235 |
+
iface.launch()
|