Update app.py
Browse files
app.py
CHANGED
|
@@ -9,6 +9,22 @@ from torch.nn import functional as F
|
|
| 9 |
import tiktoken
|
| 10 |
import gradio as gr
|
| 11 |
import asyncio
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
# Define the model architecture
|
| 13 |
class GPTConfig:
|
| 14 |
def __init__(self):
|
|
@@ -134,7 +150,7 @@ import gradio as gr
|
|
| 134 |
# [Your existing model code remains unchanged]
|
| 135 |
|
| 136 |
# Modify the generate_text function to be asynchronous
|
| 137 |
-
async def generate_text(prompt, max_length=
|
| 138 |
input_ids = torch.tensor(enc.encode(prompt)).unsqueeze(0)
|
| 139 |
generated = []
|
| 140 |
|
|
@@ -151,13 +167,16 @@ async def generate_text(prompt, max_length=100, temperature=0.7, top_k=50):
|
|
| 151 |
input_ids = torch.cat([input_ids, next_token], dim=-1)
|
| 152 |
generated.append(next_token.item())
|
| 153 |
|
| 154 |
-
|
|
|
|
| 155 |
|
| 156 |
-
if next_token.item() == enc.encode('\n')[0] and len(generated) >
|
| 157 |
break
|
| 158 |
|
| 159 |
-
await asyncio.sleep(0.
|
| 160 |
|
|
|
|
|
|
|
| 161 |
# Modify the gradio_generate function to be asynchronous
|
| 162 |
async def gradio_generate(prompt, max_length, temperature, top_k):
|
| 163 |
output = ""
|
|
@@ -178,20 +197,23 @@ css = """
|
|
| 178 |
</style>
|
| 179 |
"""
|
| 180 |
|
| 181 |
-
# 6. Gradio App Definition
|
| 182 |
with gr.Blocks(css=css) as demo:
|
| 183 |
-
gr.HTML("<div class='header'><h1>🌟 GPT-2
|
| 184 |
|
| 185 |
with gr.Row():
|
| 186 |
with gr.Column(scale=3):
|
| 187 |
-
prompt = gr.Textbox(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 188 |
with gr.Column(scale=1):
|
| 189 |
-
generate_btn = gr.Button("Generate", elem_classes="generate-btn")
|
| 190 |
|
| 191 |
with gr.Row():
|
| 192 |
-
max_length = gr.Slider(minimum=
|
| 193 |
-
temperature = gr.Slider(minimum=0.1, maximum=1.0, value=0.
|
| 194 |
-
top_k = gr.Slider(minimum=1, maximum=100, value=
|
| 195 |
|
| 196 |
output = gr.Markdown(elem_classes="output-box")
|
| 197 |
|
|
@@ -201,6 +223,6 @@ with gr.Blocks(css=css) as demo:
|
|
| 201 |
outputs=output
|
| 202 |
)
|
| 203 |
|
| 204 |
-
|
| 205 |
if __name__ == "__main__":
|
| 206 |
demo.launch()
|
|
|
|
| 9 |
import tiktoken
|
| 10 |
import gradio as gr
|
| 11 |
import asyncio
|
| 12 |
+
|
| 13 |
+
# Add the post-processing function here
|
| 14 |
+
def post_process_text(text):
|
| 15 |
+
# Ensure the text starts with a capital letter
|
| 16 |
+
text = text.capitalize()
|
| 17 |
+
|
| 18 |
+
# Remove any incomplete sentences at the end
|
| 19 |
+
sentences = text.split('.')
|
| 20 |
+
complete_sentences = sentences[:-1] if len(sentences) > 1 else sentences
|
| 21 |
+
|
| 22 |
+
# Rejoin sentences and add a period if missing
|
| 23 |
+
processed_text = '. '.join(complete_sentences)
|
| 24 |
+
if not processed_text.endswith('.'):
|
| 25 |
+
processed_text += '.'
|
| 26 |
+
|
| 27 |
+
return processed_text
|
| 28 |
# Define the model architecture
|
| 29 |
class GPTConfig:
|
| 30 |
def __init__(self):
|
|
|
|
| 150 |
# [Your existing model code remains unchanged]
|
| 151 |
|
| 152 |
# Modify the generate_text function to be asynchronous
|
| 153 |
+
async def generate_text(prompt, max_length=432, temperature=0.8, top_k=40):
|
| 154 |
input_ids = torch.tensor(enc.encode(prompt)).unsqueeze(0)
|
| 155 |
generated = []
|
| 156 |
|
|
|
|
| 167 |
input_ids = torch.cat([input_ids, next_token], dim=-1)
|
| 168 |
generated.append(next_token.item())
|
| 169 |
|
| 170 |
+
next_token_str = enc.decode([next_token.item()])
|
| 171 |
+
yield next_token_str
|
| 172 |
|
| 173 |
+
if next_token.item() == enc.encode('\n')[0] and len(generated) > 100:
|
| 174 |
break
|
| 175 |
|
| 176 |
+
await asyncio.sleep(0.02) # Slightly faster typing effect
|
| 177 |
|
| 178 |
+
if len(generated) == max_length:
|
| 179 |
+
yield "... (output truncated due to length)"
|
| 180 |
# Modify the gradio_generate function to be asynchronous
|
| 181 |
async def gradio_generate(prompt, max_length, temperature, top_k):
|
| 182 |
output = ""
|
|
|
|
| 197 |
</style>
|
| 198 |
"""
|
| 199 |
|
|
|
|
| 200 |
with gr.Blocks(css=css) as demo:
|
| 201 |
+
gr.HTML("<div class='header'><h1>🌟 GPT-2 Storyteller</h1></div>")
|
| 202 |
|
| 203 |
with gr.Row():
|
| 204 |
with gr.Column(scale=3):
|
| 205 |
+
prompt = gr.Textbox(
|
| 206 |
+
placeholder="Start your story here (e.g., 'Once upon a time in a magical forest...')",
|
| 207 |
+
label="Story Prompt",
|
| 208 |
+
elem_classes="user-input"
|
| 209 |
+
)
|
| 210 |
with gr.Column(scale=1):
|
| 211 |
+
generate_btn = gr.Button("Generate Story", elem_classes="generate-btn")
|
| 212 |
|
| 213 |
with gr.Row():
|
| 214 |
+
max_length = gr.Slider(minimum=50, maximum=500, value=432, step=1, label="Max Length")
|
| 215 |
+
temperature = gr.Slider(minimum=0.1, maximum=1.0, value=0.8, step=0.1, label="Temperature")
|
| 216 |
+
top_k = gr.Slider(minimum=1, maximum=100, value=40, step=1, label="Top-k")
|
| 217 |
|
| 218 |
output = gr.Markdown(elem_classes="output-box")
|
| 219 |
|
|
|
|
| 223 |
outputs=output
|
| 224 |
)
|
| 225 |
|
| 226 |
+
|
| 227 |
if __name__ == "__main__":
|
| 228 |
demo.launch()
|