Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,20 +1,11 @@
|
|
1 |
import streamlit as st
|
2 |
import numpy as np
|
3 |
-
import random
|
4 |
import torch
|
5 |
-
import transformers
|
6 |
from transformers import GPT2LMHeadModel, GPT2Tokenizer, Trainer, TrainingArguments, DataCollatorForLanguageModeling
|
7 |
from datasets import Dataset
|
8 |
-
from huggingface_hub import HfApi
|
9 |
-
import os
|
10 |
-
import traceback
|
11 |
-
from contextlib import contextmanager
|
12 |
-
import plotly.graph_objects as go
|
13 |
-
import plotly.express as px
|
14 |
-
from datetime import datetime
|
15 |
import time
|
16 |
-
import
|
17 |
-
import
|
18 |
|
19 |
# Advanced Cyberpunk Styling
|
20 |
def setup_advanced_cyberpunk_style():
|
@@ -22,352 +13,113 @@ def setup_advanced_cyberpunk_style():
|
|
22 |
<style>
|
23 |
@import url('https://fonts.googleapis.com/css2?family=Orbitron:wght@400;500;700&display=swap');
|
24 |
@import url('https://fonts.googleapis.com/css2?family=Share+Tech+Mono&display=swap');
|
25 |
-
|
26 |
-
.stApp {
|
27 |
-
background: linear-gradient(
|
28 |
-
45deg,
|
29 |
-
rgba(0, 0, 0, 0.9) 0%,
|
30 |
-
rgba(0, 30, 60, 0.9) 50%,
|
31 |
-
rgba(0, 0, 0, 0.9) 100%
|
32 |
-
);
|
33 |
-
color: #00ff9d;
|
34 |
-
}
|
35 |
-
|
36 |
-
.main-title {
|
37 |
-
font-family: 'Orbitron', sans-serif;
|
38 |
-
background: linear-gradient(45deg, #00ff9d, #00b8ff);
|
39 |
-
-webkit-background-clip: text;
|
40 |
-
-webkit-text-fill-color: transparent;
|
41 |
-
text-align: center;
|
42 |
-
font-size: 3.5em;
|
43 |
-
margin-bottom: 30px;
|
44 |
-
text-transform: uppercase;
|
45 |
-
letter-spacing: 3px;
|
46 |
-
animation: glow 2s ease-in-out infinite alternate;
|
47 |
-
}
|
48 |
-
|
49 |
-
@keyframes glow {
|
50 |
-
from {
|
51 |
-
text-shadow: 0 0 5px #00ff9d, 0 0 10px #00ff9d, 0 0 15px #00ff9d;
|
52 |
-
}
|
53 |
-
to {
|
54 |
-
text-shadow: 0 0 10px #00b8ff, 0 0 20px #00b8ff, 0 0 30px #00b8ff;
|
55 |
-
}
|
56 |
-
}
|
57 |
-
|
58 |
-
.cyber-box {
|
59 |
-
background: rgba(0, 0, 0, 0.7);
|
60 |
-
border: 2px solid #00ff9d;
|
61 |
-
border-radius: 10px;
|
62 |
-
padding: 20px;
|
63 |
-
margin: 10px 0;
|
64 |
-
position: relative;
|
65 |
-
overflow: hidden;
|
66 |
-
}
|
67 |
-
|
68 |
-
.cyber-box::before {
|
69 |
-
content: '';
|
70 |
-
position: absolute;
|
71 |
-
top: -2px;
|
72 |
-
left: -2px;
|
73 |
-
right: -2px;
|
74 |
-
bottom: -2px;
|
75 |
-
background: linear-gradient(45deg, #00ff9d, #00b8ff);
|
76 |
-
z-index: -1;
|
77 |
-
filter: blur(10px);
|
78 |
-
opacity: 0.5;
|
79 |
-
}
|
80 |
-
|
81 |
-
.metric-container {
|
82 |
-
background: rgba(0, 0, 0, 0.8);
|
83 |
-
border: 2px solid #00ff9d;
|
84 |
-
border-radius: 10px;
|
85 |
-
padding: 20px;
|
86 |
-
margin: 10px 0;
|
87 |
-
position: relative;
|
88 |
-
overflow: hidden;
|
89 |
-
transition: all 0.3s ease;
|
90 |
-
}
|
91 |
-
|
92 |
-
.metric-container:hover {
|
93 |
-
transform: translateY(-5px);
|
94 |
-
box-shadow: 0 5px 15px rgba(0, 255, 157, 0.3);
|
95 |
-
}
|
96 |
-
|
97 |
-
.status-text {
|
98 |
-
font-family: 'Share Tech Mono', monospace;
|
99 |
-
color: #00ff9d;
|
100 |
-
font-size: 1.2em;
|
101 |
-
margin: 0;
|
102 |
-
text-shadow: 0 0 5px #00ff9d;
|
103 |
-
}
|
104 |
-
|
105 |
-
.sidebar .stSelectbox, .sidebar .stSlider {
|
106 |
-
background-color: rgba(0, 0, 0, 0.5);
|
107 |
-
border-radius: 5px;
|
108 |
-
padding: 15px;
|
109 |
-
margin: 10px 0;
|
110 |
-
border: 1px solid #00ff9d;
|
111 |
-
}
|
112 |
-
|
113 |
-
.stButton>button {
|
114 |
-
font-family: 'Orbitron', sans-serif;
|
115 |
-
background: linear-gradient(45deg, #00ff9d, #00b8ff);
|
116 |
-
color: black;
|
117 |
-
border: none;
|
118 |
-
padding: 15px 30px;
|
119 |
-
border-radius: 5px;
|
120 |
-
text-transform: uppercase;
|
121 |
-
font-weight: bold;
|
122 |
-
letter-spacing: 2px;
|
123 |
-
transition: all 0.3s ease;
|
124 |
-
position: relative;
|
125 |
-
overflow: hidden;
|
126 |
-
}
|
127 |
-
|
128 |
-
.stButton>button:hover {
|
129 |
-
transform: scale(1.05);
|
130 |
-
box-shadow: 0 0 20px rgba(0, 255, 157, 0.5);
|
131 |
-
}
|
132 |
-
|
133 |
-
.stButton>button::after {
|
134 |
-
content: '';
|
135 |
-
position: absolute;
|
136 |
-
top: -50%;
|
137 |
-
left: -50%;
|
138 |
-
width: 200%;
|
139 |
-
height: 200%;
|
140 |
-
background: linear-gradient(
|
141 |
-
45deg,
|
142 |
-
transparent,
|
143 |
-
rgba(255, 255, 255, 0.1),
|
144 |
-
transparent
|
145 |
-
);
|
146 |
-
transform: rotate(45deg);
|
147 |
-
animation: shine 3s infinite;
|
148 |
-
}
|
149 |
-
|
150 |
-
@keyframes shine {
|
151 |
-
0% {
|
152 |
-
transform: translateX(-100%) rotate(45deg);
|
153 |
-
}
|
154 |
-
100% {
|
155 |
-
transform: translateX(100%) rotate(45deg);
|
156 |
-
}
|
157 |
-
}
|
158 |
-
|
159 |
-
.custom-info-box {
|
160 |
-
background: rgba(0, 255, 157, 0.1);
|
161 |
-
border-left: 5px solid #00ff9d;
|
162 |
-
padding: 15px;
|
163 |
-
margin: 10px 0;
|
164 |
-
font-family: 'Share Tech Mono', monospace;
|
165 |
-
}
|
166 |
-
|
167 |
-
.progress-bar-container {
|
168 |
-
width: 100%;
|
169 |
-
height: 30px;
|
170 |
-
background: rgba(0, 0, 0, 0.5);
|
171 |
-
border: 2px solid #00ff9d;
|
172 |
-
border-radius: 15px;
|
173 |
-
overflow: hidden;
|
174 |
-
position: relative;
|
175 |
-
}
|
176 |
-
|
177 |
-
.progress-bar {
|
178 |
-
height: 100%;
|
179 |
-
background: linear-gradient(45deg, #00ff9d, #00b8ff);
|
180 |
-
transition: width 0.3s ease;
|
181 |
-
}
|
182 |
</style>
|
183 |
""", unsafe_allow_html=True)
|
184 |
|
185 |
-
#
|
186 |
-
def
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
batched=True
|
196 |
-
)
|
197 |
-
tokenized_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'labels'])
|
198 |
-
return tokenized_dataset
|
199 |
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
mode='lines+markers',
|
206 |
-
name='Loss',
|
207 |
-
line=dict(color='#00ff9d', width=2),
|
208 |
-
marker=dict(size=8, symbol='diamond'),
|
209 |
-
))
|
210 |
-
|
211 |
-
fig.update_layout(
|
212 |
-
title={
|
213 |
-
'text': 'Training Progress',
|
214 |
-
'y':0.95,
|
215 |
-
'x':0.5,
|
216 |
-
'xanchor': 'center',
|
217 |
-
'yanchor': 'top',
|
218 |
-
'font': {'family': 'Orbitron', 'size': 24, 'color': '#00ff9d'}
|
219 |
-
},
|
220 |
-
paper_bgcolor='rgba(0,0,0,0.5)',
|
221 |
-
plot_bgcolor='rgba(0,0,0,0.3)',
|
222 |
-
font=dict(family='Share Tech Mono', color='#00ff9d'),
|
223 |
-
xaxis=dict(
|
224 |
-
title='Generation',
|
225 |
-
gridcolor='rgba(0,255,157,0.1)',
|
226 |
-
zerolinecolor='#00ff9d'
|
227 |
-
),
|
228 |
-
yaxis=dict(
|
229 |
-
title='Loss',
|
230 |
-
gridcolor='rgba(0,255,157,0.1)',
|
231 |
-
zerolinecolor='#00ff9d'
|
232 |
-
),
|
233 |
-
hovermode='x unified'
|
234 |
)
|
235 |
-
|
|
|
236 |
|
237 |
-
#
|
238 |
class TrainingDashboard:
|
239 |
def __init__(self):
|
240 |
self.metrics = {
|
241 |
'current_loss': 0,
|
242 |
'best_loss': float('inf'),
|
243 |
'generation': 0,
|
244 |
-
'individual': 0,
|
245 |
'start_time': time.time(),
|
246 |
'training_speed': 0
|
247 |
}
|
248 |
self.history = []
|
249 |
|
250 |
-
def update(self, loss, generation
|
251 |
self.metrics['current_loss'] = loss
|
252 |
self.metrics['generation'] = generation
|
253 |
-
self.metrics['individual'] = individual
|
254 |
if loss < self.metrics['best_loss']:
|
255 |
self.metrics['best_loss'] = loss
|
256 |
-
|
257 |
elapsed_time = time.time() - self.metrics['start_time']
|
258 |
-
self.metrics['training_speed'] =
|
259 |
-
self.history.append({
|
260 |
-
'loss': loss,
|
261 |
-
'timestamp': datetime.now().strftime('%H:%M:%S')
|
262 |
-
})
|
263 |
|
264 |
def display(self):
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
"""
|
275 |
-
|
276 |
-
|
277 |
-
self.metrics['individual'],
|
278 |
-
self.metrics['population_size']
|
279 |
-
), unsafe_allow_html=True)
|
280 |
-
|
281 |
-
with col2:
|
282 |
-
st.markdown("""
|
283 |
-
<div class="metric-container">
|
284 |
-
<h3 style="color: #00ff9d;">Performance</h3>
|
285 |
-
<p class="status-text">Current Loss: {:.4f}</p>
|
286 |
-
<p class="status-text">Best Loss: {:.4f}</p>
|
287 |
-
</div>
|
288 |
-
""".format(
|
289 |
-
self.metrics['current_loss'],
|
290 |
-
self.metrics['best_loss']
|
291 |
-
), unsafe_allow_html=True)
|
292 |
-
|
293 |
-
with col3:
|
294 |
-
st.markdown("""
|
295 |
-
<div class="metric-container">
|
296 |
-
<h3 style="color: #00ff9d;">Training Metrics</h3>
|
297 |
-
<p class="status-text">Speed: {:.2f} iter/s</p>
|
298 |
-
<p class="status-text">Runtime: {:.2f}m</p>
|
299 |
-
</div>
|
300 |
-
""".format(
|
301 |
-
self.metrics['training_speed'],
|
302 |
-
(time.time() - self.metrics['start_time']) / 60
|
303 |
-
), unsafe_allow_html=True)
|
304 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
305 |
def main():
|
306 |
setup_advanced_cyberpunk_style()
|
307 |
-
|
308 |
st.markdown('<h1 class="main-title">Neural Evolution GPT-2 Training Hub</h1>', unsafe_allow_html=True)
|
309 |
-
|
310 |
-
# Initialize dashboard
|
311 |
-
dashboard = TrainingDashboard()
|
312 |
-
|
313 |
-
# Advanced Sidebar
|
314 |
-
with st.sidebar:
|
315 |
-
st.markdown("""
|
316 |
-
<div style="text-align: center; padding: 20px;">
|
317 |
-
<h2 style="font-family: 'Orbitron'; color: #00ff9d;">Control Panel</h2>
|
318 |
-
</div>
|
319 |
-
""", unsafe_allow_html=True)
|
320 |
-
|
321 |
-
# Configuration Tabs
|
322 |
-
tab1, tab2, tab3 = st.tabs(["🔧 Setup", "⚙️ Parameters", "📊 Monitoring"])
|
323 |
-
|
324 |
-
with tab1:
|
325 |
-
hf_token = st.text_input("🔑 HuggingFace Token", type="password")
|
326 |
-
repo_name = st.text_input("📁 Repository Name", "my-gpt2-model")
|
327 |
-
data_source = st.selectbox('📊 Data Source', ('DEMO', 'Upload Text File'))
|
328 |
-
|
329 |
-
with tab2:
|
330 |
-
population_size = st.slider("Population Size", 4, 20, 6)
|
331 |
-
num_generations = st.slider("Generations", 1, 10, 3)
|
332 |
-
num_parents = st.slider("Parents", 2, population_size, 2)
|
333 |
-
mutation_rate = st.slider("Mutation Rate", 0.0, 1.0, 0.1)
|
334 |
-
|
335 |
-
# Advanced Parameters
|
336 |
-
with st.expander("🔬 Advanced Settings"):
|
337 |
-
learning_rate_min = st.number_input("Min Learning Rate", 1e-6, 1e-4, 1e-5)
|
338 |
-
learning_rate_max = st.number_input("Max Learning Rate", 1e-5, 1e-3, 5e-5)
|
339 |
-
batch_size_options = st.multiselect("Batch Sizes", [2, 4, 8, 16], default=[2, 4, 8])
|
340 |
-
|
341 |
-
with tab3:
|
342 |
-
st.markdown("""
|
343 |
-
<div class="cyber-box">
|
344 |
-
<h3 style="color: #00ff9d;">System Status</h3>
|
345 |
-
<p>GPU: {}</p>
|
346 |
-
<p>Memory Usage: {:.2f}GB</p>
|
347 |
-
</div>
|
348 |
-
""".format(
|
349 |
-
'CUDA' if torch.cuda.is_available() else 'CPU',
|
350 |
-
torch.cuda.memory_allocated() / 1e9 if torch.cuda.is_available() else 0
|
351 |
-
), unsafe_allow_html=True)
|
352 |
|
353 |
-
#
|
354 |
-
|
|
|
|
|
|
|
|
|
355 |
|
356 |
-
#
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
|
364 |
-
|
365 |
-
|
366 |
-
|
367 |
-
<div class="progress-bar-container">
|
368 |
-
<div class="progress-bar" style="width: {progress * 100}%"></div>
|
369 |
-
</div>
|
370 |
-
""", unsafe_allow_html=True)
|
371 |
|
372 |
if __name__ == "__main__":
|
373 |
-
main()
|
|
|
1 |
import streamlit as st
|
2 |
import numpy as np
|
|
|
3 |
import torch
|
|
|
4 |
from transformers import GPT2LMHeadModel, GPT2Tokenizer, Trainer, TrainingArguments, DataCollatorForLanguageModeling
|
5 |
from datasets import Dataset
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
import time
|
7 |
+
from datetime import datetime
|
8 |
+
import plotly.graph_objects as go
|
9 |
|
10 |
# Advanced Cyberpunk Styling
|
11 |
def setup_advanced_cyberpunk_style():
|
|
|
13 |
<style>
|
14 |
@import url('https://fonts.googleapis.com/css2?family=Orbitron:wght@400;500;700&display=swap');
|
15 |
@import url('https://fonts.googleapis.com/css2?family=Share+Tech+Mono&display=swap');
|
16 |
+
/* Additional styling as provided previously */
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
</style>
|
18 |
""", unsafe_allow_html=True)
|
19 |
|
20 |
+
# Initialize Model and Tokenizer
|
21 |
+
def initialize_model():
|
22 |
+
model = GPT2LMHeadModel.from_pretrained("gpt2")
|
23 |
+
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
|
24 |
+
return model, tokenizer
|
25 |
|
26 |
+
# Prepare Dataset
|
27 |
+
def prepare_dataset(data, tokenizer, block_size=128):
|
28 |
+
def tokenize_function(examples):
|
29 |
+
return tokenizer(examples['text'], truncation=True, max_length=block_size, padding='max_length')
|
|
|
|
|
|
|
|
|
30 |
|
31 |
+
raw_dataset = Dataset.from_dict({'text': data})
|
32 |
+
tokenized_dataset = raw_dataset.map(tokenize_function, batched=True, remove_columns=['text'])
|
33 |
+
tokenized_dataset = tokenized_dataset.map(
|
34 |
+
lambda examples: {'labels': examples['input_ids']},
|
35 |
+
batched=True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
)
|
37 |
+
tokenized_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'labels'])
|
38 |
+
return tokenized_dataset
|
39 |
|
40 |
+
# Training Dashboard Class
|
41 |
class TrainingDashboard:
|
42 |
def __init__(self):
|
43 |
self.metrics = {
|
44 |
'current_loss': 0,
|
45 |
'best_loss': float('inf'),
|
46 |
'generation': 0,
|
|
|
47 |
'start_time': time.time(),
|
48 |
'training_speed': 0
|
49 |
}
|
50 |
self.history = []
|
51 |
|
52 |
+
def update(self, loss, generation):
|
53 |
self.metrics['current_loss'] = loss
|
54 |
self.metrics['generation'] = generation
|
|
|
55 |
if loss < self.metrics['best_loss']:
|
56 |
self.metrics['best_loss'] = loss
|
|
|
57 |
elapsed_time = time.time() - self.metrics['start_time']
|
58 |
+
self.metrics['training_speed'] = generation / elapsed_time
|
59 |
+
self.history.append({'loss': loss, 'timestamp': datetime.now().strftime('%H:%M:%S')})
|
|
|
|
|
|
|
60 |
|
61 |
def display(self):
|
62 |
+
st.write(f"**Generation:** {self.metrics['generation']}")
|
63 |
+
st.write(f"**Current Loss:** {self.metrics['current_loss']:.4f}")
|
64 |
+
st.write(f"**Best Loss:** {self.metrics['best_loss']:.4f}")
|
65 |
+
st.write(f"**Training Speed:** {self.metrics['training_speed']:.2f} generations/sec")
|
66 |
+
|
67 |
+
# Display Progress Bar
|
68 |
+
def display_progress(progress):
|
69 |
+
st.markdown(f"""
|
70 |
+
<div class="progress-bar-container">
|
71 |
+
<div class="progress-bar" style="width: {progress * 100}%"></div>
|
72 |
+
</div>
|
73 |
+
""", unsafe_allow_html=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
|
75 |
+
# Fitness Calculation (Placeholder for actual loss computation)
|
76 |
+
def compute_loss(model, dataset):
|
77 |
+
# Placeholder for real loss computation with Trainer API or custom logic
|
78 |
+
trainer = Trainer(
|
79 |
+
model=model,
|
80 |
+
args=TrainingArguments(output_dir="./results", per_device_train_batch_size=2, num_train_epochs=1),
|
81 |
+
train_dataset=dataset,
|
82 |
+
data_collator=DataCollatorForLanguageModeling(tokenizer=model.config._name_or_path, mlm=False),
|
83 |
+
)
|
84 |
+
train_result = trainer.train()
|
85 |
+
return train_result.training_loss
|
86 |
+
|
87 |
+
# Training Loop with Loading Screen
|
88 |
+
def training_loop(dashboard, model, dataset, num_generations, population_size):
|
89 |
+
with st.spinner("Training in progress..."):
|
90 |
+
for generation in range(1, num_generations + 1):
|
91 |
+
# Simulated population loop
|
92 |
+
for individual in range(population_size):
|
93 |
+
loss = compute_loss(model, dataset)
|
94 |
+
dashboard.update(loss, generation)
|
95 |
+
progress = generation / num_generations
|
96 |
+
display_progress(progress)
|
97 |
+
dashboard.display()
|
98 |
+
time.sleep(1) # Simulate delay for each individual training
|
99 |
+
|
100 |
+
# Main Function
|
101 |
def main():
|
102 |
setup_advanced_cyberpunk_style()
|
|
|
103 |
st.markdown('<h1 class="main-title">Neural Evolution GPT-2 Training Hub</h1>', unsafe_allow_html=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
+
# Load Model and Tokenizer
|
106 |
+
model, tokenizer = initialize_model()
|
107 |
+
|
108 |
+
# Prepare Data
|
109 |
+
data = ["Sample training text"] * 10 # Replace with real data
|
110 |
+
train_dataset = prepare_dataset(data, tokenizer)
|
111 |
|
112 |
+
# Initialize Dashboard
|
113 |
+
dashboard = TrainingDashboard()
|
114 |
+
|
115 |
+
# Sidebar Configuration
|
116 |
+
st.sidebar.markdown("### Training Parameters")
|
117 |
+
num_generations = st.sidebar.slider("Generations", 1, 20, 5)
|
118 |
+
population_size = st.sidebar.slider("Population Size", 4, 20, 6)
|
119 |
+
|
120 |
+
# Run Training
|
121 |
+
if st.button("Start Training"):
|
122 |
+
training_loop(dashboard, model, train_dataset, num_generations, population_size)
|
|
|
|
|
|
|
|
|
123 |
|
124 |
if __name__ == "__main__":
|
125 |
+
main()
|