Update app.py
Browse files
app.py
CHANGED
@@ -1,138 +1,180 @@
|
|
1 |
import gradio as gr
|
2 |
-
import
|
3 |
-
import
|
|
|
4 |
from io import BytesIO
|
5 |
-
import
|
6 |
-
import os
|
7 |
-
|
8 |
from train_tokenizer import train_tokenizer
|
9 |
-
from tokenizers import Tokenizer
|
10 |
from datasets import load_dataset
|
|
|
|
|
|
|
11 |
|
12 |
-
def
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
|
36 |
-
def
|
37 |
-
|
38 |
-
|
39 |
-
"""
|
40 |
-
encoded = tokenizer.encode(test_text)
|
41 |
-
decoded = tokenizer.decode(encoded.ids)
|
42 |
-
|
43 |
-
# Μέτρηση των Unknown tokens
|
44 |
-
unknown_tokens = sum(1 for t in encoded.tokens if t == "<unk>")
|
45 |
-
unknown_percent = (unknown_tokens / len(encoded.tokens) * 100) if encoded.tokens else 0
|
46 |
-
|
47 |
-
# Υπολογισμός μήκους των tokens
|
48 |
-
token_lengths = [len(t) for t in encoded.tokens]
|
49 |
-
avg_length = np.mean(token_lengths) if token_lengths else 0
|
50 |
-
|
51 |
-
# Έλεγχος κάλυψης κώδικα: παραδείγματα συμβόλων
|
52 |
-
code_symbols = ['{', '}', '(', ')', ';', '//', 'printf']
|
53 |
-
code_coverage = {sym: (sym in test_text and sym in encoded.tokens) for sym in code_symbols}
|
54 |
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
"token_length_distribution": img_buffer.getvalue()
|
72 |
-
}
|
73 |
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
-
def train_and_test(
|
77 |
-
|
78 |
-
raise gr.Error("Πρέπει να παρέχετε αρχεία ή όνομα dataset!")
|
79 |
-
|
80 |
try:
|
81 |
-
|
82 |
-
iterator = create_iterator(files, dataset_name, split)
|
83 |
|
84 |
-
# Προσθήκη progress bar
|
85 |
with gr.Progress() as progress:
|
86 |
-
progress(0.
|
87 |
tokenizer = train_tokenizer(iterator, vocab_size, min_freq)
|
88 |
-
|
89 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
|
91 |
except Exception as e:
|
92 |
raise gr.Error(f"Σφάλμα εκπαίδευσης: {str(e)}")
|
93 |
-
|
94 |
-
# Αποθήκευση και φόρτωση του tokenizer για επικύρωση
|
95 |
-
with tempfile.NamedTemporaryFile(delete=False, suffix=".json") as tmp:
|
96 |
-
tokenizer.save(tmp.name)
|
97 |
-
trained_tokenizer = Tokenizer.from_file(tmp.name)
|
98 |
-
os.unlink(tmp.name)
|
99 |
-
|
100 |
-
# Εκτενής επικύρωση με το δοκιμαστικό κείμενο
|
101 |
-
validation = enhanced_validation(trained_tokenizer, test_text)
|
102 |
-
|
103 |
-
return {
|
104 |
-
"validation_metrics": {k: v for k, v in validation.items() if k != "token_length_distribution"},
|
105 |
-
"histogram": validation["token_length_distribution"]
|
106 |
-
}
|
107 |
|
108 |
-
#
|
109 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
110 |
-
gr.Markdown("##
|
111 |
|
112 |
with gr.Row():
|
113 |
with gr.Column():
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
test_text = gr.Textbox(
|
123 |
-
value='
|
124 |
label="Test Text"
|
125 |
)
|
126 |
-
train_btn = gr.Button("Εκπαίδευση
|
127 |
-
|
128 |
with gr.Column():
|
129 |
-
|
130 |
-
|
|
|
131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
train_btn.click(
|
133 |
fn=train_and_test,
|
134 |
-
inputs=[
|
135 |
outputs=[results_json, results_plot]
|
136 |
)
|
|
|
137 |
if __name__ == "__main__":
|
138 |
demo.launch()
|
|
|
1 |
import gradio as gr
|
2 |
+
import requests
|
3 |
+
import json
|
4 |
+
import re
|
5 |
from io import BytesIO
|
6 |
+
import matplotlib.pyplot as plt
|
|
|
|
|
7 |
from train_tokenizer import train_tokenizer
|
|
|
8 |
from datasets import load_dataset
|
9 |
+
from tokenizers import Tokenizer
|
10 |
+
import tempfile
|
11 |
+
import os
|
12 |
|
13 |
+
def fetch_splits(dataset_name):
|
14 |
+
try:
|
15 |
+
response = requests.get(
|
16 |
+
f"https://datasets-server.huggingface.co/splits?dataset={dataset_name}",
|
17 |
+
timeout=10
|
18 |
+
)
|
19 |
+
response.raise_for_status()
|
20 |
+
data = response.json()
|
21 |
+
|
22 |
+
splits_info = {}
|
23 |
+
for split in data['splits']:
|
24 |
+
config = split['config']
|
25 |
+
split_name = split['split']
|
26 |
+
if config not in splits_info:
|
27 |
+
splits_info[config] = []
|
28 |
+
splits_info[config].append(split_name)
|
29 |
+
|
30 |
+
return {
|
31 |
+
"splits": splits_info,
|
32 |
+
"viewer_template": f"https://huggingface.co/datasets/{dataset_name}/embed/viewer/{{config}}/{{split}}"
|
33 |
+
}
|
34 |
+
except Exception as e:
|
35 |
+
raise gr.Error(f"Σφάλμα κατάττην ανάκτηση splits: {str(e)}")
|
36 |
|
37 |
+
def update_components(dataset_name):
|
38 |
+
if not dataset_name:
|
39 |
+
return [gr.Dropdown.update(choices=[], value=None), gr.Dropdown.update(choices=[]), gr.HTML.update(value="")]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
+
try:
|
42 |
+
splits_data = fetch_splits(dataset_name)
|
43 |
+
config_choices = list(splits_data['splits'].keys())
|
44 |
+
|
45 |
+
# Δημιουργία iframe preview για το πρώτο config
|
46 |
+
first_config = config_choices[0] if config_choices else None
|
47 |
+
iframe_html = f"""
|
48 |
+
<iframe
|
49 |
+
src="{splits_data['viewer_template'].format(config=first_config, split='train')}"
|
50 |
+
frameborder="0"
|
51 |
+
width="100%"
|
52 |
+
height="560px"
|
53 |
+
></iframe>
|
54 |
+
""" if first_config else "Δεν βρέθηκαν διαθέσιμα δεδομένα"
|
55 |
+
|
56 |
+
return [
|
57 |
+
gr.Dropdown.update(choices=config_choices, value=first_config),
|
58 |
+
gr.Dropdown.update(choices=splits_data['splits'].get(first_config, [])),
|
59 |
+
gr.HTML.update(value=iframe_html)
|
60 |
+
]
|
61 |
+
except Exception as e:
|
62 |
+
raise gr.Error(f"Σφάλμα: {str(e)}")
|
63 |
+
|
64 |
+
def update_split_choices(dataset_name, config):
|
65 |
+
if not dataset_name or not config:
|
66 |
+
return gr.Dropdown.update(choices=[])
|
67 |
|
68 |
+
try:
|
69 |
+
splits_data = fetch_splits(dataset_name)
|
70 |
+
return gr.Dropdown.update(choices=splits_data['splits'].get(config, []))
|
71 |
+
except:
|
72 |
+
return gr.Dropdown.update(choices=[])
|
|
|
|
|
73 |
|
74 |
+
def create_iterator(dataset_name, config, split):
|
75 |
+
try:
|
76 |
+
dataset = load_dataset(
|
77 |
+
dataset_name,
|
78 |
+
name=config,
|
79 |
+
split=split,
|
80 |
+
streaming=True
|
81 |
+
)
|
82 |
+
for example in dataset:
|
83 |
+
yield example.get('text', '')
|
84 |
+
except Exception as e:
|
85 |
+
raise gr.Error(f"Σφάλμα φόρτωσης dataset: {str(e)}")
|
86 |
|
87 |
+
def train_and_test(dataset_name, config, split, vocab_size, min_freq, test_text):
|
88 |
+
# Εκπαίδευση και validation logic
|
|
|
|
|
89 |
try:
|
90 |
+
iterator = create_iterator(dataset_name, config, split)
|
|
|
91 |
|
|
|
92 |
with gr.Progress() as progress:
|
93 |
+
progress(0.2, desc="Δημιουργία tokenizer...")
|
94 |
tokenizer = train_tokenizer(iterator, vocab_size, min_freq)
|
95 |
+
|
96 |
+
# Αποθήκευση και φόρτωση tokenizer
|
97 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=".json") as f:
|
98 |
+
tokenizer.save(f.name)
|
99 |
+
trained_tokenizer = Tokenizer.from_file(f.name)
|
100 |
+
os.unlink(f.name)
|
101 |
+
|
102 |
+
# Validation
|
103 |
+
encoded = trained_tokenizer.encode(test_text)
|
104 |
+
decoded = trained_tokenizer.decode(encoded.ids)
|
105 |
+
|
106 |
+
# Δημιουργία γραφήματος
|
107 |
+
token_lengths = [len(t) for t in encoded.tokens]
|
108 |
+
fig = plt.figure()
|
109 |
+
plt.hist(token_lengths, bins=20)
|
110 |
+
plt.xlabel('Μήκος Token')
|
111 |
+
plt.ylabel('Συχνότητα')
|
112 |
+
img_buffer = BytesIO()
|
113 |
+
plt.savefig(img_buffer, format='png')
|
114 |
+
plt.close()
|
115 |
+
|
116 |
+
return {
|
117 |
+
"Πρωτότυπο Κείμενο": test_text,
|
118 |
+
"Αποκωδικοποιημένο": decoded,
|
119 |
+
"Αριθμός Tokens": len(encoded.tokens),
|
120 |
+
"Αγνώστων Tokens": sum(1 for t in encoded.tokens if t == "<unk>")
|
121 |
+
}, img_buffer.getvalue()
|
122 |
|
123 |
except Exception as e:
|
124 |
raise gr.Error(f"Σφάλμα εκπαίδευσης: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
|
126 |
+
# Gradio Interface
|
127 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
128 |
+
gr.Markdown("## Wikipedia Tokenizer Trainer")
|
129 |
|
130 |
with gr.Row():
|
131 |
with gr.Column():
|
132 |
+
dataset_name = gr.Textbox(
|
133 |
+
label="Dataset Name",
|
134 |
+
value="wikimedia/wikipedia",
|
135 |
+
placeholder="π.χ. 'wikimedia/wikipedia'"
|
136 |
+
)
|
137 |
+
config = gr.Dropdown(
|
138 |
+
label="Config",
|
139 |
+
choices=[],
|
140 |
+
interactive=True
|
141 |
+
)
|
142 |
+
split = gr.Dropdown(
|
143 |
+
label="Split",
|
144 |
+
choices=[],
|
145 |
+
value="train"
|
146 |
+
)
|
147 |
+
vocab_size = gr.Slider(20000, 100000, value=50000, label="Μέγεθος Λεξιλογίου")
|
148 |
+
min_freq = gr.Slider(1, 100, value=3, label="Ελάχιστη Συχνότητα")
|
149 |
test_text = gr.Textbox(
|
150 |
+
value='Η Ακρόπολη είναι σύμβολο της αρχαίας ελληνικής πολιτισμικής κληρονομιάς.',
|
151 |
label="Test Text"
|
152 |
)
|
153 |
+
train_btn = gr.Button("Εκπαίδευση", variant="primary")
|
154 |
+
|
155 |
with gr.Column():
|
156 |
+
preview = gr.HTML(label="Dataset Preview")
|
157 |
+
results_json = gr.JSON(label="Αποτελέσματα")
|
158 |
+
results_plot = gr.Image(label="Κατανομή Μηκών Tokens")
|
159 |
|
160 |
+
# Event handlers
|
161 |
+
dataset_name.change(
|
162 |
+
fn=update_components,
|
163 |
+
inputs=dataset_name,
|
164 |
+
outputs=[config, split, preview]
|
165 |
+
)
|
166 |
+
|
167 |
+
config.change(
|
168 |
+
fn=update_split_choices,
|
169 |
+
inputs=[dataset_name, config],
|
170 |
+
outputs=split
|
171 |
+
)
|
172 |
+
|
173 |
train_btn.click(
|
174 |
fn=train_and_test,
|
175 |
+
inputs=[dataset_name, config, split, vocab_size, min_freq, test_text],
|
176 |
outputs=[results_json, results_plot]
|
177 |
)
|
178 |
+
|
179 |
if __name__ == "__main__":
|
180 |
demo.launch()
|