PhotographerAlpha7
commited on
Commit
•
5fb8b32
1
Parent(s):
e3778e7
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,205 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import pandas as pd
|
3 |
+
from datasets import Dataset
|
4 |
+
from transformers import GPT2Tokenizer, GPT2LMHeadModel, Trainer, TrainingArguments
|
5 |
+
import torch
|
6 |
+
import os
|
7 |
+
import matplotlib.pyplot as plt
|
8 |
+
import json
|
9 |
+
import io
|
10 |
+
|
11 |
+
# Variables globales pour stocker les colonnes détectées
|
12 |
+
columns = []
|
13 |
+
|
14 |
+
# Fonction pour lire le fichier et détecter les colonnes
|
15 |
+
def read_file(data_file):
|
16 |
+
global columns
|
17 |
+
try:
|
18 |
+
# Charger les données
|
19 |
+
file_extension = os.path.splitext(data_file.name)[1]
|
20 |
+
if file_extension == '.csv':
|
21 |
+
df = pd.read_csv(data_file.name)
|
22 |
+
elif file_extension == '.json':
|
23 |
+
df = pd.read_json(data_file.name)
|
24 |
+
elif file_extension == '.xlsx':
|
25 |
+
df = pd.read_excel(data_file.name)
|
26 |
+
else:
|
27 |
+
return "Invalid file format. Please upload a CSV, JSON, or Excel file."
|
28 |
+
|
29 |
+
# Détecter les colonnes
|
30 |
+
columns = df.columns.tolist()
|
31 |
+
return columns
|
32 |
+
except Exception as e:
|
33 |
+
return f"An error occurred: {str(e)}"
|
34 |
+
|
35 |
+
# Fonction pour entraîner le modèle
|
36 |
+
def train_model(data_file, model_name, epochs, batch_size, learning_rate, output_dir, prompt_col, description_col):
|
37 |
+
try:
|
38 |
+
# Charger les données
|
39 |
+
file_extension = os.path.splitext(data_file.name)[1]
|
40 |
+
if file_extension == '.csv':
|
41 |
+
df = pd.read_csv(data_file.name)
|
42 |
+
elif file_extension == '.json':
|
43 |
+
df = pd.read_json(data_file.name)
|
44 |
+
elif file_extension == '.xlsx':
|
45 |
+
df = pd.read_excel(data_file.name)
|
46 |
+
|
47 |
+
# Prévisualisation des données
|
48 |
+
preview = df.head().to_string(index=False)
|
49 |
+
|
50 |
+
# Préparer le texte d'entraînement
|
51 |
+
df['text'] = df[prompt_col] + ': ' + df[description_col]
|
52 |
+
dataset = Dataset.from_pandas(df[['text']])
|
53 |
+
|
54 |
+
# Initialiser le tokenizer et le modèle GPT-2
|
55 |
+
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
|
56 |
+
model = GPT2LMHeadModel.from_pretrained(model_name)
|
57 |
+
|
58 |
+
# Ajouter un token de padding si nécessaire
|
59 |
+
if tokenizer.pad_token is None:
|
60 |
+
tokenizer.add_special_tokens({'pad_token': '[PAD]'})
|
61 |
+
model.resize_token_embeddings(len(tokenizer))
|
62 |
+
|
63 |
+
# Tokenizer les données
|
64 |
+
def tokenize_function(examples):
|
65 |
+
tokens = tokenizer(examples['text'], padding="max_length", truncation=True, max_length=128)
|
66 |
+
tokens['labels'] = tokens['input_ids'].copy()
|
67 |
+
return tokens
|
68 |
+
|
69 |
+
tokenized_datasets = dataset.map(tokenize_function, batched=True)
|
70 |
+
|
71 |
+
# Ajustement des hyperparamètres
|
72 |
+
training_args = TrainingArguments(
|
73 |
+
output_dir=output_dir,
|
74 |
+
overwrite_output_dir=True,
|
75 |
+
num_train_epochs=int(epochs),
|
76 |
+
per_device_train_batch_size=int(batch_size),
|
77 |
+
per_device_eval_batch_size=int(batch_size),
|
78 |
+
warmup_steps=1000,
|
79 |
+
weight_decay=0.01,
|
80 |
+
learning_rate=float(learning_rate),
|
81 |
+
logging_dir="./logs",
|
82 |
+
logging_steps=10,
|
83 |
+
save_steps=500,
|
84 |
+
save_total_limit=2,
|
85 |
+
evaluation_strategy="steps",
|
86 |
+
eval_steps=500,
|
87 |
+
load_best_model_at_end=True,
|
88 |
+
metric_for_best_model="eval_loss"
|
89 |
+
)
|
90 |
+
|
91 |
+
# Configuration du Trainer
|
92 |
+
trainer = Trainer(
|
93 |
+
model=model,
|
94 |
+
args=training_args,
|
95 |
+
train_dataset=tokenized_datasets,
|
96 |
+
eval_dataset=tokenized_datasets,
|
97 |
+
)
|
98 |
+
|
99 |
+
# Entraînement et évaluation
|
100 |
+
trainer.train()
|
101 |
+
eval_results = trainer.evaluate()
|
102 |
+
|
103 |
+
# Sauvegarder le modèle fine-tuné
|
104 |
+
model.save_pretrained(output_dir)
|
105 |
+
tokenizer.save_pretrained(output_dir)
|
106 |
+
|
107 |
+
# Générer un graphique des pertes d'entraînement et de validation
|
108 |
+
train_loss = [x['loss'] for x in trainer.state.log_history if 'loss' in x]
|
109 |
+
eval_loss = [x['eval_loss'] for x in trainer.state.log_history if 'eval_loss' in x]
|
110 |
+
plt.plot(train_loss, label='Training Loss')
|
111 |
+
plt.plot(eval_loss, label='Validation Loss')
|
112 |
+
plt.xlabel('Steps')
|
113 |
+
plt.ylabel('Loss')
|
114 |
+
plt.title('Training and Validation Loss')
|
115 |
+
plt.legend()
|
116 |
+
plt.savefig(os.path.join(output_dir, 'training_eval_loss.png'))
|
117 |
+
|
118 |
+
return f"Training completed successfully.\nPreview of data:\n{preview}", eval_results
|
119 |
+
except Exception as e:
|
120 |
+
return f"An error occurred: {str(e)}"
|
121 |
+
|
122 |
+
# Fonction de génération de texte
|
123 |
+
def generate_text(prompt, temperature, top_k, max_length, repetition_penalty, use_comma):
|
124 |
+
try:
|
125 |
+
model_name = "./fine-tuned-gpt2"
|
126 |
+
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
|
127 |
+
model = GPT2LMHeadModel.from_pretrained(model_name)
|
128 |
+
|
129 |
+
if use_comma:
|
130 |
+
prompt = prompt.replace('.', ',')
|
131 |
+
|
132 |
+
inputs = tokenizer(prompt, return_tensors="pt", padding=True)
|
133 |
+
attention_mask = inputs.attention_mask
|
134 |
+
outputs = model.generate(
|
135 |
+
inputs.input_ids,
|
136 |
+
attention_mask=attention_mask,
|
137 |
+
max_length=int(max_length),
|
138 |
+
temperature=float(temperature),
|
139 |
+
top_k=int(top_k),
|
140 |
+
repetition_penalty=float(repetition_penalty),
|
141 |
+
num_return_sequences=1,
|
142 |
+
pad_token_id=tokenizer.eos_token_id
|
143 |
+
)
|
144 |
+
return tokenizer.decode(outputs[0], skip_special_tokens=True)
|
145 |
+
except Exception as e:
|
146 |
+
return f"An error occurred: {str(e)}"
|
147 |
+
|
148 |
+
# Fonction pour configurer les presets
|
149 |
+
def set_preset(preset):
|
150 |
+
if preset == "Default":
|
151 |
+
return 5, 8, 3e-5
|
152 |
+
elif preset == "Fast Training":
|
153 |
+
return 3, 16, 5e-5
|
154 |
+
elif preset == "High Accuracy":
|
155 |
+
return 10, 4, 1e-5
|
156 |
+
|
157 |
+
# Interface Gradio
|
158 |
+
with gr.Blocks() as ui:
|
159 |
+
gr.Markdown("# Model-Fine-Tuner | by Dimonapatrick243")
|
160 |
+
|
161 |
+
with gr.Tab("Train Model"):
|
162 |
+
with gr.Row():
|
163 |
+
data_file = gr.File(label="Upload Data File (CSV, JSON, Excel)")
|
164 |
+
model_name = gr.Textbox(label="Model Name", value="gpt2")
|
165 |
+
output_dir = gr.Textbox(label="Output Directory", value="./fine-tuned-gpt2")
|
166 |
+
|
167 |
+
with gr.Row():
|
168 |
+
preset = gr.Radio(["Default", "Fast Training", "High Accuracy"], label="Preset")
|
169 |
+
epochs = gr.Number(label="Epochs", value=5)
|
170 |
+
batch_size = gr.Number(label="Batch Size", value=8)
|
171 |
+
learning_rate = gr.Number(label="Learning Rate", value=3e-5)
|
172 |
+
|
173 |
+
preset.change(set_preset, preset, [epochs, batch_size, learning_rate])
|
174 |
+
|
175 |
+
# Champs pour sélectionner les colonnes
|
176 |
+
with gr.Row():
|
177 |
+
design_col = gr.Dropdown(label="Design Column")
|
178 |
+
description_col = gr.Dropdown(label="Description Column")
|
179 |
+
|
180 |
+
# Détection des colonnes lors du téléchargement du fichier
|
181 |
+
data_file.upload(read_file, inputs=data_file, outputs=[design_col, description_col])
|
182 |
+
|
183 |
+
train_button = gr.Button("Train Model")
|
184 |
+
train_output = gr.Textbox(label="Training Output")
|
185 |
+
train_graph = gr.Image(label="Training and Validation Loss Graph")
|
186 |
+
|
187 |
+
train_button.click(train_model, inputs=[data_file, model_name, epochs, batch_size, learning_rate, output_dir, design_col, description_col], outputs=[train_output, train_graph])
|
188 |
+
|
189 |
+
with gr.Tab("Generate Text"):
|
190 |
+
with gr.Row():
|
191 |
+
with gr.Column():
|
192 |
+
temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=1.0, value=0.7)
|
193 |
+
top_k = gr.Slider(label="Top K", minimum=1, maximum=100, value=50)
|
194 |
+
max_length = gr.Slider(label="Max Length", minimum=10, maximum=1024, value=128)
|
195 |
+
repetition_penalty = gr.Slider(label="Repetition Penalty", minimum=1.0, maximum=2.0, value=1.2)
|
196 |
+
use_comma = gr.Checkbox(label="Use Comma", value=True)
|
197 |
+
|
198 |
+
with gr.Column():
|
199 |
+
prompt = gr.Textbox(label="Prompt")
|
200 |
+
generate_button = gr.Button("Generate Text")
|
201 |
+
generated_text = gr.Textbox(label="Generated Text")
|
202 |
+
|
203 |
+
generate_button.click(generate_text, inputs=[prompt, temperature, top_k, max_length, repetition_penalty, use_comma], outputs=generated_text)
|
204 |
+
|
205 |
+
ui.launch()
|