Spaces:
Sleeping
Sleeping
File size: 9,381 Bytes
c643b82 24cf4d3 c643b82 a9281cb c643b82 a9281cb 117a333 a9281cb 117a333 a9281cb 117a333 a9281cb c643b82 24cf4d3 117a333 24cf4d3 c643b82 117a333 c643b82 24cf4d3 117a333 24cf4d3 c643b82 24cf4d3 c643b82 a9281cb c643b82 2022648 c643b82 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 |
import os
import json
import gradio as gr
import torch
import pandas as pd
from transformers import (
TrainingArguments,
Trainer,
AutoModelForCausalLM,
AutoTokenizer,
DataCollatorForLanguageModeling
)
from datasets import Dataset
from peft import (
prepare_model_for_kbit_training,
LoraConfig,
get_peft_model
)
# Constants
MODEL_NAME = "deepseek-ai/DeepSeek-R1"
OUTPUT_DIR = "finetuned_models"
LOGS_DIR = "training_logs"
def save_uploaded_file(file_obj):
"""Save uploaded file and return its path"""
try:
os.makedirs('uploads', exist_ok=True)
import tempfile
# Create a temporary file
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.csv', dir='uploads')
# Write the content
if isinstance(file_obj, (bytes, bytearray)):
temp_file.write(file_obj)
else:
content = file_obj.read()
if isinstance(content, str):
temp_file.write(content.encode('utf-8'))
else:
temp_file.write(content)
temp_file.close()
return temp_file.name
except Exception as e:
raise Exception(f"Error saving file: {str(e)}")
def prepare_training_data(df):
"""Convert DataFrame into Q&A format"""
formatted_data = []
try:
for _, row in df.iterrows():
# Clean and validate the data
chunk_id = str(row['chunk_id']).strip()
text = str(row['text']).strip()
if chunk_id and text: # Only include non-empty pairs
# Format each conversation in the required structure
formatted_text = f"User: {chunk_id}\nAssistant: {text}"
formatted_data.append({"text": formatted_text})
if not formatted_data:
raise ValueError("No valid training pairs found in the data")
return formatted_data
except Exception as e:
raise Exception(f"Error preparing training data: {str(e)}")
def prepare_training_components(
data_path,
learning_rate,
num_epochs,
batch_size,
model_name=MODEL_NAME
):
"""Prepare model, tokenizer, and training arguments"""
print(f"Loading data from: {data_path}") # Debug logging
"""Prepare model, tokenizer, and training arguments"""
# Create output directory with timestamp
import time
timestamp = time.strftime("%Y%m%d_%H%M%S")
specific_output_dir = os.path.join(OUTPUT_DIR, f"run_{timestamp}")
os.makedirs(specific_output_dir, exist_ok=True)
os.makedirs(LOGS_DIR, exist_ok=True)
# Load data and convert to Q&A format
try:
df = pd.read_csv(data_path, encoding='utf-8')
print(f"Loaded CSV with {len(df)} rows") # Debug logging
formatted_data = prepare_training_data(df)
print(f"Prepared {len(formatted_data)} training examples") # Debug logging
except Exception as e:
print(f"Error loading CSV: {str(e)}") # Debug logging
raise
# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_name,
trust_remote_code=True,
torch_dtype=torch.float16,
load_in_8bit=True
)
# LoRA Configuration
lora_config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=[
"q_proj", "k_proj", "v_proj", "o_proj",
"gate_proj", "up_proj", "down_proj"
],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM"
)
# Prepare model
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, lora_config)
# Training Arguments
training_args = TrainingArguments(
output_dir=specific_output_dir,
num_train_epochs=num_epochs,
per_device_train_batch_size=batch_size,
learning_rate=learning_rate,
fp16=True,
gradient_accumulation_steps=8,
gradient_checkpointing=True,
logging_dir=os.path.join(LOGS_DIR, f"run_{timestamp}"),
logging_steps=10,
save_strategy="epoch",
evaluation_strategy="epoch",
save_total_limit=2,
)
# Convert to datasets format
dataset = Dataset.from_dict({
'text': [item['text'] for item in formatted_data]
})
# Create data collator
data_collator = DataCollatorForLanguageModeling(
tokenizer=tokenizer,
mlm=False
)
return {
'model': model,
'tokenizer': tokenizer,
'training_args': training_args,
'dataset': dataset,
'data_collator': data_collator,
'output_dir': specific_output_dir
}
def train_model(
file,
learning_rate=2e-4,
num_epochs=3,
batch_size=4,
progress=gr.Progress()
):
"""Training function for Gradio interface"""
if file is None:
return "Please upload a file first."
try:
# File validation
progress(0.1, desc="Validating file...")
file_path = save_uploaded_file(file)
# Prepare components
progress(0.2, desc="Preparing training components...")
components = prepare_training_components(
file_path,
learning_rate,
num_epochs,
batch_size
)
# Initialize trainer
progress(0.4, desc="Initializing trainer...")
trainer = Trainer(
model=components['model'],
args=components['training_args'],
train_dataset=components['dataset'],
data_collator=components['data_collator'],
)
# Train
progress(0.5, desc="Training model...")
trainer.train()
# Save model and tokenizer
progress(0.9, desc="Saving model...")
trainer.save_model()
components['tokenizer'].save_pretrained(components['output_dir'])
progress(1.0, desc="Training complete!")
return f"Training completed! Model saved in {components['output_dir']}"
except Exception as e:
error_msg = f"Error during training: {str(e)}"
print(error_msg) # Log the error
return error_msg
"""Training function for Gradio interface"""
try:
# Save uploaded file
file_path = save_uploaded_file(file)
# Prepare components
progress(0.2, desc="Preparing training components...")
components = prepare_training_components(
file_path,
learning_rate,
num_epochs,
batch_size
)
# Initialize trainer
progress(0.4, desc="Initializing trainer...")
trainer = Trainer(
model=components['model'],
args=components['training_args'],
train_dataset=components['dataset'],
data_collator=components['data_collator'],
)
# Train
progress(0.5, desc="Training model...")
trainer.train()
# Save model and tokenizer
progress(0.9, desc="Saving model...")
trainer.save_model()
components['tokenizer'].save_pretrained(components['output_dir'])
progress(1.0, desc="Training complete!")
return f"Training completed! Model saved in {components['output_dir']}"
except Exception as e:
return f"Error during training: {str(e)}"
# Create Gradio interface
def create_interface():
demo = gr.Interface(
# Configure Gradio to handle larger file uploads
upload_size_limit=100
)
with gr.Row():
with gr.Column():
file_input = gr.File(
label="Upload Training Data (CSV)",
type="binary",
file_types=[".csv"]
)
learning_rate = gr.Slider(
minimum=1e-5,
maximum=1e-3,
value=2e-4,
label="Learning Rate"
)
num_epochs = gr.Slider(
minimum=1,
maximum=10,
value=3,
step=1,
label="Number of Epochs"
)
batch_size = gr.Slider(
minimum=1,
maximum=8,
value=4,
step=1,
label="Batch Size"
)
train_button = gr.Button("Start Training")
with gr.Column():
output = gr.Textbox(label="Training Status")
train_button.click(
fn=train_model,
inputs=[file_input, learning_rate, num_epochs, batch_size],
outputs=output
)
gr.Markdown("""
## Instructions
1. Upload your training data in CSV format with columns:
- chunk_id (questions)
- text (answers)
2. Adjust training parameters if needed
3. Click 'Start Training'
4. Wait for training to complete
""")
return demo
if __name__ == "__main__":
# Create necessary directories
os.makedirs(OUTPUT_DIR, exist_ok=True)
os.makedirs(LOGS_DIR, exist_ok=True)
# Launch Gradio interface
demo = create_interface()
demo.launch(share=True) |