Spaces:
Paused
Paused
Update train.py
Browse files
train.py
CHANGED
|
@@ -1,3 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
import sys
|
| 3 |
from transformers import (
|
|
@@ -6,55 +11,104 @@ from transformers import (
|
|
| 6 |
Trainer,
|
| 7 |
TrainingArguments,
|
| 8 |
)
|
| 9 |
-
from datasets import load_dataset, Dataset
|
| 10 |
from huggingface_hub import login
|
| 11 |
import torch
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
def create_local_dataset():
|
| 15 |
-
"""Create a small local dataset for testing
|
| 16 |
print("\nπ Creating local dataset for testing...")
|
| 17 |
|
| 18 |
-
#
|
| 19 |
-
train_texts = [
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
"I loved this movie, it was amazing!",
|
| 21 |
-
"The acting was terrible and boring.",
|
| 22 |
"Great film, highly recommend it.",
|
| 23 |
-
"Waste of time, don't watch it.",
|
| 24 |
"Fantastic story and characters!",
|
| 25 |
-
"Poorly made and disappointing.",
|
| 26 |
"Excellent cinematography!",
|
| 27 |
-
"The plot made no sense at all.",
|
| 28 |
"Brilliant performance by the lead actor.",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
"I fell asleep halfway through.",
|
| 30 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
-
|
|
|
|
|
|
|
| 33 |
|
| 34 |
-
|
| 35 |
"Pretty good movie overall.",
|
| 36 |
-
"Not worth the ticket price.",
|
| 37 |
"Enjoyable with some flaws.",
|
| 38 |
-
"Complete disaster.",
|
| 39 |
"Decent watch for a rainy day.",
|
| 40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
|
| 42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
train_dataset = Dataset.from_dict({
|
| 45 |
-
'text': train_texts,
|
| 46 |
-
'label': train_labels
|
| 47 |
})
|
| 48 |
test_dataset = Dataset.from_dict({
|
| 49 |
-
'text': test_texts,
|
| 50 |
-
'label': test_labels
|
| 51 |
})
|
| 52 |
|
| 53 |
print(f"β
Created local dataset: {len(train_dataset)} train, {len(test_dataset)} test")
|
| 54 |
return train_dataset, test_dataset
|
| 55 |
|
| 56 |
|
| 57 |
-
# --- Main Training Function (FIXED) ---
|
| 58 |
def setup_training():
|
| 59 |
"""Setup and run the fine-tuning process"""
|
| 60 |
|
|
@@ -62,26 +116,55 @@ def setup_training():
|
|
| 62 |
print("π TRACKMODEL - FINE-TUNING STARTED")
|
| 63 |
print("=" * 60)
|
| 64 |
|
| 65 |
-
|
| 66 |
-
print(f"β
Transformers version: {__import__('transformers').__version__}")
|
| 67 |
-
print(f"β
Datasets version: {__import__('datasets').__version__}")
|
| 68 |
-
|
| 69 |
-
# 1. Load Dataset - FIXED with trust_remote_code=True
|
| 70 |
print("\nπ Step 1: Loading dataset...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
try:
|
| 72 |
-
#
|
| 73 |
-
|
| 74 |
-
print(f"β
Loaded {len(dataset['train'])} training samples")
|
| 75 |
-
print(f"β
Loaded {len(dataset['test'])} test samples")
|
| 76 |
|
| 77 |
-
#
|
| 78 |
-
|
| 79 |
train_subset = dataset["train"].select(range(500))
|
| 80 |
test_subset = dataset["test"].select(range(100))
|
|
|
|
| 81 |
|
| 82 |
-
except Exception as
|
| 83 |
-
print(f"
|
| 84 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
train_subset, test_subset = create_local_dataset()
|
| 86 |
|
| 87 |
# 2. Load Model and Tokenizer
|
|
@@ -98,7 +181,6 @@ def setup_training():
|
|
| 98 |
print("β
Model and tokenizer loaded successfully")
|
| 99 |
except Exception as e:
|
| 100 |
print(f"β Failed to load model: {str(e)}")
|
| 101 |
-
print("π‘ Try: pip install transformers==4.30.0")
|
| 102 |
return None
|
| 103 |
|
| 104 |
# 3. Tokenize the dataset
|
|
@@ -163,7 +245,6 @@ def setup_training():
|
|
| 163 |
print("β
Training complete!")
|
| 164 |
except Exception as e:
|
| 165 |
print(f"β Training failed: {str(e)}")
|
| 166 |
-
print("π‘ If out of memory, reduce batch size or dataset size")
|
| 167 |
return None
|
| 168 |
|
| 169 |
# 6. Save the model
|
|
@@ -193,7 +274,6 @@ def setup_training():
|
|
| 193 |
return "./my_finetuned_model"
|
| 194 |
|
| 195 |
|
| 196 |
-
# --- Push to Hugging Face Hub (Unchanged) ---
|
| 197 |
def push_to_hub():
|
| 198 |
"""Push the fine-tuned model to your Hugging Face profile"""
|
| 199 |
print("\nπ€ Pushing model to Hugging Face Hub...")
|
|
@@ -206,13 +286,10 @@ def push_to_hub():
|
|
| 206 |
print("π Please login manually:")
|
| 207 |
login()
|
| 208 |
|
| 209 |
-
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
| 210 |
-
|
| 211 |
model = AutoModelForSequenceClassification.from_pretrained("./my_finetuned_model")
|
| 212 |
tokenizer = AutoTokenizer.from_pretrained("./my_finetuned_model")
|
| 213 |
|
| 214 |
-
|
| 215 |
-
repo_name = "Surajnitp/sentiment-trackmodel"
|
| 216 |
print(f"π€ Uploading to: https://huggingface.co/{repo_name}")
|
| 217 |
model.push_to_hub(repo_name)
|
| 218 |
tokenizer.push_to_hub(repo_name)
|
|
@@ -221,7 +298,6 @@ def push_to_hub():
|
|
| 221 |
print(f"π https://huggingface.co/{repo_name}")
|
| 222 |
|
| 223 |
|
| 224 |
-
# --- Main Execution ---
|
| 225 |
if __name__ == "__main__":
|
| 226 |
# Train the model
|
| 227 |
model_path = setup_training()
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Trackmodel - Fine-Tuning Script
|
| 3 |
+
This script downloads a pre-trained model and fine-tunes it on IMDb dataset
|
| 4 |
+
"""
|
| 5 |
+
|
| 6 |
import os
|
| 7 |
import sys
|
| 8 |
from transformers import (
|
|
|
|
| 11 |
Trainer,
|
| 12 |
TrainingArguments,
|
| 13 |
)
|
| 14 |
+
from datasets import load_dataset, Dataset
|
| 15 |
from huggingface_hub import login
|
| 16 |
import torch
|
| 17 |
|
| 18 |
+
print(f"β
PyTorch version: {torch.__version__}")
|
| 19 |
+
print(f"β
Transformers version: {__import__('transformers').__version__}")
|
| 20 |
+
print(f"β
Datasets version: {__import__('datasets').__version__}")
|
| 21 |
+
|
| 22 |
+
|
| 23 |
def create_local_dataset():
|
| 24 |
+
"""Create a small local dataset for testing when IMDb fails"""
|
| 25 |
print("\nπ Creating local dataset for testing...")
|
| 26 |
|
| 27 |
+
# Training data (500 samples)
|
| 28 |
+
train_texts = []
|
| 29 |
+
train_labels = []
|
| 30 |
+
|
| 31 |
+
# Positive examples (250)
|
| 32 |
+
positive_texts = [
|
| 33 |
"I loved this movie, it was amazing!",
|
|
|
|
| 34 |
"Great film, highly recommend it.",
|
|
|
|
| 35 |
"Fantastic story and characters!",
|
|
|
|
| 36 |
"Excellent cinematography!",
|
|
|
|
| 37 |
"Brilliant performance by the lead actor.",
|
| 38 |
+
"This movie exceeded all my expectations.",
|
| 39 |
+
"Absolutely wonderful film!",
|
| 40 |
+
"The best movie I've seen this year.",
|
| 41 |
+
"Incredible acting and beautiful visuals.",
|
| 42 |
+
"A masterpiece of modern cinema.",
|
| 43 |
+
]
|
| 44 |
+
for i in range(25):
|
| 45 |
+
for text in positive_texts:
|
| 46 |
+
if len(train_texts) < 250:
|
| 47 |
+
train_texts.append(text)
|
| 48 |
+
train_labels.append(1)
|
| 49 |
+
|
| 50 |
+
# Negative examples (250)
|
| 51 |
+
negative_texts = [
|
| 52 |
+
"The acting was terrible and boring.",
|
| 53 |
+
"Waste of time, don't watch it.",
|
| 54 |
+
"Poorly made and disappointing.",
|
| 55 |
+
"The plot made no sense at all.",
|
| 56 |
"I fell asleep halfway through.",
|
| 57 |
+
"This movie was a complete disaster.",
|
| 58 |
+
"Absolutely terrible film.",
|
| 59 |
+
"Worst movie I've ever seen.",
|
| 60 |
+
"Boring and completely predictable.",
|
| 61 |
+
"A complete waste of money.",
|
| 62 |
+
]
|
| 63 |
+
for i in range(25):
|
| 64 |
+
for text in negative_texts:
|
| 65 |
+
if len(train_texts) < 500:
|
| 66 |
+
train_texts.append(text)
|
| 67 |
+
train_labels.append(0)
|
| 68 |
|
| 69 |
+
# Test data (100 samples)
|
| 70 |
+
test_texts = []
|
| 71 |
+
test_labels = []
|
| 72 |
|
| 73 |
+
test_positive = [
|
| 74 |
"Pretty good movie overall.",
|
|
|
|
| 75 |
"Enjoyable with some flaws.",
|
|
|
|
| 76 |
"Decent watch for a rainy day.",
|
| 77 |
+
"Interesting plot with good acting.",
|
| 78 |
+
"A solid film worth watching.",
|
| 79 |
+
]
|
| 80 |
+
for i in range(10):
|
| 81 |
+
for text in test_positive:
|
| 82 |
+
if len(test_texts) < 50:
|
| 83 |
+
test_texts.append(text)
|
| 84 |
+
test_labels.append(1)
|
| 85 |
|
| 86 |
+
test_negative = [
|
| 87 |
+
"Not worth the ticket price.",
|
| 88 |
+
"Complete disaster of a movie.",
|
| 89 |
+
"Terrible acting and boring.",
|
| 90 |
+
"Predictable and unoriginal.",
|
| 91 |
+
"This movie was a mistake.",
|
| 92 |
+
]
|
| 93 |
+
for i in range(10):
|
| 94 |
+
for text in test_negative:
|
| 95 |
+
if len(test_texts) < 100:
|
| 96 |
+
test_texts.append(text)
|
| 97 |
+
test_labels.append(0)
|
| 98 |
|
| 99 |
train_dataset = Dataset.from_dict({
|
| 100 |
+
'text': train_texts[:500],
|
| 101 |
+
'label': train_labels[:500]
|
| 102 |
})
|
| 103 |
test_dataset = Dataset.from_dict({
|
| 104 |
+
'text': test_texts[:100],
|
| 105 |
+
'label': test_labels[:100]
|
| 106 |
})
|
| 107 |
|
| 108 |
print(f"β
Created local dataset: {len(train_dataset)} train, {len(test_dataset)} test")
|
| 109 |
return train_dataset, test_dataset
|
| 110 |
|
| 111 |
|
|
|
|
| 112 |
def setup_training():
|
| 113 |
"""Setup and run the fine-tuning process"""
|
| 114 |
|
|
|
|
| 116 |
print("π TRACKMODEL - FINE-TUNING STARTED")
|
| 117 |
print("=" * 60)
|
| 118 |
|
| 119 |
+
# 1. Load Dataset
|
|
|
|
|
|
|
|
|
|
|
|
|
| 120 |
print("\nπ Step 1: Loading dataset...")
|
| 121 |
+
|
| 122 |
+
train_subset = None
|
| 123 |
+
test_subset = None
|
| 124 |
+
|
| 125 |
+
# Try loading from Hugging Face first
|
| 126 |
try:
|
| 127 |
+
# Try different loading methods
|
| 128 |
+
print("π Attempting to load IMDb dataset...")
|
|
|
|
|
|
|
| 129 |
|
| 130 |
+
# Method 1: Standard load
|
| 131 |
+
dataset = load_dataset("imdb", trust_remote_code=True)
|
| 132 |
train_subset = dataset["train"].select(range(500))
|
| 133 |
test_subset = dataset["test"].select(range(100))
|
| 134 |
+
print(f"β
Loaded IMDb dataset: {len(train_subset)} train, {len(test_subset)} test")
|
| 135 |
|
| 136 |
+
except Exception as e1:
|
| 137 |
+
print(f"β οΈ Method 1 failed: {str(e1)[:100]}...")
|
| 138 |
+
|
| 139 |
+
try:
|
| 140 |
+
# Method 2: Try without trust_remote_code
|
| 141 |
+
print("π Attempting method 2...")
|
| 142 |
+
dataset = load_dataset("imdb")
|
| 143 |
+
train_subset = dataset["train"].select(range(500))
|
| 144 |
+
test_subset = dataset["test"].select(range(100))
|
| 145 |
+
print(f"β
Loaded IMDb dataset: {len(train_subset)} train, {len(test_subset)} test")
|
| 146 |
+
|
| 147 |
+
except Exception as e2:
|
| 148 |
+
print(f"β οΈ Method 2 failed: {str(e2)[:100]}...")
|
| 149 |
+
|
| 150 |
+
try:
|
| 151 |
+
# Method 3: Try with specific config
|
| 152 |
+
print("π Attempting method 3...")
|
| 153 |
+
dataset = load_dataset("imdb", split="train+test", trust_remote_code=True)
|
| 154 |
+
# Split back into train/test
|
| 155 |
+
train_size = int(len(dataset) * 0.8)
|
| 156 |
+
train_subset = dataset.select(range(500))
|
| 157 |
+
test_subset = dataset.select(range(500, 600))
|
| 158 |
+
print(f"β
Loaded IMDb dataset: {len(train_subset)} train, {len(test_subset)} test")
|
| 159 |
+
|
| 160 |
+
except Exception as e3:
|
| 161 |
+
print(f"β All IMDb loading methods failed!")
|
| 162 |
+
print("π Using local dataset instead...")
|
| 163 |
+
train_subset, test_subset = create_local_dataset()
|
| 164 |
+
|
| 165 |
+
# If still None, use local dataset
|
| 166 |
+
if train_subset is None or test_subset is None:
|
| 167 |
+
print("π Using local dataset...")
|
| 168 |
train_subset, test_subset = create_local_dataset()
|
| 169 |
|
| 170 |
# 2. Load Model and Tokenizer
|
|
|
|
| 181 |
print("β
Model and tokenizer loaded successfully")
|
| 182 |
except Exception as e:
|
| 183 |
print(f"β Failed to load model: {str(e)}")
|
|
|
|
| 184 |
return None
|
| 185 |
|
| 186 |
# 3. Tokenize the dataset
|
|
|
|
| 245 |
print("β
Training complete!")
|
| 246 |
except Exception as e:
|
| 247 |
print(f"β Training failed: {str(e)}")
|
|
|
|
| 248 |
return None
|
| 249 |
|
| 250 |
# 6. Save the model
|
|
|
|
| 274 |
return "./my_finetuned_model"
|
| 275 |
|
| 276 |
|
|
|
|
| 277 |
def push_to_hub():
|
| 278 |
"""Push the fine-tuned model to your Hugging Face profile"""
|
| 279 |
print("\nπ€ Pushing model to Hugging Face Hub...")
|
|
|
|
| 286 |
print("π Please login manually:")
|
| 287 |
login()
|
| 288 |
|
|
|
|
|
|
|
| 289 |
model = AutoModelForSequenceClassification.from_pretrained("./my_finetuned_model")
|
| 290 |
tokenizer = AutoTokenizer.from_pretrained("./my_finetuned_model")
|
| 291 |
|
| 292 |
+
repo_name = "Surajnitp/sentiment-trackmodel"
|
|
|
|
| 293 |
print(f"π€ Uploading to: https://huggingface.co/{repo_name}")
|
| 294 |
model.push_to_hub(repo_name)
|
| 295 |
tokenizer.push_to_hub(repo_name)
|
|
|
|
| 298 |
print(f"π https://huggingface.co/{repo_name}")
|
| 299 |
|
| 300 |
|
|
|
|
| 301 |
if __name__ == "__main__":
|
| 302 |
# Train the model
|
| 303 |
model_path = setup_training()
|