File size: 1,571 Bytes
c40dcd1 |
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 |
# agent_1_train.py
from datasets import load_dataset
from transformers import GPT2Config, GPT2LMHeadModel, GPT2Tokenizer
from transformers import DataCollatorForLanguageModeling, Trainer, TrainingArguments
# Load dataset
dataset = load_dataset("wikitext", "wikitext-2-raw-v1")
# Tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
tokenizer.pad_token = tokenizer.eos_token
# Tiny GPT config (~20M params)
config = GPT2Config(
vocab_size=tokenizer.vocab_size,
n_positions=128,
n_ctx=128,
n_embd=256,
n_layer=4,
n_head=4
)
model = GPT2LMHeadModel(config)
# Tokenize dataset
def tokenize_function(examples):
return tokenizer(examples['text'], truncation=True, max_length=128, padding="max_length")
tokenized_datasets = dataset.map(tokenize_function, batched=True)
tokenized_datasets.set_format(type='torch', columns=['input_ids'])
# Data collator
data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)
# Training arguments
training_args = TrainingArguments(
output_dir="./tiny-gpt",
num_train_epochs=3,
per_device_train_batch_size=2,
save_steps=500,
save_total_limit=2,
logging_steps=50,
learning_rate=5e-4,
fp16=False
)
# Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=tokenized_datasets['train'],
tokenizer=tokenizer,
data_collator=data_collator
)
# Train model
trainer.train()
# Save model
model.save_pretrained("./tiny-gpt")
tokenizer.save_pretrained("./tiny-gpt")
print("Training complete! Model saved in ./tiny-gpt")
|