File size: 2,692 Bytes
3cef3e3 a570747 185d502 a570747 3cef3e3 d81adb9 3cef3e3 816facc 3cef3e3 f05adaf 3cef3e3 f05adaf 3cef3e3 f05adaf 3cef3e3 816facc 3cef3e3 816facc 3cef3e3 |
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 |
import torch
from transformers import T5ForConditionalGeneration, T5Tokenizer, Trainer, TrainingArguments
from datasets import load_dataset
import os
# Set Hugging Face cache environment variables
os.environ["HF_HOME"] = "/app/hf_cache"
os.environ["HF_DATASETS_CACHE"] = "/app/hf_cache"
os.environ["TRANSFORMERS_CACHE"] = "/app/hf_cache"
# Load dataset (Replace this with your dataset)
dataset = load_dataset("tatsu-lab/alpaca") # Example alternative dataset
# Check available dataset splits
print("Dataset splits available:", dataset)
# If "test" split is missing, use a portion of "train" split
if "test" not in dataset:
dataset = dataset["train"].train_test_split(test_size=0.1) # Split 10% for testing
# Tokenize dataset
tokenized_datasets = dataset.map(tokenize_function, batched=True)
# Assign datasets
train_dataset = tokenized_datasets["train"]
eval_dataset = tokenized_datasets["test"] # This is now safely created
# Debug output
print("Dataset successfully split into train and test sets")
# Load model and tokenizer
model_name = "t5-large"
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)
# Tokenization function
def tokenize_function(examples):
print("Sample data structure:", examples) # Move print inside function
inputs = examples["input"] # Make sure "input" matches dataset keys
targets = examples["output"]
model_inputs = tokenizer(inputs, max_length=512, truncation=True, padding="max_length")
labels = tokenizer(targets, max_length=512, truncation=True, padding="max_length")
model_inputs["labels"] = labels["input_ids"]
return model_inputs
return model_inputs
# Apply tokenization
tokenized_datasets = dataset.map(tokenize_function, batched=True)
# Training arguments
training_args = TrainingArguments(
output_dir="./t5-finetuned",
per_device_train_batch_size=2, # Smaller batch to avoid memory errors
per_device_eval_batch_size=2, # Smaller eval batch
save_total_limit=1, # Keep only 1 checkpoint
num_train_epochs=1, # Quick test with 1 epoch
logging_steps=50, # More frequent logging
evaluation_strategy="epoch", # Evaluate only at the end of the epoch
save_strategy="epoch", # Save only at the end of the epoch
push_to_hub=False # Avoid pushing test model to Hugging Face Hub
)
# Trainer setup
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset, # Now correctly assigned
eval_dataset=eval_dataset, # No more KeyError
)
# Train the model
trainer.train()
# Save and push model to Hugging Face Hub
trainer.push_to_hub("your-hf-username/t5-cover-letter")
|