add models

models/model_t5-small.py

@@ -0,0 +1,101 @@
+############################################################
+##### Imports
+############################################################
+
+
+import evaluate
+import numpy as np
+
+from datasets import load_dataset
+from transformers import (
+    AutoTokenizer,
+    DataCollatorForSeq2Seq,
+    AutoModelForSeq2SeqLM,
+    Seq2SeqTrainingArguments,
+    Seq2SeqTrainer,
+)
+
+
+############################################################
+##### Fine-Tune T5-small Model
+############################################################
+
+
+def preprocess_function(examples):
+    inputs = [prefix + example[source_lang] for example in examples["translation"]]
+    targets = [example[target_lang] for example in examples["translation"]]
+    model_inputs = tokenizer(inputs, text_target=targets, max_length=128, truncation=True)
+    return model_inputs
+
+
+def postprocess_text(preds, labels):
+    preds = [pred.strip() for pred in preds]
+    labels = [[label.strip()] for label in labels]
+
+    return preds, labels
+
+
+def compute_metrics(eval_preds):
+    preds, labels = eval_preds
+    if isinstance(preds, tuple):
+        preds = preds[0]
+    decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True)
+
+    labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
+    decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
+
+    decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
+
+    result = metric.compute(predictions=decoded_preds, references=decoded_labels)
+    result = {"bleu": result["score"]}
+
+    prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
+    result["gen_len"] = np.mean(prediction_lens)
+    result = {k: round(v, 4) for k, v in result.items()}
+    return result
+
+
+if __name__ == "__main__":
+
+    checkpoint = "google-t5/t5-small"
+
+    books = load_dataset("opus_books", "de-en")
+    books = books["train"].train_test_split(test_size=0.2)
+
+    source_lang = "en"
+    target_lang = "de"
+    prefix = "Translate English to German: "
+
+    tokenizer = AutoTokenizer.from_pretrained(checkpoint)
+    tokenized_books = books.map(preprocess_function, batched=True)
+
+    training_args = Seq2SeqTrainingArguments(
+        output_dir="my_awesome_opus_books_model",
+        evaluation_strategy="epoch",
+        learning_rate=2e-3,
+        per_device_train_batch_size=16,
+        per_device_eval_batch_size=16,
+        weight_decay=0.01,
+        save_total_limit=3,
+        num_train_epochs=10,
+        predict_with_generate=True,
+        fp16=True,
+        push_to_hub=False,
+    )
+
+    metric = evaluate.load("sacrebleu")
+
+    data_collator = DataCollatorForSeq2Seq(tokenizer=tokenizer, model=checkpoint)
+    model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
+
+    trainer = Seq2SeqTrainer(
+        model=model,
+        args=training_args,
+        train_dataset=tokenized_books["train"],
+        eval_dataset=tokenized_books["test"],
+        tokenizer=tokenizer,
+        data_collator=data_collator,
+        compute_metrics=compute_metrics,
+    )
+
+    trainer.train()