app.py

from fastapi import FastAPI
from datasets import load_dataset
from transformers import AutoTokenizer
from transformers import DataCollatorForSeq2Seq
import evaluate
import numpy as np
from transformers import AutoModelForSeq2SeqLM, Seq2SeqTrainingArguments, Seq2SeqTrainer

app = FastAPI()

@app.get("/")
def summarize():
    # Example: Loading a dataset as part of the API
    billsum = load_dataset("billsum", split="ca_test")

#    import pandas as pd

#    df = pd.read_csv("squad_sample_train.tsv", sep="\t")
#    print(df.head())  # Debugging step
#    return {"Hello": "World!", "dataset_length": len(billsum)}
#    return df.head()
    checkpoint = "google-t5/t5-small"
    tokenizer = AutoTokenizer.from_pretrained(checkpoint)

    prefix = "summarize: "


    def preprocess_function(examples):
        inputs = [prefix + doc for doc in examples["text"]]
        model_inputs = tokenizer(inputs, max_length=1024, truncation=True)

        labels = tokenizer(text_target=examples["summary"], max_length=128, truncation=True)

        model_inputs["labels"] = labels["input_ids"]
        return model_inputs

    tokenized_billsum = billsum.map(preprocess_function, batched=True)

    data_collator = DataCollatorForSeq2Seq(tokenizer=tokenizer, model=checkpoint)
   
    rouge = evaluate.load("rouge")

    def compute_metrics(eval_pred):
        predictions, labels = eval_pred
        decoded_preds = tokenizer.batch_decode(predictions, skip_special_tokens=True)
        labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
        decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)

        result = rouge.compute(predictions=decoded_preds, references=decoded_labels, use_stemmer=True)

        prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in predictions]
        result["gen_len"] = np.mean(prediction_lens)

        return {k: round(v, 4) for k, v in result.items()}

    model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
    
    training_args = Seq2SeqTrainingArguments(
        output_dir="my_awesome_billsum_model",
        eval_strategy="no",
        learning_rate=2e-5,
        per_device_train_batch_size=16,  # Increase batch size
        per_device_eval_batch_size=16,
        weight_decay=0.01,
        save_total_limit=3,
        num_train_epochs=1,  # Reduce epochs
        predict_with_generate=True,
        fp16=True,  # Keep mixed precision
        push_to_hub=False,
    #    optim="adamw_bnb_8bit",  # Use 8-bit optimizer
        logging_steps=100,  # Reduce logging overhead
        dataloader_num_workers=4,  # Speed up data loading
        save_strategy="epoch",  # Reduce checkpointing overhead
        gradient_accumulation_steps=4  # Effective larger batch size
    )

    trainer = Seq2SeqTrainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_billsum["train"],
        eval_dataset=tokenized_billsum["test"],
        processing_class=tokenizer,
        data_collator=data_collator,
        compute_metrics=compute_metrics,
    )

    trainer.train()

    return data_collator
#    return type(tokenized_billsum)

"""from fastapi import FastAPI
from datasets import load_dataset
from transformers import AutoTokenizer

app = FastAPI()

#@app.get("/")
# Load dataset and tokenizer
billsum = load_dataset("billsum", split="ca_test")  # Load a small sample
tokenizer = AutoTokenizer.from_pretrained("t5-small")
prefix = "summarize: "  # Example prefix for text generation

@app.get("/")
def preprocess_function(examples):
    inputs = [prefix + doc for doc in examples["text"]]
    model_inputs = tokenizer(inputs, max_length=1024, truncation=True)

    labels = tokenizer(text_target=examples["summary"], max_length=128, truncation=True)
    model_inputs["labels"] = labels["input_ids"]

    return model_inputs

#@app.get("/")
def get_tokenized_data():
    tokenized_billsum = billsum.map(preprocess_function, batched=True)

    # Convert to list of dictionaries
    json_serializable_output = tokenized_billsum.to_pandas().to_dict(orient="records")

    return {"tokenized_data": json_serializable_output}  # Ensure JSON format"""