train.py

# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved

import os
import time
import sys
import torch
import logging
import json
import numpy as np
import random
import pickle

import torch.distributed as dist
from torch.utils.data import DataLoader, RandomSampler

from src.options import Options
from src import data, beir_utils, slurm, dist_utils, utils
from src import moco, inbatch


logger = logging.getLogger(__name__)


def train(opt, model, optimizer, scheduler, step):

    run_stats = utils.WeightedAvgStats()

    tb_logger = utils.init_tb_logger(opt.output_dir)

    logger.info("Data loading")
    if isinstance(model, torch.nn.parallel.DistributedDataParallel):
        tokenizer = model.module.tokenizer
    else:
        tokenizer = model.tokenizer
    collator = data.Collator(opt=opt)
    train_dataset = data.load_data(opt, tokenizer)
    logger.warning(f"Data loading finished for rank {dist_utils.get_rank()}")

    train_sampler = RandomSampler(train_dataset)
    train_dataloader = DataLoader(
        train_dataset,
        sampler=train_sampler,
        batch_size=opt.per_gpu_batch_size,
        drop_last=True,
        num_workers=opt.num_workers,
        collate_fn=collator,
    )

    epoch = 1

    model.train()
    while step < opt.total_steps:
        train_dataset.generate_offset()

        logger.info(f"Start epoch {epoch}")
        for i, batch in enumerate(train_dataloader):
            step += 1

            batch = {key: value.cuda() if isinstance(value, torch.Tensor) else value for key, value in batch.items()}
            train_loss, iter_stats = model(**batch, stats_prefix="train")

            train_loss.backward()
            optimizer.step()

            scheduler.step()
            model.zero_grad()

            run_stats.update(iter_stats)

            if step % opt.log_freq == 0:
                log = f"{step} / {opt.total_steps}"
                for k, v in sorted(run_stats.average_stats.items()):
                    log += f" | {k}: {v:.3f}"
                    if tb_logger:
                        tb_logger.add_scalar(k, v, step)
                log += f" | lr: {scheduler.get_last_lr()[0]:0.3g}"
                log += f" | Memory: {torch.cuda.max_memory_allocated()//1e9} GiB"

                logger.info(log)
                run_stats.reset()

            if step % opt.eval_freq == 0:
                if isinstance(model, torch.nn.parallel.DistributedDataParallel):
                    encoder = model.module.get_encoder()
                else:
                    encoder = model.get_encoder()
                eval_model(
                    opt, query_encoder=encoder, doc_encoder=encoder, tokenizer=tokenizer, tb_logger=tb_logger, step=step
                )

                if dist_utils.is_main():
                    utils.save(model, optimizer, scheduler, step, opt, opt.output_dir, f"lastlog")

                model.train()

            if dist_utils.is_main() and step % opt.save_freq == 0:
                utils.save(model, optimizer, scheduler, step, opt, opt.output_dir, f"step-{step}")

            if step > opt.total_steps:
                break
        epoch += 1


def eval_model(opt, query_encoder, doc_encoder, tokenizer, tb_logger, step):
    for datasetname in opt.eval_datasets:
        metrics = beir_utils.evaluate_model(
            query_encoder,
            doc_encoder,
            tokenizer,
            dataset=datasetname,
            batch_size=opt.per_gpu_eval_batch_size,
            norm_doc=opt.norm_doc,
            norm_query=opt.norm_query,
            beir_dir=opt.eval_datasets_dir,
            score_function=opt.score_function,
            lower_case=opt.lower_case,
            normalize_text=opt.eval_normalize_text,
        )

        message = []
        if dist_utils.is_main():
            for metric in ["NDCG@10", "Recall@10", "Recall@100"]:
                message.append(f"{datasetname}/{metric}: {metrics[metric]:.2f}")
                if tb_logger is not None:
                    tb_logger.add_scalar(f"{datasetname}/{metric}", metrics[metric], step)
            logger.info(" | ".join(message))


if __name__ == "__main__":
    logger.info("Start")

    options = Options()
    opt = options.parse()

    torch.manual_seed(opt.seed)
    slurm.init_distributed_mode(opt)
    slurm.init_signal_handler()

    directory_exists = os.path.isdir(opt.output_dir)
    if dist.is_initialized():
        dist.barrier()
    os.makedirs(opt.output_dir, exist_ok=True)
    if not directory_exists and dist_utils.is_main():
        options.print_options(opt)
    if dist.is_initialized():
        dist.barrier()
    utils.init_logger(opt)

    os.environ["TOKENIZERS_PARALLELISM"] = "false"

    if opt.contrastive_mode == "moco":
        model_class = moco.MoCo
    elif opt.contrastive_mode == "inbatch":
        model_class = inbatch.InBatch
    else:
        raise ValueError(f"contrastive mode: {opt.contrastive_mode} not recognised")

    if not directory_exists and opt.model_path == "none":
        model = model_class(opt)
        model = model.cuda()
        optimizer, scheduler = utils.set_optim(opt, model)
        step = 0
    elif directory_exists:
        model_path = os.path.join(opt.output_dir, "checkpoint", "latest")
        model, optimizer, scheduler, opt_checkpoint, step = utils.load(
            model_class,
            model_path,
            opt,
            reset_params=False,
        )
        logger.info(f"Model loaded from {opt.output_dir}")
    else:
        model, optimizer, scheduler, opt_checkpoint, step = utils.load(
            model_class,
            opt.model_path,
            opt,
            reset_params=False if opt.continue_training else True,
        )
        if not opt.continue_training:
            step = 0
        logger.info(f"Model loaded from {opt.model_path}")

    logger.info(utils.get_parameters(model))

    if dist.is_initialized():
        model = torch.nn.parallel.DistributedDataParallel(
            model,
            device_ids=[opt.local_rank],
            output_device=opt.local_rank,
            find_unused_parameters=False,
        )
        dist.barrier()

    logger.info("Start training")
    train(opt, model, optimizer, scheduler, step)