|
|
from tqdm import tqdm |
|
|
import torch |
|
|
from trainer.build import TRAINER_REGISTRY, BaseTrainer |
|
|
|
|
|
@TRAINER_REGISTRY.register() |
|
|
class OpenVocabTrainer(BaseTrainer): |
|
|
def __init__(self, cfg): |
|
|
super().__init__(cfg) |
|
|
self.best_metric = -1 |
|
|
|
|
|
def forward(self, data_dict, mode): |
|
|
return self.model(data_dict, mode) |
|
|
|
|
|
def backward(self, loss, mode=None): |
|
|
self.accelerator.backward(loss) |
|
|
|
|
|
if self.grad_norm is not None and self.accelerator.sync_gradients: |
|
|
self.accelerator.clip_grad_norm_(self.model.parameters(), self.grad_norm) |
|
|
|
|
|
self.optimizer.step() |
|
|
self.scheduler.step() |
|
|
self.optimizer.zero_grad() |
|
|
|
|
|
def train_step(self, epoch, mode = None): |
|
|
self.model.train() |
|
|
loader = self.data_loaders[self.mode] |
|
|
is_main = self.accelerator.is_main_process |
|
|
|
|
|
pbar = tqdm(loader, disable=not is_main, desc=f"[Epoch {epoch + 1}/{self.epochs}]") |
|
|
|
|
|
for data_dict in pbar: |
|
|
with self.accelerator.accumulate(self.model): |
|
|
|
|
|
data_dict = self.forward(data_dict, mode=mode) |
|
|
|
|
|
|
|
|
loss, losses = self.loss(data_dict) |
|
|
|
|
|
|
|
|
self.backward(loss, mode=mode) |
|
|
|
|
|
|
|
|
self.global_step += 1 |
|
|
log_dict = {'step': self.global_step, **losses} |
|
|
|
|
|
if mode == 'qa': |
|
|
metrics = self.evaluator["train"].batch_metrics(data_dict) |
|
|
log_dict.update(metrics) |
|
|
self.log(log_dict, mode="train") |
|
|
|
|
|
@torch.no_grad() |
|
|
def eval_step(self, epoch, mode): |
|
|
self.model.eval() |
|
|
loader = self.data_loaders["val"] |
|
|
pbar = tqdm(range(len(loader)), disable=(not self.accelerator.is_main_process)) |
|
|
for i, data_dict in enumerate(loader): |
|
|
data_dict = self.forward(data_dict, mode = mode) |
|
|
loss, losses = self.loss(data_dict) |
|
|
log_dict = {'epoch': epoch, **losses} |
|
|
pbar.update(1) |
|
|
self.log(log_dict, mode="val") |
|
|
|
|
|
@torch.no_grad() |
|
|
def test_step(self): |
|
|
self.model.eval() |
|
|
loader = self.data_loaders["test"] |
|
|
pbar = tqdm(range(len(loader)), disable=(not self.accelerator.is_main_process)) |
|
|
for i, data_dict in enumerate(loader): |
|
|
data_dict = self.forward(data_dict) |
|
|
self.evaluator["val"].update(data_dict) |
|
|
pbar.update(1) |
|
|
is_best, results = self.evaluator["val"].record() |
|
|
self.log(results, mode="test") |
|
|
self.evaluator["val"].reset() |
|
|
return results |
|
|
|
|
|
def run(self): |
|
|
num_trainable_params = 0 |
|
|
for name, param in self.model.named_parameters(): |
|
|
if param.requires_grad: |
|
|
num_trainable_params += param.numel() |
|
|
print(name) |
|
|
|
|
|
print(f"Total number of trainable parameters: {num_trainable_params:,}") |
|
|
if self.mode in ['warmup', "pretrain"]: |
|
|
start_epoch = self.exp_tracker.epoch |
|
|
self.global_step = start_epoch * len(self.data_loaders[self.mode]) |
|
|
|
|
|
for epoch in range(start_epoch, self.epochs): |
|
|
self.exp_tracker.step() |
|
|
self.train_step(epoch, mode=self.mode) |
|
|
|
|
|
self.accelerator.wait_for_everyone() |
|
|
if self.accelerator.is_main_process: |
|
|
if self.epochs_per_save and (epoch + 1) % self.epochs_per_save == 0: |
|
|
self.save(f"ckpt_{epoch+1}.pth") |
|
|
|
|
|
self.save(f"ckpt_{epoch+1}.pth") |
|
|
self.accelerator.end_training() |
|
|
else: |
|
|
if self.mode == "train": |
|
|
start_epoch = self.exp_tracker.epoch |
|
|
self.global_step = start_epoch * len(self.data_loaders["train"]) |
|
|
for epoch in range(start_epoch, self.epochs): |
|
|
self.exp_tracker.step() |
|
|
self.train_step(epoch) |
|
|
if self.epochs_per_eval and (epoch + 1) % self.epochs_per_eval == 0: |
|
|
is_best = self.eval_step(epoch) |
|
|
self.accelerator.print(f"[Epoch {epoch + 1}/{self.epochs}] finished eval, is_best: {is_best}") |
|
|
else: |
|
|
is_best = False |
|
|
|
|
|
self.accelerator.wait_for_everyone() |
|
|
if self.accelerator.is_main_process: |
|
|
if is_best: |
|
|
self.save("best.pth") |
|
|
if self.epochs_per_save and (epoch + 1) % self.epochs_per_save == 0: |
|
|
self.save(f"ckpt_{epoch+1}.pth") |
|
|
|
|
|
self.test_step() |
|
|
if self.mode == "train": |
|
|
self.accelerator.end_training() |
