[checkpoint] checkpoint for ColoTensor Model (#1196)

colossalai/utils/checkpoint/__init__.py

@@ -0,0 +1,3 @@
+from .module_checkpoint import save_checkpoint, load_checkpoint
+
+__all__ = ['save_checkpoint', 'load_checkpoint']

colossalai/utils/checkpoint/module_checkpoint.py

@@ -0,0 +1,73 @@
+import torch
+import torch.nn as nn
+import torch.distributed as dist
+import collections
+from torch.optim.lr_scheduler import CosineAnnealingLR as _CosineAnnealingLR
+from colossalai.utils.model.colo_init_context import colo_state_dict
+
+def save_checkpoint(dire,
+                    epoch: int,
+                    model: torch.nn.Module,
+                    optimizer: torch.optim.Optimizer = None,
+                    lr_scheduler: torch.optim.lr_scheduler._LRScheduler = None,
+                    *args,
+                    **kwargs):
+    """save_checkpoint 
+    save a model, whose parameters are `ColoTensor`s.
+    Args:
+        dire (_type_): _description_
+        epoch (int): _description_
+        model (torch.nn.Module): _description_
+        optimizer (torch.optim.Optimizer, optional): _description_. Defaults to None.
+        lr_scheduler (torch.optim.lr_scheduler._LRScheduler, optional): _description_. Defaults to None.
+    """
+    model_state = {
+        'epoch': epoch,
+        'model': colo_state_dict(model, state_dict_func=nn.Module.state_dict)
+    }
+    if dist.get_rank() == 0:
+        torch.save(model_state, dire + '/epoch_{}_model.pth'.format(epoch))
+    lr_scheduler_dict = lr_scheduler.state_dict()
+    lr_scheduler_dict['after_scheduler'] = lr_scheduler_dict['after_scheduler'].state_dict()
+    optim_state = {
+        'epoch': epoch,
+        'optimizer': optimizer.state_dict(),
+        'lr_scheduler': lr_scheduler_dict
+    }
+    torch.save(optim_state, dire + '/epoch_{}_optim_rank_{}.pth'.format(epoch, dist.get_rank()))
+
+
+
+
+def load_checkpoint(dire,
+                    epoch: int,
+                    rank: int,
+                    model: torch.nn.Module,
+                    optimizer: torch.optim.Optimizer = None,
+                    lr_scheduler: torch.optim.lr_scheduler._LRScheduler = None,
+                    *args,
+                    **kwargs):
+    """load_checkpoint 
+    load a model, whose parameters are `ColoTensor`s.
+    Args:
+        dire (_type_): _description_
+        epoch (int): _description_
+        rank (int): _description_
+        model (torch.nn.Module): _description_
+        optimizer (torch.optim.Optimizer, optional): _description_. Defaults to None.
+        lr_scheduler (torch.optim.lr_scheduler._LRScheduler, optional): _description_. Defaults to None.
+    """
+    model_state = torch.load(dire + '/epoch_{}_model.pth'.format(epoch))
+    model_state['model'] = collections.OrderedDict([(k.split('.', 1)[1], v) for k, v in model_state['model'].items()])
+    model.load_state_dict(model_state['model'])
+    optim_state = torch.load(dire + '/epoch_{}_optim_rank_{}.pth'.format(epoch, rank))
+    optimizer.load_state_dict(optim_state['optimizer'])
+    lr_scheduler_dict = optim_state['lr_scheduler']
+    after_scheduler_dict = lr_scheduler_dict['after_scheduler']
+    lr_scheduler_dict['after_scheduler'] = _CosineAnnealingLR(
+        optimizer, 
+        after_scheduler_dict['T_max'],
+        after_scheduler_dict['eta_min'],
+        after_scheduler_dict['last_epoch']
+        )
+    lr_scheduler.load_state_dict(lr_scheduler_dict)

colossalai/utils/model/colo_init_context.py

@@ -38,15 +38,18 @@ def colo_state_dict(self, destination=None, prefix='', keep_vars=False, state_di
     # build param to spec mapping
     mapping1 = dict()
     mapping2 = dict()
+    mapping3 = dict()
     # gather all params
     has_dist_parameter = False
     with torch.no_grad():
         for param in self.parameters():
-            if isinstance(param, ColoParameter) and param.has_compute_spec():
+            if isinstance(param, ColoParameter):
                 has_dist_parameter = True
                 mapping1[id(param)] = copy(param.dist_spec)
                 mapping2[id(param)] = copy(param.compute_spec)
+                mapping3[id(param)] = param.get_process_group()
                 param.set_dist_spec(distspec.replicate())
+                param.process_group = None
 
     # TODO: fix when keep_vars = True
     # when keep_vars = False, the state_dict_func will call detach to create
@@ -64,6 +67,7 @@ def colo_state_dict(self, destination=None, prefix='', keep_vars=False, state_di
             if param_id in mapping1:
                 dist_spec = mapping1[id(param)]
                 compute_spec = mapping2[id(param)]
+                param.process_group = mapping3[id(param)]
                 param.set_tensor_spec(dist_spec, compute_spec)
     return ret
 

tests/test_utils/test_colo_checkpoint.py

@@ -0,0 +1,211 @@
+from abc import ABC, abstractmethod
+import os, sys, shutil
+import torch
+import torch.nn as nn
+import pytest
+import copy
+import operator
+import colossalai
+from colossalai.context.parallel_mode import ParallelMode
+import torch.multiprocessing as mp
+import torch.distributed as dist
+from colossalai.testing import rerun_if_address_is_in_use
+from colossalai.utils.cuda import get_current_device
+from colossalai.utils import free_port
+from colossalai.utils.model.colo_init_context import ColoInitContext
+from colossalai.tensor import ColoTensorSpec, ComputePattern, ComputeSpec, DistSpecManager, distspec, ProcessGroup, ColoTensor
+from colossalai.core import global_context as gpc
+from functools import partial
+from colossalai.nn.parallel.data_parallel import ColoDDP
+from colossalai.utils.checkpoint import save_checkpoint, load_checkpoint
+from colossalai.nn.lr_scheduler import CosineAnnealingWarmupLR
+
+
+class DummyDataGenerator(ABC):
+
+    def __init__(self, length=10):
+        self.length = length
+
+    @abstractmethod
+    def generate(self):
+        pass
+
+    def __iter__(self):
+        self.step = 0
+        return self
+
+    def __next__(self):
+        if self.step < self.length:
+            self.step += 1
+            return self.generate()
+        else:
+            raise StopIteration
+
+    def __len__(self):
+        return self.length
+
+
+class DummyDataLoader(DummyDataGenerator):
+    batch_size = 128
+    category = 16
+    feature_size = 256
+
+    def generate(self):
+        image_dict = {}
+        image_dict['pixel_values'] = torch.rand(
+            DummyDataLoader.batch_size, DummyDataLoader.feature_size, device=get_current_device()) * 2 - 1
+        image_dict['label'] = torch.randint(DummyDataLoader.category, (DummyDataLoader.batch_size,),
+                                            dtype=torch.int64,
+                                            device=get_current_device())
+        return image_dict
+
+
+class MLP(nn.Module):
+
+    def __init__(self, in_features, out_features, hidden_features=None):
+        super().__init__()
+        if hidden_features is None:
+            hidden_features = out_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.activation = nn.ReLU()
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.activation(x)
+        x = self.fc2(x)
+        return x
+
+
+def init_1d_row_for_linear_weight_spec(model, pg: ProcessGroup):
+    spec = (distspec.shard([-1], [pg.tp_world_size()]), ComputeSpec(ComputePattern.TP1D))
+    with DistSpecManager.no_grad():
+        for n, p in model.named_parameters():
+            if 'weight' in n:
+                p.set_process_group(pg)
+                p.set_tensor_spec(*spec)
+
+
+def check_param_equal(model, torch_model):
+    for p, torch_p in zip(model.parameters(), torch_model.parameters()):
+        assert torch.allclose(torch_p, p, rtol=1e-3, atol=1e-1)
+
+
+def remove(path):
+    """ param <path> could either be relative or absolute. """
+    if os.path.isfile(path) or os.path.islink(path):
+        os.remove(path)
+    elif os.path.isdir(path):
+        shutil.rmtree(path)
+    else:
+        raise ValueError("file {} is not a file or dir.".format(path))
+
+
+def run_checkpoint(init_spec_func, use_ddp, test_epoch, pg):
+    train_dataloader = DummyDataLoader(length=16)
+    with ColoInitContext(device=get_current_device()):
+        model = MLP(256, 16, 64)
+        model_reload = MLP(256, 16, 64)
+        model_ref = MLP(256, 16, 64)
+    model = model.cuda()
+    model_reload = model_reload.cuda()
+    model_ref = model_ref.cuda()
+    if use_ddp:
+        model = ColoDDP(model, pg)
+        model_reload = ColoDDP(model_reload, pg)
+        model_ref = ColoDDP(model_ref, pg)
+
+    criterion = torch.nn.CrossEntropyLoss()
+    optimizer = torch.optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.999), eps=1e-08, weight_decay=0)
+    optimizer_reload = torch.optim.Adam(model_reload.parameters(),
+                                        lr=0.001,
+                                        betas=(0.9, 0.999),
+                                        eps=1e-08,
+                                        weight_decay=0)
+    optimizer_ref = torch.optim.Adam(model_ref.parameters(), lr=0.001, betas=(0.9, 0.999), eps=1e-08, weight_decay=0)
+
+    lr_scheduler = CosineAnnealingWarmupLR(optimizer=optimizer, total_steps=20, warmup_steps=5)
+    lr_scheduler_reload = CosineAnnealingWarmupLR(optimizer=optimizer_reload, total_steps=20, warmup_steps=5)
+    lr_scheduler_ref = CosineAnnealingWarmupLR(optimizer=optimizer_ref, total_steps=20, warmup_steps=5)
+
+    init_spec_func(model, pg)
+    init_spec_func(model_ref, pg)
+
+    for epoch in range(0, 20):
+        if epoch <= test_epoch:
+            for i, image_dict in enumerate(train_dataloader):
+                if use_ddp:
+                    model.zero_grad()
+                else:
+                    optimizer.zero_grad()
+                logits = model(image_dict['pixel_values'])
+                loss = criterion(logits, image_dict['label'])
+                if use_ddp:
+                    model.backward(loss)
+                else:
+                    loss.backward()
+                optimizer.step()
+
+            if epoch == test_epoch:
+                for ref_p, p in zip(model_ref.parameters(), model.parameters()):
+                    ref_p.data.copy_(p)
+                optimizer_ref = copy.deepcopy(optimizer)
+                lr_scheduler_ref = copy.deepcopy(lr_scheduler)
+
+                check_param_equal(model, model_ref)
+                save_checkpoint('./checkpoint', epoch, model, optimizer, lr_scheduler)
+                dist.barrier()
+        else:
+            if epoch == test_epoch + 1:
+                load_checkpoint('./checkpoint', test_epoch, dist.get_rank(), model_reload, optimizer_reload,
+                                lr_scheduler_reload)
+                init_spec_func(model_reload, pg)
+            for i, image_dict in enumerate(train_dataloader):
+                if use_ddp:
+                    model_ref.zero_grad()
+                    model_reload.zero_grad()
+                else:
+                    optimizer_ref.zero_grad()
+                    optimizer_reload.zero_grad()
+                logits_ref = model_ref(image_dict['pixel_values'])
+                logits_reload = model_reload(image_dict['pixel_values'])
+                loss_ref = criterion(logits_ref, image_dict['label'])
+                loss_reload = criterion(logits_reload, image_dict['label'])
+                if use_ddp:
+                    model_ref.backward(loss_ref)
+                    model_reload.backward(loss_reload)
+                else:
+                    loss_ref.backward()
+                    loss_reload.backward()
+                optimizer_ref.step()
+                optimizer_reload.step()
+        lr_scheduler.step()
+
+    check_param_equal(model_ref, model_reload)
+
+
+def run_dist(rank, world_size, port, use_ddp, test_epoch):
+    if use_ddp and world_size == 1:
+        return
+    tp_world_size = world_size // 2 if use_ddp else world_size
+    config = dict(parallel=dict(tensor=dict(mode="1d", size=tp_world_size),))
+    colossalai.launch(config=config, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
+    pg = ProcessGroup(tp_degree=world_size)
+    run_checkpoint(init_1d_row_for_linear_weight_spec, use_ddp, test_epoch, pg)
+
+
+@pytest.mark.dist
+@pytest.mark.parametrize('world_size', [4])
+@pytest.mark.parametrize('use_ddp', [True])
+@pytest.mark.parametrize('test_epoch', [1, 2, 3])
+@rerun_if_address_is_in_use()
+def test_checkpoint(world_size, use_ddp, test_epoch):
+    if not os.path.isdir('./checkpoint'):
+        os.mkdir('./checkpoint')
+    run_func = partial(run_dist, world_size=world_size, port=free_port(), use_ddp=use_ddp, test_epoch=test_epoch)
+    mp.spawn(run_func, nprocs=world_size)
+    remove('./checkpoint')
+
+
+if __name__ == '__main__':
+    test_checkpoint(4, True, 1)