def test_ignored_modules_transformer(self):
"""Tests that ignored modules' parameters are not flattened for a
transformer model with shared parameters."""
# Initialize an FSDP-wrapped transformer model that has FSDP ignore
# the `nn.Transformer` module's parameters
model: nn.Module = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.NO_FSDP,
CUDAInitMode.CUDA_BEFORE,
deterministic=True,
)
wrapped_model = FSDP(
model,
self.process_group,
ignored_modules=[model.transformer],
)
# Check that the wrapped model's flattened parameter does not include
# the ignored transformer module's parameters
nonwrapped_model: nn.Module = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.NO_FSDP,
CUDAInitMode.CUDA_BEFORE,
deterministic=True,
)
total_numel = sum(p.numel() for p in nonwrapped_model.parameters())
ignored_numel = sum(p.numel()
for p in nonwrapped_model.transformer.parameters())
nonignored_numel = total_numel - ignored_numel
with FSDP.summon_full_params(wrapped_model):
flat_param_numel = wrapped_model.params[0].numel()
self.assertEqual(flat_param_numel, nonignored_numel)
# Check that we can run a few iterations
optim = torch.optim.Adam(wrapped_model.parameters(), lr=1e-3)
self._train_model(wrapped_model, optim, 3)
def test_param_change_after_init(self, mixed_precision):
"""
Tests that changing FSDP model parameter values in-place after FSDP
initialization persist.
"""
# Establish reference behavior
fsdp_kwargs = {}
if mixed_precision:
fsdp_kwargs["mixed_precision"] = MixedPrecision()
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
fsdp_kwargs,
deterministic=True,
)
input = fsdp_model.module.get_input(torch.device("cuda"))
ref_output = fsdp_model(*input)
# Initialize the same model but change its first parameter value
# in-place after FSDP initialization
new_fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
fsdp_kwargs,
deterministic=True,
)
first_param = next(new_fsdp_model.parameters())
nn.init.normal_(first_param.data)
new_output = new_fsdp_model(*input)
self.assertNotEqual(
ref_output,
new_output,
msg="new_output did not reflect change to param after init",
)
def _init_model(
self,
nested_model: bool,
sharding_strategy: ShardingStrategy,
device: torch.device,
):
fsdp_kwargs = {"sharding_strategy": sharding_strategy}
if nested_model:
model = NestedWrappedModule.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
fsdp_kwargs,
)
fsdp_model: FSDP = FSDP(
model,
self.process_group,
**fsdp_kwargs,
).to(device)
else:
fsdp_model: FSDP = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_BEFORE,
fsdp_kwargs,
)
return fsdp_model
def _init_transformer_model(
self,
wrap: bool,
device: torch.device = torch.device("cuda"),
group=None,
optim_class: Type[torch.optim.Optimizer] = torch.optim.Adam,
use_multiple_param_groups: bool = False,
use_diff_optim_inputs: bool = False,
):
if use_multiple_param_groups or use_diff_optim_inputs:
# Keep these as arguments for parity with `_init_nested_model()`;
# these settings are not implemented since the transformer is
# wrapped with FSDP at the top-level, which means that there is
# only a single flattened parameter, making these booleans vacuous
raise NotImplementedError()
if group is None:
group = dist.distributed_c10d._get_default_group()
model = TransformerWithSharedParams.init(
group,
FSDPInitMode.RECURSIVE if wrap else FSDPInitMode.NO_FSDP,
CUDAInitMode.CUDA_BEFORE,
deterministic=True,
)
optim = optim_class(model.parameters(), lr=0.01)
return model, optim, None
def test_transformer_no_grad(self, mixed_precision):
"""Tests that for an FSDP-wrapped transformer model with shared
parameters, after training for one iteration, running a forward pass in
``eval()`` mode gives the same output as running a forward pass in
``torch.no_grad()``."""
fsdp_kwargs = {}
if mixed_precision:
fsdp_kwargs["mixed_precision"] = MixedPrecision(
param_dtype=torch.float16,
reduce_dtype=torch.float16,
buffer_dtype=torch.float16,
)
else:
fsdp_kwargs["mixed_precision"] = None
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
fsdp_kwargs,
)
self._train_for_several_steps(
fsdp_model,
num_steps=1,
autocast=False,
mixed_precision=fsdp_kwargs["mixed_precision"]
)
input = fsdp_model.module.get_input(torch.device("cuda"))
# Run a forward in eval mode
fsdp_model.eval()
ref_output = fsdp_model(*input)
# Run a forward in `no_grad()` and compare
with torch.no_grad():
no_grad_output = fsdp_model(*input)
self.assertEqual(ref_output, no_grad_output)
def _test_mixed_precision_embedding_table(self, mp_config):
# Basic test to ensure int inputs are not casted which would break
# modules such as embedding tables.
param_dtype = mp_config.param_dtype or torch.float32
orig_reduce_scatter = dist._reduce_scatter_base
test_reduce_scatter = partial(
self._reduce_scatter_base_validate_mp,
orig_reduce_scatter,
mp_config,
)
with patch_reduce_scatter(test_reduce_scatter, param_dtype):
# TODO: `test_mp_embedding_reduce()` fails if we do not wrap the
# entire `TransformerWithSharedParams` with a single top-level FSDP
model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.NO_FSDP,
CUDAInitMode.CUDA_BEFORE,
{"mixed_precision": mp_config},
)
fsdp_model = FSDP(model, mixed_precision=mp_config)
optim = torch.optim.SGD(fsdp_model.parameters(), lr=0.1)
for _ in range(6):
inp = fsdp_model.module.get_input(torch.device("cuda"))
# This would fail if we casted integer module inputs such as for
# embedding tables.
output = fsdp_model(*inp)
loss = fsdp_model.module.get_loss(inp, output).cuda()
self.assertEqual(loss.dtype, param_dtype)
fsdp_model.module.run_backward(loss)
optim.step()
def test_pre_backward_hook_registration(self, cuda_first: bool):
"""Tests that FSDP pre-backward hooks are registered on forward pass
outputs."""
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_BEFORE if cuda_first else CUDAInitMode.CUDA_AFTER,
)
self._test_pre_backward_hook_registration(fsdp_model)
def test_transformer_module_apply(self):
"""Tests that ``apply()`` modifies parameter values in-place on an
FSDP-wrapped transformer model with shared parameters."""
transformer = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
)
self._check_apply(transformer)
def test_apply_in_summon_raises_error(self):
"""Tests that calling ``apply()`` on an FSDP instance inside the
``summon_full_params()`` context raises an error."""
transformer = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
)
with transformer.summon_full_params(transformer):
with self.assertRaisesRegex(ValueError, "expected to be in states"):
transformer.apply(self._init_linear_weights)
def test_pre_backward_hook_registration_after_state_dict(self):
"""Tests that FSDP pre-backward hooks are registered on forward pass
outputs after saving and loading the model from a checkpoint."""
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
)
self._train_for_several_steps(fsdp_model, num_steps=2, autocast=False)
state_dict = fsdp_model.state_dict()
fsdp_model.load_state_dict(state_dict)
self._test_pre_backward_hook_registration(fsdp_model)
def test_register_functions_called(self, cuda_first: bool, mixed_precision: bool):
"""Tests that ``_register_{pre|post}_backward_hooks()`` are called
during the FSDP forward."""
fsdp_kwargs = {}
if mixed_precision:
fsdp_kwargs["mixed_precision"] = MixedPrecision()
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_BEFORE if cuda_first else CUDAInitMode.CUDA_AFTER,
fsdp_kwargs,
)
input = fsdp_model.module.get_input(torch.device("cuda"))
fsdp_model._register_pre_backward_hooks = mock.MagicMock(return_value=None)
fsdp_model._register_post_backward_hooks = mock.MagicMock(return_value=None)
self.assertFalse(fsdp_model._register_post_backward_hooks.called)
self.assertFalse(fsdp_model._register_pre_backward_hooks.called)
fsdp_model(*input)
self.assertTrue(fsdp_model._register_post_backward_hooks.called)
self.assertTrue(fsdp_model._register_pre_backward_hooks.called)
def test_device_id_auto_wrap(self):
"""Tests that ``auto_wrap_policy`` propagates ``device_id`` to all
nested FSDP instances."""
auto_wrap_policy = functools.partial(
transformer_auto_wrap_policy,
transformer_layer_cls={TransformerEncoderLayer, TransformerDecoderLayer},
)
fsdp_kwargs = {
"auto_wrap_policy": auto_wrap_policy,
"device_id": torch.cuda.current_device(),
}
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_BEFORE,
fsdp_kwargs,
)
for fsdp_module in FSDP.fsdp_modules(fsdp_model):
self.assertEqual(
fsdp_module.device_id,
torch.device("cuda", torch.cuda.current_device()),
)
def test_transformer_auto_wrap_policy(self):
"""Tests the ``transformer_auto_wrap_policy``."""
auto_wrap_policy = functools.partial(
transformer_auto_wrap_policy,
transformer_layer_cls={
TransformerEncoderLayer, TransformerDecoderLayer
},
)
fsdp_kwargs = {"auto_wrap_policy": auto_wrap_policy}
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_BEFORE,
fsdp_kwargs,
)
modules = list(fsdp_model.modules())
encoder_layers = set(fsdp_model.module.transformer.encoder.layers)
decoder_layers = set(fsdp_model.module.transformer.decoder.layers)
for module in modules:
if module is fsdp_model or module in encoder_layers or module in decoder_layers:
self.assertTrue(isinstance(module, FSDP))
else:
self.assertFalse(isinstance(module, FSDP))
def test_state_dict_rank0_offload_save_load_flow(self):
"""Tests saving a model checkpoint only on rank 0 and loading it only
on rank 0 with ``sync_module_states=True`` to emulate the workflow to
avoid redundant CPU memory usage."""
auto_wrap_policy = partial(
transformer_auto_wrap_policy,
transformer_layer_cls={
TransformerEncoderLayer, TransformerDecoderLayer
},
)
fsdp_kwargs = {"auto_wrap_policy": auto_wrap_policy}
fsdp_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_BEFORE,
fsdp_kwargs,
)
# Force model parameters and buffers to be nonzero
with FSDP.summon_full_params(fsdp_model):
for tensor in itertools.chain(fsdp_model.parameters(),
fsdp_model.buffers()):
if torch.count_nonzero(tensor) == 0:
with torch.no_grad():
tensor.add_(
torch.tensor(1,
dtype=tensor.dtype,
device=tensor.device))
with self._get_state_dict_mgr(fsdp_model, "state_dict", True):
state_dict = deepcopy(_get_state_dict(fsdp_model))
# Initialize a non-wrapped model on all ranks
new_model = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.NO_FSDP,
CUDAInitMode.CUDA_BEFORE,
)
_zero_model(new_model, zero_buffers=True)
# Only load the checkpoint on rank 0
if self.rank == 0:
new_model.load_state_dict(state_dict, strict=True)
_assert_module_states(
new_model,
process_group=self.process_group,
assert_fn=self.assertNotEqual,
)
# Broadcast the module states from rank 0 with `sync_module_states=True`
new_fsdp_model = FSDP(
new_model,
device_id=torch.cuda.current_device(),
auto_wrap_policy=auto_wrap_policy,
sync_module_states=True,
)
# Check FSDP models are equal across ranks
with FSDP.summon_full_params(new_fsdp_model):
_assert_module_states(
new_fsdp_model,
process_group=self.process_group,
assert_fn=self.assertEqual,
)
# Check FSDP models correctly loaded the checkpoint
with FullyShardedDataParallel.summon_full_params(fsdp_model):
with FullyShardedDataParallel.summon_full_params(new_fsdp_model):
params = list(fsdp_model.parameters())
params_new = list(new_fsdp_model.parameters())
self.assertEqual(params, params_new)
def _test_grad_acc(
self,
batch_dim: int,
configs: List[_GradAccConfig],
cpu_offload: CPUOffload,
backward_prefetch: Optional[BackwardPrefetch],
):
"""
Tests gradient accumulation by comparing a run that trains sequentially
through some batches while accumulating gradients with a run that
trains on the concatenation of those batches in a single iteration.
The last iteration always synchronizes gradients regardless of what is
specified by the last element of ``configs``.
Arguments:
batch_dim (int): Batch dimension in the input tensor to be passed
into the model for the forward pass.
configs (List[_GradAccConfig]): :class:`list` of configurations
specifying how gradients are accumulated; for example, a list
corresponding to [(False, 2), (True, 2), (False, 2)] indicates
to accumulate over 2 + 2 + 2 = 6 total iterations, where the
first two do not use ``no_sync()``, the middle two do use
``no_sync()``, and the final two again do not use
``no_sync()``.
cpu_offload (CPUOffload): Configures CPU offloading.
backward_prefetch (Optional[BackwardPrefetch]): Specifies at which
point to prefetch the next layer's full parameters during the
backward pass, if at all.
"""
# Gradient accumulation outside `no_sync()` is not currently compatible
# with CPU offloading
if cpu_offload.offload_params and \
any(not config.use_no_sync for config in configs):
return
old_allow_tf32 = torch.backends.cuda.matmul.allow_tf32
try:
# Disable TF32 to prevent floating point drift
torch.backends.cuda.matmul.allow_tf32 = False
# Initialize the FSDP model and optimizer
fsdp_kwargs = {
"cpu_offload": cpu_offload,
"backward_prefetch": backward_prefetch,
}
fsdp_model: FSDP = TransformerWithSharedParams.init(
self.process_group,
FSDPInitMode.RECURSIVE,
CUDAInitMode.CUDA_AFTER,
fsdp_kwargs,
deterministic=True,
add_bn=
False, # disable BN since the test uses varying batch sizes
)
device = torch.device("cuda")
optim = torch.optim.SGD(
fsdp_model.parameters(),
lr=0.01,
momentum=0.9,
)
# Generate the sequence of batches, each containing the same data
# but permuted
def permute_tensor(x: torch.Tensor):
return x.view(-1)[torch.randperm(x.numel())].view_as(x)
batch: Tuple[torch.Tensor, ...] = \
fsdp_model.module.get_input(device)
batches: List[Tuple[torch.Tensor, ...]] = [batch]
num_iters_to_acc = sum(config.num_iters for config in configs)
for _ in range(num_iters_to_acc - 1):
batches.append(tuple(permute_tensor(t) for t in batch))
for (batch1, batch2) in itertools.combinations(batches, r=2):
for t1, t2 in zip(batch1, batch2):
assert not torch.all(t1 == t2), \
"Check the test to make sure that batches are distinct"
# Concatenate the batches along the given batch dimension
concat_batch: Tuple[torch.Tensor, ...] = tuple(
torch.cat(ts, dim=batch_dim) for ts in zip(*batches))
# Establish reference gradients using the concatenated batch
fsdp_model.zero_grad()
output = fsdp_model(*concat_batch)
ref_loss = fsdp_model.module.get_loss(concat_batch, output)
ref_loss.backward()
ref_grads = [
p.grad.detach().clone() for p in fsdp_model.parameters()
]
# Compute and accumulate the gradients
fsdp_model.zero_grad()
losses = []
batch_idx = 0
for config in configs:
sync_context = fsdp_model.no_sync() if config.use_no_sync \
else contextlib.suppress()
with sync_context:
for _ in range(config.num_iters):
if batch_idx == num_iters_to_acc - 1:
break # always sync on the last iteration
batch = batches[batch_idx]
batch_idx += 1
output = fsdp_model(*batch)
loss = fsdp_model.module.get_loss(batch, output)
loss.backward()
losses.append(loss)
output = fsdp_model(*batches[-1])
loss = fsdp_model.module.get_loss(batches[-1], output)
loss.backward()
losses.append(loss)
acc_loss = sum(losses)
acc_grads = [
p.grad.detach().clone() for p in fsdp_model.parameters()
]
# Compare the losses and gradients
torch.testing.assert_close(ref_loss, acc_loss)
self.assertEqual(len(ref_grads), len(acc_grads))
for ref_grad, acc_grad in zip(ref_grads, acc_grads):
self.assertEqual(ref_grad.device, acc_grad.device)
self.assertEqual(ref_grad.size(), acc_grad.size())
self.assertEqual(ref_grad.dtype, acc_grad.dtype)
torch.testing.assert_close(ref_grad, acc_grad)
# Check that the optimizer step does not error
optim.step()
finally:
torch.backends.cuda.matmul.allow_tf32 = old_allow_tf32