Skip to content

[checkpoint]: unpad padded parameters when saving sharded MoE checkpoints #6449

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
LiRunGuo wants to merge 1 commit into
base: main
Choose a base branch
from
+8 −4
Open

[checkpoint]: unpad padded parameters when saving sharded MoE checkpoints #6449

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions colossalai/checkpoint_io/moe_checkpoint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
)
from colossalai.interface import ModelWrapper, OptimizerWrapper
from colossalai.tensor.moe_tensor.api import is_moe_tensor
from colossalai.tensor.padded_tensor import is_padded_tensor, to_unpadded_tensor

try:
from torch.nn.modules.module import _EXTRA_STATE_KEY_SUFFIX
Expand Down Expand Up @@ -87,6 +88,8 @@ def _model_sharder(
continue
# Gather tensor pieces when using tensor parallel.
param_ = gather_distributed_param(param, keep_vars=False)
if is_padded_tensor(param_):
param_ = to_unpadded_tensor(param_)
block, block_size = state_dict_sharder.append_param(prefix + name, param_)
if block is not None:
yield block, block_size
Expand Down
9 changes: 5 additions & 4 deletions tests/test_moe/test_moe_checkpoint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,12 +82,15 @@ def check_optimizer_snapshot_equal(snapshot1, snapshot2, param2name, moe_dp_grou
num_attention_heads=2,
num_key_value_heads=2,
num_hidden_layers=2,
# not a multiple of make_vocab_size_divisible_by * tp_size, so the embedding is padded with tp
vocab_size=1000,
),
MixtralForCausalLM,
],
],
)
def check_moe_checkpoint(test_config):
@parameterize("plugin_config", [{"pp_size": 2, "ep_size": 2, "tp_size": 1}, {"pp_size": 2, "ep_size": 1, "tp_size": 2}])
def check_moe_checkpoint(test_config, plugin_config):
dtype, precision = torch.float16, "fp16"
config, model_cls = test_config
torch.cuda.set_device(dist.get_rank())
Expand All @@ -106,9 +109,7 @@ def check_moe_checkpoint(test_config):
seed_all(10086)
model = deepcopy(orig_model)
optimizer = SGD(model.parameters(), lr=1e-3)
plugin = MoeHybridParallelPlugin(
pp_size=2, ep_size=2, tp_size=1, microbatch_size=1, zero_stage=1, precision=precision
)
plugin = MoeHybridParallelPlugin(**plugin_config, microbatch_size=1, zero_stage=1, precision=precision)
booster = Booster(plugin=plugin)
model, optimizer, *_ = booster.boost(model=model, optimizer=optimizer)
# initialize grads
Expand Down