Skip to content

Remove padding for CPU TBE op to reduce the memory waste (#14) #61

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

Closed
wants to merge 1 commit into from
Closed

Remove padding for CPU TBE op to reduce the memory waste (#14) #61

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
7 changes: 5 additions & 2 deletions torchrec/distributed/quant_embedding_kernel.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -143,14 +143,17 @@ def _to_data_type(dtype: torch.dtype) -> DataType:
itertools.chain(module.named_buffers(), module.named_parameters())
)
device = next(iter(state_dict.values())).device
use_cpu = not (device is not None and device.type == "cuda")

# Adjust config to quantized version.
# This obviously doesn't work for column-wise sharding.
# pyre-ignore [29]
config = copy.deepcopy(module.config())
config.data_type = data_type
for table in config.embedding_tables:
table.local_cols = rounded_row_size_in_bytes(table.local_cols, sparse_type)
table.local_cols = rounded_row_size_in_bytes(
table.local_cols, sparse_type, use_cpu
)
if table.local_metadata is not None:
table.local_metadata.shard_sizes = [
table.local_rows,
Expand All @@ -163,7 +166,7 @@ def _to_data_type(dtype: torch.dtype) -> DataType:
shard_meta.shard_sizes = [
shard_meta.shard_sizes[0],
rounded_row_size_in_bytes(
shard_meta.shard_sizes[1], sparse_type
shard_meta.shard_sizes[1], sparse_type, use_cpu
),
]
table.global_metadata.size = torch.Size(
Expand Down