Skip to content

Commit

Permalink
Update API
Browse files Browse the repository at this point in the history
Signed-off-by: Dashiell Stander <[email protected]>
  • Loading branch information
dashstander committed Oct 3, 2023
1 parent 3f43f07 commit 6189c0d
Show file tree
Hide file tree
Showing 2 changed files with 1 addition and 3 deletions.
1 change: 0 additions & 1 deletion megatron/model/transformer.py
Original file line number Diff line number Diff line change
Expand Up @@ -344,7 +344,6 @@ def __init__(
neox_args,
self.attention_type,
self.num_attention_heads_per_partition,
mpu=mpu,
)
else:
if self.use_flash_attention:
Expand Down
3 changes: 1 addition & 2 deletions megatron/model/utils.py
Original file line number Diff line number Diff line change
Expand Up @@ -238,7 +238,7 @@ def _set_use_cache(modules, value: bool):
recursive_setattr(modules, "use_cache", value, assert_type=bool)


def configure_sparse_attention(neox_args, attention_type, num_attention_heads, mpu):
def configure_sparse_attention(neox_args, attention_type, num_attention_heads):
from deepspeed.ops.sparse_attention import (
SparseSelfAttention,
VariableSparsityConfig,
Expand Down Expand Up @@ -337,7 +337,6 @@ def configure_sparse_attention(neox_args, attention_type, num_attention_heads, m
sparsity_config=sparsity_config,
max_seq_length=neox_args.seq_length,
attn_mask_mode="add",
mpu=mpu,
)


Expand Down

0 comments on commit 6189c0d

Please sign in to comment.