diff options
Diffstat (limited to 'modules/sub_quadratic_attention.py')
| -rw-r--r-- | modules/sub_quadratic_attention.py | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/modules/sub_quadratic_attention.py b/modules/sub_quadratic_attention.py index 497568eb..4cb561ef 100644 --- a/modules/sub_quadratic_attention.py +++ b/modules/sub_quadratic_attention.py @@ -15,7 +15,7 @@ import torch from torch import Tensor from torch.utils.checkpoint import checkpoint import math -from typing import Optional, NamedTuple, List +from typing import Optional, NamedTuple def narrow_trunc( @@ -58,7 +58,7 @@ def _summarize_chunk( scale: float, ) -> AttnChunk: attn_weights = torch.baddbmm( - torch.empty(1, 1, 1, device=query.device, dtype=query.dtype), + torch.zeros(1, 1, 1, device=query.device, dtype=query.dtype), query, key.transpose(1,2), alpha=scale, @@ -97,7 +97,7 @@ def _query_chunk_attention( ) return summarize_chunk(query, key_chunk, value_chunk) - chunks: List[AttnChunk] = [ + chunks: list[AttnChunk] = [ chunk_scanner(chunk) for chunk in torch.arange(0, k_tokens, kv_chunk_size) ] acc_chunk = AttnChunk(*map(torch.stack, zip(*chunks))) @@ -121,7 +121,7 @@ def _get_attention_scores_no_kv_chunking( scale: float, ) -> Tensor: attn_scores = torch.baddbmm( - torch.empty(1, 1, 1, device=query.device, dtype=query.dtype), + torch.zeros(1, 1, 1, device=query.device, dtype=query.dtype), query, key.transpose(1,2), alpha=scale, |