chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[GDN] Remove unnecessary contiguous() (#13604)

Browse Source
This commit is contained in:
Binyao Jiang
2025-11-19 16:07:49 -08:00
committed by GitHub
parent 17b24aca6e
commit f88b2aa6af
1 changed files with 4 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
python/sglang/srt/layers/attention/hybrid_linear_attn_backend.py
View File

@@ -649,11 +649,9 @@ class GDNAttnBackend(MambaAttnBackendBase):
if is_target_verify:
batch_size = seq_len // forward_batch.spec_info.draft_token_num
draft_token_num = forward_batch.spec_info.draft_token_num
mixed_qkv_reshaped = (
mixed_qkv.view(batch_size, draft_token_num, -1)
.transpose(1, 2)
.contiguous()
)
mixed_qkv_reshaped = mixed_qkv.view(
batch_size, draft_token_num, -1
).transpose(1, 2)
mixed_qkv_processed = causal_conv1d_update(
mixed_qkv_reshaped,
conv_states_to_use,
@@ -666,9 +664,7 @@ class GDNAttnBackend(MambaAttnBackendBase):
retrieve_next_sibling=retrieve_next_sibling,
retrieve_parent_token=retrieve_parent_token,
)
mixed_qkv = (
mixed_qkv_processed.transpose(1, 2).contiguous().view(seq_len, -1)
)
mixed_qkv = mixed_qkv_processed.transpose(1, 2).view(seq_len, -1)
else:
mixed_qkv = causal_conv1d_fn(
mixed_qkv.transpose(0, 1),