EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

fix page first per layer pf2lf kernel (#8915)

Browse Source 
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
This commit is contained in:
huangtingwei
2025-08-10 08:16:11 +08:00
committed by GitHub
parent 5c31b35db2
commit 86497d99f2
5 changed files with 15 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
sgl-kernel/python/sgl_kernel/kvcacheio.py
View File

@@ -34,6 +34,7 @@ def transfer_kv_per_layer_pf_lf(
dst_v: torch.Tensor,
src_indices: torch.Tensor,
dst_indices: torch.Tensor,
layer_id: int,
item_size: int,
src_layout_dim: int,
block_quota: int = 2,
@@ -46,6 +47,7 @@ def transfer_kv_per_layer_pf_lf(
dst_v,
src_indices,
dst_indices,
layer_id,
item_size,
src_layout_dim,
block_quota,
@@ -144,6 +146,7 @@ def transfer_kv_per_layer_mla_pf_lf(
dst: torch.Tensor,
src_indices: torch.Tensor,
dst_indices: torch.Tensor,
layer_id: int,
item_size: int,
src_layout_dim: int,
block_quota: int = 2,
@@ -154,6 +157,7 @@ def transfer_kv_per_layer_mla_pf_lf(
dst,
src_indices,
dst_indices,
layer_id,
item_size,
src_layout_dim,
block_quota,