apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[XPU] refactor of block_attn param 'pos_emb_type' (#5511)

Browse Source
This commit is contained in:
Lucas
2025-12-12 14:30:09 +08:00
committed by GitHub
parent 4eb55332f6
commit 888c4b992d
6 changed files with 25 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
fastdeploy/model_executor/forward_meta.py
View File

@@ -253,8 +253,6 @@ class XPUForwardMeta(ForwardMeta):
dec_batch: Optional[paddle.Tensor] = None
#
total_enc_len: Optional[paddle.Tensor] = None
# position embedding type in rope, supports 'NORMAL' or 'HALF_HEAD_DIM'
pos_emb_type: Optional[str] = "NORMAL"
# for pd_disaggregation
kv_signal_sender: Optional[paddle.Tensor] = None

2
fastdeploy/model_executor/layers/attention/xpu_attn_backend.py
View File

@@ -213,7 +213,7 @@ class XPUAttentionBackend(AttentionBackend):
None, # smooth
metadata.kv_signal_data_list[layer.layer_id], # kv_signal_data
forward_meta.kv_signal_sender, # kv_signal_sender
forward_meta.pos_emb_type,
layer.use_neox_rotary_style,
self.rope_3d,
)