apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[XPU] Refactor get_padding_offset to single kernel. (#7029)

Browse Source 
* [XPU] Refactor get_padding_offset to single kernel.

* add unittest.

* fix codestyle.

* remove cum_offsets_now.

* remove max_len.
This commit is contained in:
Jiajun Ji
2026-04-13 11:04:50 +08:00
committed by GitHub
parent 26d6a20c2f
commit cb03958b52
7 changed files with 199 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/xpu_pre_and_post_process.py
+1 -3
View File
@@ -106,7 +106,6 @@ def xpu_pre_process(
use_cudagraph=False,
) -> XPUForwardMeta:
""" """
max_len = input_ids.shape[1]
token_num_cpu = paddle.sum(seq_lens_this_time).cpu()
if use_speculate_method:
@@ -124,14 +123,13 @@ def xpu_pre_process(
share_inputs["cu_seqlens_q_output"] = cu_seqlens_q_output
share_inputs["batch_id_per_token_output"] = batch_id_per_token_output
else:
cum_offsets_now = paddle.cumsum(max_len - seq_lens_this_time, dtype="int32")
(
ids_remove_padding,
cum_offsets,
batch_id_per_token,
cu_seqlens_q,
cu_seqlens_k,
) = get_padding_offset(input_ids, cum_offsets_now, token_num_cpu, seq_lens_this_time)
) = get_padding_offset(input_ids, seq_lens_this_time, token_num_cpu)
share_inputs["batch_id_per_token"] = batch_id_per_token
share_inputs["cu_seqlens_q"] = cu_seqlens_q