apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-04 08:16:42 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Feature] block sparse attention (#3668)

Browse Source 
* 支持稀疏attn

* fix bug

* code style

* fix moba attn get kv shape

* 修复a100编译

* codestyle

* code style

* code style

* code style

* fix conflict

* 增加单侧

* code style

* 增加eblite 加载时间

* fix bug

* for ci

* for ci

* for ci

* for ci

* 支持mlp block size 128

* 增加小算子单测

* fix 单测 mlp

* 将环境变量加入到config里面

* fix rollout config

* 修复显存

* add test server

* add test server

* fix mlp  最后一层使用full attn
This commit is contained in:
yangjianfengo1
2025-08-29 19:46:30 +08:00
committed by GitHub
parent ccd52b5596
commit 3754a9906d
31 changed files with 6553 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
custom_ops/gpu_ops/cpp_extensions.cc
View File

@@ -845,15 +845,15 @@ void SpeculateStepPaddle(
const int max_draft_tokens);
void MergePrefillDecodeOutput(
const paddle::Tensor &encoder_res,
const paddle::Tensor &decoder_res,
const paddle::Tensor &seq_lens_encoder,
const paddle::Tensor &seq_lens_decoder,
const paddle::Tensor &seq_lens_this_time,
const paddle::Tensor &cu_seq_q,
const int head_num,
const int head_dim,
const int max_token);
const paddle::Tensor &encoder_res,
const paddle::Tensor &decoder_res,
const paddle::Tensor &seq_lens_encoder,
const paddle::Tensor &seq_lens_decoder,
const paddle::Tensor &seq_lens_this_time,
const paddle::Tensor &cu_seq_q,
const int head_num,
const int head_dim,
const int max_token);
std::vector<paddle::Tensor> TopPSamplingReject(const paddle::Tensor &probs,
const paddle::Tensor &top_p,