apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 17:11:21 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

[Speculative Decoding][MTP]Support attn mask offset (#4641)
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* [MTP]Merge support attn (#4591)

* support mask_offset in speculate decoding

* fix dummpy run output

* add unit test

* fix unit test import

* support attn_mask_offset in mtp mode

* add update_attn_mask op

* fix unit test && fix code-style
This commit is contained in:
freeliuzc
2025-11-03 10:08:01 +08:00
committed by GitHub
parent f44f4bafd1
commit 11398790d3
13 changed files with 638 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files
custom_ops/gpu_ops/cpp_extensions.cc
+18 -1
View File
@@ -782,7 +782,8 @@ void SpeculateUpdate(const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& stop_flags,
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& is_block_step,
const paddle::Tensor& stop_nums);
const paddle::Tensor& stop_nums,
const paddle::Tensor& mask_rollback);
void SpeculateSetValueByFlagsAndIdx(const paddle::Tensor& pre_ids_all,
const paddle::Tensor& accept_tokens,
@@ -1047,6 +1048,18 @@ void SpeculateGetTargetLogits(const paddle::Tensor& target_logits,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& accept_num);
std::vector<paddle::Tensor> UpdateAttnMaskOffsets(
const paddle::Tensor& ids_remove_padding,
const paddle::Tensor& seq_lens_this_time, // only on cpu
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& attn_mask_offsets_full,
const paddle::Tensor& attn_mask_offsets_decoder,
const paddle::Tensor& is_block_step,
const paddle::Tensor& decode_states,
const paddle::Tensor& mask_rollback);
PYBIND11_MODULE(fastdeploy_ops, m) {
m.def("get_expert_token_num",
&GetExpertTokenNum,
@@ -1632,4 +1645,8 @@ PYBIND11_MODULE(fastdeploy_ops, m) {
m.def("speculate_get_target_logits",
&SpeculateGetTargetLogits,
"speculate_get_target_logits function");
m.def("update_attn_mask_offsets",
&UpdateAttnMaskOffsets,
"update attention mask");
}