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

[Model Runner] Refactor execute_model for GPU async scheduling (#6176)

Browse Source
This commit is contained in:
sunxin
2026-01-28 14:19:33 +08:00
committed by GitHub
parent ce06c6dfb3
commit 27f8799f04
9 changed files with 188 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files
custom_ops/gpu_ops/update_inputs.cu
+2 -7
View File
@@ -68,7 +68,7 @@ __global__ void update_inputs_kernel(bool* not_need_stop,
}
void UpdateInputs(const paddle::Tensor& stop_flags,
const paddle::Tensor& not_need_stop, // only on cpu
const paddle::Tensor& not_need_stop, // on gpu
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
@@ -86,9 +86,8 @@ void UpdateInputs(const paddle::Tensor& stop_flags,
const int max_bsz = stop_flags.shape()[0];
const int now_bsz = seq_lens_this_time.shape()[0];
const int input_ids_stride = input_ids.shape()[1];
auto not_need_stop_gpu = not_need_stop.copy_to(stop_flags.place(), false);
update_inputs_kernel<1024><<<1, 1024, 0, cu_stream>>>(
const_cast<bool*>(not_need_stop_gpu.data<bool>()),
const_cast<bool*>(not_need_stop.data<bool>()),
const_cast<int*>(seq_lens_this_time.data<int>()),
const_cast<int*>(seq_lens_encoder.data<int>()),
const_cast<int*>(seq_lens_decoder.data<int>()),
@@ -99,10 +98,6 @@ void UpdateInputs(const paddle::Tensor& stop_flags,
now_bsz,
max_bsz,
input_ids_stride);
auto not_need_stop_cpu =
not_need_stop_gpu.copy_to(not_need_stop.place(), false);
bool* not_need_stop_data = const_cast<bool*>(not_need_stop.data<bool>());
not_need_stop_data[0] = not_need_stop_cpu.data<bool>()[0];
}
PD_BUILD_STATIC_OP(update_inputs)