ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-15 18:42:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Tune the input initialization to avoid over-flow in silu

Browse Source
This commit is contained in:
Qianfeng Zhang
2025-04-09 10:03:32 +00:00
parent 86c0e45987
commit dd2cd2cbcb
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
example/ck_tile/18_hstu_attention/example_hstu_attention.cpp
View File

@@ -290,9 +290,9 @@ bool run(const ck_tile::ArgParser& arg_parser)
ck_tile::HostTensor<InOutDataType> o_host_ref(
std::array<ck_tile::index_t, 4>{batches_for_alloc, seqlen, num_head, hdim_v});
ck_tile::FillNormalDistributionIntegerValue<InOutDataType>{-2.f, 2.f, seed}(q_host);
ck_tile::FillNormalDistributionIntegerValue<InOutDataType>{-2.f, 2.f, seed}(k_host);
ck_tile::FillNormalDistributionIntegerValue<InOutDataType>{-2.f, 2.f, seed}(v_host);
ck_tile::FillNormalDistribution<InOutDataType>{0.f, 1.f, seed}(q_host);
ck_tile::FillNormalDistribution<InOutDataType>{0.f, 1.f, seed}(k_host);
ck_tile::FillNormalDistributionIntegerValue<InOutDataType>{-3.f, 3.f, seed}(v_host);
ck_tile::DeviceMem q_dev(q_host.get_element_space_size_in_bytes());
ck_tile::DeviceMem k_dev(k_host.get_element_space_size_in_bytes());