minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[CPU] Add GPT-OSS model optimization for CPU (#16775)

Browse Source 
Co-authored-by: mingfeima <mingfei.ma@intel.com>
Co-authored-by: jianan-gu <jianan.gu@intel.com>
This commit is contained in:
blzheng
2026-05-29 16:05:26 +08:00
committed by GitHub
co-authored by mingfeima jianan-gu
parent 5601b7139d
commit 3ecf2c76ad
35 changed files with 2000 additions and 530 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sgl-kernel/csrc/cpu/flash_attn.cpp
+8 -5
View File
@@ -161,9 +161,10 @@ void flash_attn_kernel_impl(
}
}
flash_attn_softmax<scalar_t, BLOCK_M, BLOCK_N>::apply(
s_i, s_delta, v_prime, s_prime, m_prime, m_size, n_size, padded_n_size, head_size_v, sm_scale);
for (int row = 0; row < m_size; ++row) {
flash_attn_softmax<scalar_t, BLOCK_M, BLOCK_N>::apply(
s_i, s_delta, v_prime, s_prime, m_prime, m_size, n_size, padded_n_size, head_size_v, sm_scale, row);
}
// get value and pack
pack_vnni2<scalar_t>(
/* dst */ Btmp,
@@ -344,8 +345,10 @@ void flash_attn_varlen_kernel_impl(
}
}
flash_attn_softmax<scalar_t, BLOCK_M, BLOCK_N>::apply(
s_i, s_delta, v_prime, s_prime, m_prime, m_size, n_size, padded_n_size, head_size_v, sm_scale);
for (int row = 0; row < m_size; ++row) {
flash_attn_softmax<scalar_t, BLOCK_M, BLOCK_N>::apply(
s_i, s_delta, v_prime, s_prime, m_prime, m_size, n_size, padded_n_size, head_size_v, sm_scale, row);
}
// get value and pack
pack_vnni2<scalar_t>(