[CPU] Bump up CPU kernels to latest version (#50387)

Signed-off-by: jiang1.li <jiang1.li@intel.com>
Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
Li, Jiang
2026-07-30 19:18:43 +08:00
committed by GitHub
co-authored by Claude
parent aeeb36b1f1
commit 072a472782
15 changed files with 2181 additions and 1290 deletions
+3 -2
View File
@@ -110,7 +110,7 @@ std::tuple<at::Tensor, at::Tensor> chunk_gated_delta_rule_cpu(
const at::Tensor& g, const at::Tensor& beta,
const at::Tensor& initial_state, bool output_final_state,
const at::Tensor& cu_seqlens, bool head_first, bool use_qk_l2norm_in_kernel,
double eps = 1e-5);
const at::Tensor& initial_state_indices, double eps = 1e-5);
at::Tensor fused_sigmoid_gating_delta_rule_update_cpu(
const at::Tensor& A_log, const at::Tensor& dt_bias, const at::Tensor& q,
@@ -546,7 +546,8 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
"Tensor g, Tensor beta, "
"Tensor initial_state, bool output_final_state, Tensor cu_seqlens, bool "
"head_first, "
"bool use_qk_l2norm_in_kernel, float eps=1e-5) -> (Tensor, Tensor)");
"bool use_qk_l2norm_in_kernel, Tensor initial_state_indices, float "
"eps=1e-5) -> (Tensor, Tensor)");
ops.impl("chunk_gated_delta_rule_cpu", torch::kCPU,
&chunk_gated_delta_rule_cpu);
ops.def(