From e947d11ea06abf75970bb1d9e96b8a4b57ddc29e Mon Sep 17 00:00:00 2001
From: coderfeli <coderfeli@163.com>
Date: Wed, 5 Feb 2025 08:19:41 +0000
Subject: [PATCH] save outputs

---
 example/65_gemm_multiply_multiply/moe_gemm_fp16.cpp | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/example/65_gemm_multiply_multiply/moe_gemm_fp16.cpp b/example/65_gemm_multiply_multiply/moe_gemm_fp16.cpp
index 271ea52818..8cd93e95e7 100644
--- a/example/65_gemm_multiply_multiply/moe_gemm_fp16.cpp
+++ b/example/65_gemm_multiply_multiply/moe_gemm_fp16.cpp
@@ -80,9 +80,10 @@ struct MultiplyMultiply
     // }
 };
 
+
 void preShuffleBuffer(const F16* src, F16* dst, int N, int K, int NXdl)
 {
-    int KPack = 8;
+    int KPack = 16 / sizeof(F16);
     int NLane = NXdl;
     int KLane = 64 / NLane;
 
@@ -226,7 +227,8 @@ int main(int argc, char* argv[])
         else
             sorted_token_ids.mData[i] = tokens;
     }
-
+    expert_ids.savetxt("expert_ids.txt", "int");
+    sorted_token_ids.savetxt("sorted_token_ids.txt", "int");
     Tensor<A0DataType> a0_t_k(HostTensorDescriptor({tokens, K}, {K, 1}));
     Tensor<B0DataType> b0_e_n_k(HostTensorDescriptor({experts, N, K}, {N*K, K, 1}));
     Tensor<B0DataType> b0_preshuffled(HostTensorDescriptor({experts, N, K}, {N*K, K, 1}));