From 9e1ce2ba265e2de461685f9206fb9f90cdad1d01 Mon Sep 17 00:00:00 2001
From: Arthur Islamov <arthur@islamov.ai>
Date: Mon, 11 Sep 2023 23:49:47 +0400
Subject: [PATCH] Everything works

---
 .../contrib_ops/cpu/bert/attention_cpu_base.h  |  6 ------
 .../cpu/bert/multihead_attention.cc            | 18 ------------------
 2 files changed, 24 deletions(-)

diff --git a/onnxruntime/contrib_ops/cpu/bert/attention_cpu_base.h b/onnxruntime/contrib_ops/cpu/bert/attention_cpu_base.h
index 040bc7f7bf326..b761b1afd8529 100644
--- a/onnxruntime/contrib_ops/cpu/bert/attention_cpu_base.h
+++ b/onnxruntime/contrib_ops/cpu/bert/attention_cpu_base.h
@@ -200,12 +200,6 @@ class AttentionCPUBase : public AttentionBase {
       });
     }
 
-    std::cout << "Probs before softmax.";
-    for (size_t i = 0; i < total_sequence_length * sequence_length * batch_size * num_heads_; ++i) {
-      std::cout << attention_probs[i] << " ";
-    }
-    std::cout << std::endl;
-
     // attention_probs(B, N, S, T) = Softmax(attention_probs)
     {
       const int N = batch_size * num_heads_ * sequence_length;
diff --git a/onnxruntime/contrib_ops/cpu/bert/multihead_attention.cc b/onnxruntime/contrib_ops/cpu/bert/multihead_attention.cc
index 87008d011b5c8..0b55cb7804c61 100644
--- a/onnxruntime/contrib_ops/cpu/bert/multihead_attention.cc
+++ b/onnxruntime/contrib_ops/cpu/bert/multihead_attention.cc
@@ -139,30 +139,12 @@ Status AddBiasTranspose(const Tensor* qkv,                   // Input: Q/K/V dat
                                });
   }
 
-  std::cout << "After bias add.";
-  std::cout << std::endl;
-  auto tensor = qkv_with_bias.GetMutable<Tensor>();
-  auto data = tensor->MutableData<float>();
-  for (size_t i = 0; i < batch_size * sequence_length * hidden_size; ++i) {
-    std::cout << data[i] << " ";
-  }
-  std::cout << std::endl;
-
   // Reshape Q from BxSxD to BxSxNxH
   ORT_RETURN_IF_ERROR(Reshape_BSD_to_BSNH(qkv_with_bias.GetMutable<Tensor>(), batch_size, sequence_length, num_heads, head_size));
 
   // Transpose Q from BxSxNxH to BxNxSxH
   ORT_RETURN_IF_ERROR(Transpose_BSNH_to_BNSH(qkv_with_bias.GetMutable<Tensor>(), qkv_with_bias_transposed));
 
-  std::cout << "After transpose.";
-  std::cout << std::endl;
-  tensor = qkv_with_bias_transposed.GetMutable<Tensor>();
-  data = tensor->MutableData<float>();
-  for (size_t i = 0; i < batch_size * sequence_length * hidden_size; ++i) {
-    std::cout << data[i] << " ";
-  }
-  std::cout << std::endl;
-
   return Status::OK();
 }