diff --git a/megatron/model/fused_softmax.py b/megatron/model/fused_softmax.py
index 6fd055f0d..0b4cd99dd 100644
--- a/megatron/model/fused_softmax.py
+++ b/megatron/model/fused_softmax.py
@@ -16,7 +16,12 @@
 
 import torch
 import torch.nn as nn
+
+from megatron import logging
 from megatron.enums import AttnMaskType
+from megatron.model import utils
+
+logger = logging.get_logger(__name__)
 
 class ScaledUpperTriangMaskedSoftmax(torch.autograd.Function):
     """
@@ -134,6 +139,13 @@ def forward(self, input, mask):
     def is_kernel_available(self, mask, b, np, sq, sk):
         attn_batches = b * np
 
+        print("is_kernel_available")
+        print(f"self.self.scaled_masked_softmax_fusion = {self.scaled_masked_softmax_fusion}")
+        print(f"self.input_in_float16 = {self.input_in_float16}")
+        print(f"mask is not None = {mask is not None}")
+        print(f"sq = {sq}")
+        print(f"sk = {sk}")
+        print(f"attn_batches = {attn_batches}")
         if (
             self.scaled_masked_softmax_fusion  # user want to fuse
             and self.input_in_float16  # input must be fp16
@@ -149,10 +161,12 @@ def is_kernel_available(self, mask, b, np, sq, sk):
                     if attn_batches % batch_per_block == 0:
                         return True
                 else:
+                    print(sq,  batch_per_block)
                     if sq % batch_per_block == 0:
                         return True
         return False
 
+    @utils.log_debug_usage(logger, "Using fused softmax")
     def forward_fused_softmax(self, input, mask):
         b, np, sq, sk = input.size()
         scale = self.scale if self.scale is not None else 1.0
@@ -168,6 +182,7 @@ def forward_fused_softmax(self, input, mask):
             # input is 4D tensor (b, np, sq, sk)
             return ScaledMaskedSoftmax.apply(input, mask, scale)
 
+    @utils.log_debug_usage(logger, "Using torch softmax")
     def forward_torch_softmax(self, input, mask):
         if self.input_in_float16 and self.softmax_in_fp32:
             input = input.float()
diff --git a/tests/test_model.py b/tests/test_model.py
index 189fcce01..dfdebfe7c 100644
--- a/tests/test_model.py
+++ b/tests/test_model.py
@@ -6,7 +6,7 @@
 import torch
 
 from megatron import initialize_megatron, get_args, get_tokenizer, global_vars
-from megatron.testing_utils import TestCasePlus, mockenv_context
+from megatron.testing_utils import TestCasePlus, mockenv_context, CaptureStdout
 from megatron.training import setup_model_and_optimizer
 from pretrain_gpt import model_provider as gpt_model_provider, get_batch_pipe as get_gpt_batch_pipe
 from pretrain_prefix_lm import model_provider as prefix_lm_model_provider, get_batch_pipe as get_prefix_lm_batch_pipe
@@ -49,6 +49,10 @@ def get_default_args():
         "--checkpoint-activations": "",
 
         # DATA_ARGS
+
+        # LOGGING_ARGS
+        "--log-level": "debug",
+        "--log-level-replica": "info",
     }
 
 
@@ -257,10 +261,16 @@ def test_prefix_lm_wo_reset_attention_mask(self):
                     # Make sure that the last prefix token predicts the first token.
                     self.assertTrue(loss_mask[batch_id, id -1] == 1)
 
-                model(*input_batch)
+                with CaptureStdout() as cs:
+                    model(*input_batch)
+
+                self.assertIn("Using fused softmax", cs.out)
+                self.assertIn("Using torch softmax", cs.out)
 
                 #TODO: Check all invariants
 
+
+
     def test_gpt_rotary_embeddings(self):
         """Test rotary embeddings"""
         command_args = get_default_args()