diff --git a/llmfoundry/command_utils/train.py b/llmfoundry/command_utils/train.py
index 3b94e57a6c..46d7e2c61a 100644
--- a/llmfoundry/command_utils/train.py
+++ b/llmfoundry/command_utils/train.py
@@ -65,11 +65,6 @@
 log = logging.getLogger(__name__)
 
 
-ic.configureOutput(includeContext=True)
-install()
-
-ic.disable()
-
 def validate_config(train_config: TrainConfig):
     """Validates compatible model and dataloader selection."""
     # Validate the rest of the config
diff --git a/llmfoundry/models/layers/blocks.py b/llmfoundry/models/layers/blocks.py
index cf07e453fa..c88cf33d1b 100644
--- a/llmfoundry/models/layers/blocks.py
+++ b/llmfoundry/models/layers/blocks.py
@@ -206,7 +206,6 @@ def forward(
                 m = self.norm_2(x)
 
         n = self.apply_ffn(attention_mask, m)
-        ic(x.shape, x.device, m.shape, m.device, n.shape, n.device)
         # In the following line we move the `x` tensor to the same devices as the output of ffn layer. This operation should be a no-op during training.
         # This is done to fix pipeline parallel generation using hf.generate. Please see this comment for details: https://github.com/mosaicml/llm-foundry/pull/1332#issue-2386827204
         x = x.to(device=n.device,
diff --git a/llmfoundry/models/mpt/modeling_mpt.py b/llmfoundry/models/mpt/modeling_mpt.py
index 770af5a9cf..e3f6f0575e 100644
--- a/llmfoundry/models/mpt/modeling_mpt.py
+++ b/llmfoundry/models/mpt/modeling_mpt.py
@@ -966,7 +966,6 @@ def forward(
             if prev_layer_key_value is not None:
                 extra_kwargs['prev_layer_key_value'] = prev_layer_key_value
 
-            ic(type(x), type(past_key_value), type(attn_bias), type(attention_mask))
             x, attn_weights, present = block(
                 x,
                 past_key_value=past_key_value,
@@ -1144,7 +1143,6 @@ def forward(
             use_cache if use_cache is not None else self.config.use_cache
         )
 
-        ic(type(input_ids))
         outputs = self.transformer(
             input_ids=input_ids,
             past_key_values=past_key_values,
@@ -1157,7 +1155,6 @@ def forward(
             inputs_embeds=inputs_embeds,
             position_ids=position_ids,
         )
-        ic(outputs)
 
         if self.lm_head is not None:
             logits = self.lm_head(outputs.last_hidden_state)
diff --git a/llmfoundry/models/utils/tp_strategy.py b/llmfoundry/models/utils/tp_strategy.py
index 158daa6425..35c87b86dd 100644
--- a/llmfoundry/models/utils/tp_strategy.py
+++ b/llmfoundry/models/utils/tp_strategy.py
@@ -17,7 +17,6 @@ def ffn_tp_strategy(model: ComposerModel) -> dict[str, ParallelStyle]:
     # generate layer plan
     layer_plan: dict[str, ParallelStyle] = {}
     for name, _ in model.named_modules():
-        ic(name)
         if name.split('.')[-2:] == ['ffn', 'up_proj']:
             layer_plan[name] = ColwiseParallel(
                 input_layouts = Replicate(),