Handle state_dict prefix

Remove submodule comprehensive debugging tests
Eliminate mask and generate_mask function usage

Author: jennychristopher

comprehensive_submodule_tests/comprehensive_attention_norm_test.py

@@ -92,17 +92,9 @@ def test_image_transformer_inference(batch, num_chunks, mesh_device):
         pt_attention_input,
         force_replicated=True,
     )
-    tt_mask = ttnn.from_torch(
-        attention_mask,
-        device=mesh_device,
-        dtype=ttnn.bfloat16,
-        layout=ttnn.TILE_LAYOUT,
-        memory_config=ttnn.DRAM_MEMORY_CONFIG,
-        mesh_mapper=ttnn.ReplicateTensorToMesh(mesh_device),
-    )
 
     with torch.no_grad():
-        tt_out = tt_model(attention_input, mask=tt_mask, position_embeddings=(cos_t, sin_t))
+        tt_out = tt_model(attention_input, position_embeddings=(cos_t, sin_t))
         reference_output = reference_model(
             pt_attention_input, attention_mask=attention_mask, position_embeddings=(cos, sin)
         )[0]

comprehensive_submodule_tests/comprehensive_conv2d_test.py

@@ -105,16 +105,7 @@ def test_vision_attention(mesh_device, seq_len, batch_size):
         mesh_mapper=ttnn.ReplicateTensorToMesh(mesh_device),
     )
 
-    tt_mask = ttnn.from_torch(
-        attention_mask,
-        device=mesh_device,
-        dtype=ttnn.bfloat8_b,
-        layout=ttnn.TILE_LAYOUT,
-        memory_config=ttnn.DRAM_MEMORY_CONFIG,
-        mesh_mapper=ttnn.ReplicateTensorToMesh(mesh_device),
-    )
-
-    tt_out = tt_model(attention_input, position_embeddings=(cos_t, sin_t), mask=tt_mask)
+    tt_out = tt_model(attention_input, position_embeddings=(cos_t, sin_t))
     tt_output_torch = ttnn.to_torch(tt_out, mesh_composer=ConcatMeshToTensor(mesh_device, dim=-1))[
         :, :, :, : tt_out.shape[-1]
     ]