#7637: Update bcast unary op (scalar input) to use passed in memory c…

…onfig instead of default interleaved - Manually switch to DRAM output memory config for usages that now run out of L1 - Raise falcon7b batch 1 expected perf
tenstorrent · Jul 5, 2024 · aaab87a · aaab87a
1 parent afec31e
commit aaab87a
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 4 deletions.
diff --git a/models/demos/ttnn_falcon7b/tests/test_perf_device_falcon.py b/models/demos/ttnn_falcon7b/tests/test_perf_device_falcon.py
@@ -10,7 +10,7 @@
 @pytest.mark.parametrize(
     "batch_size, test, expected_perf",
     [
-        [1, "BFLOAT16-L1-falcon_7b-layers_32-prefill_seq256", 3.44],
+        [1, "BFLOAT16-L1-falcon_7b-layers_32-prefill_seq256", 3.666],
         [32, "BFLOAT16-L1-falcon_7b-layers_32-decode_batch32", 139],
     ],
 )

diff --git a/tests/ttnn/unit_tests/operations/test_add.py b/tests/ttnn/unit_tests/operations/test_add.py
@@ -156,7 +156,7 @@ def test_add_attention_scores_to_scalar(device, shape, scalar):
     input_tensor = ttnn.from_torch(
         torch_input_tensor, layout=ttnn.TILE_LAYOUT, device=device, memory_config=ttnn.L1_MEMORY_CONFIG
     )
-    output_tensor = ttnn.add(input_tensor, scalar, memory_config=ttnn.L1_MEMORY_CONFIG)
+    output_tensor = ttnn.add(input_tensor, scalar, memory_config=ttnn.DRAM_MEMORY_CONFIG)
     output_tensor = ttnn.to_torch(output_tensor)
 
     assert ttnn.pearson_correlation_coefficient(torch_output_tensor, output_tensor) >= 0.99988

diff --git a/ttnn/cpp/ttnn/operations/eltwise/binary/binary.hpp b/ttnn/cpp/ttnn/operations/eltwise/binary/binary.hpp
@@ -114,7 +114,7 @@ struct Binary {
             input_tensor_a,
             scalar,
             dtype,
-            operation::DEFAULT_OUTPUT_MEMORY_CONFIG,
+            memory_config,
             optional_output_tensor,
             activations);
     }
@@ -141,7 +141,7 @@ struct Binary {
             input_tensor_a,
             scalar_tensor_device,
             dtype,
-            operation::DEFAULT_OUTPUT_MEMORY_CONFIG,
+            memory_config,
             optional_output_tensor,
             activations);
     }