#0: Update BERT end-to-end times and compile times

tenstorrent · May 21, 2024 · e1404ac · e1404ac
1 parent 38145ff
commit e1404ac
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 5 deletions.
diff --git a/models/demos/metal_BERT_large_11/tests/test_perf_bert11.py b/models/demos/metal_BERT_large_11/tests/test_perf_bert11.py
@@ -150,9 +150,9 @@ def run_perf_bert11(
 @pytest.mark.parametrize(
     "batch_size, model_config_str, expected_inference_time, expected_compile_time, inference_iterations",
     (
-        [7, "BFLOAT8_B-SHARDED", 0.0358, 3, 10],
-        [8, "BFLOAT8_B-SHARDED", 0.0358, 3, 10],
-        [12, "BFLOAT8_B-SHARDED", 0.0358, 3, 10],
+        [7, "BFLOAT8_B-SHARDED", 0.0329, 3.1, 10],
+        [8, "BFLOAT8_B-SHARDED", 0.0329, 3.1, 10],
+        [12, "BFLOAT8_B-SHARDED", 0.0329, 3.1, 10],
     ),
 )
 def test_perf_bare_metal(

diff --git a/tests/ttnn/integration_tests/bert/test_performance.py b/tests/ttnn/integration_tests/bert/test_performance.py
@@ -58,8 +58,8 @@ def preprocess_inputs(
 def get_expected_times(bert):
     return {
         ttnn_bert: (0.1, 0.1),
-        ttnn_optimized_bert: (5.4, 0.07),
-        ttnn_optimized_sharded_bert: (5, 0.07),
+        ttnn_optimized_bert: (5.5, 0.07),
+        ttnn_optimized_sharded_bert: (5.1, 0.07),
     }[bert]