Arm backend: Reduce atol of some model tests (pytorch#20109)

Change-Id: If516eed4f503d38f9193cc574b70aee36afe64be

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils
@Sebastian-Larsson @robell @rascani

Signed-off-by: Yufeng Shi <yufeng.shi@arm.com>

Author: YufengShi-dudu

backends/arm/test/models/stable_diffusion/test_CLIPTextModelWithProjection.py

@@ -109,7 +109,6 @@ def test_clip_text_with_projection_tosa_INT():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=0.8,
             frobenius_threshold=None,
             cosine_threshold=None,
         )
@@ -132,7 +131,6 @@ def test_clip_text_with_projection_vgf_no_quant():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=4,
             transform_passes=[
                 ConvertInt64ConstOpsToInt32Pass(),
                 ConvertInt64OutputOpsToInt32Pass(),
@@ -159,7 +157,6 @@ def test_clip_text_with_projection_vgf_quant():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=0.8,
             quantize=True,
         )
         pipeline.change_args(

backends/arm/test/models/stable_diffusion/test_SD3Transformer2DModel.py

@@ -117,8 +117,6 @@ def test_sd3_transformer_tosa_FP():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            rtol=1.0,  # TODO: MLETORCH-875: Reduce tolerance of SD3Transformer2DModel with FP and INT
-            atol=4.0,
         )
         pipeline.change_args(
             "check_count.exir", TestSD3Transformer2DModel.ops_after_partitioner_FP
@@ -137,9 +135,7 @@ def test_sd3_transformer_tosa_INT():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            qtol=1.0,  # TODO: MLETORCH-875: Reduce tolerance of SD3Transformer2DModel with FP and INT
-            rtol=1.0,
-            atol=4.0,
+            atol=0.1,  # TODO: MLETORCH-875: Reduce tolerance of SD3Transformer2DModel with FP and INT
             frobenius_threshold=None,
             cosine_threshold=None,
         )
@@ -161,8 +157,6 @@ def test_sd3_transformer_vgf_no_quant():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            rtol=1.0,  # TODO: MLETORCH-875: Reduce tolerance of SD3Transformer2DModel with FP and INT,
-            atol=4.0,
             quantize=False,
         )
         pipeline.change_args(
@@ -184,9 +178,7 @@ def test_sd3_transformer_vgf_quant():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            qtol=1.0,
-            rtol=1.0,  # TODO: MLETORCH-875: Reduce tolerance of SD3Transformer2DModel with FP and INT,
-            atol=4.0,
+            atol=0.1,  # TODO: MLETORCH-875: Reduce tolerance of SD3Transformer2DModel with FP and INT
             quantize=True,
         )
         pipeline.change_args(

backends/arm/test/models/stable_diffusion/test_vae_AutoencoderKL.py

@@ -79,7 +79,7 @@ def test_vae_tosa_INT():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=1.0,  # TODO: MLETORCH-990 Reduce tolerance of vae(AutoencoderKL) with INT
+            atol=0.1,  # TODO: MLETORCH-990 Reduce tolerance of vae(AutoencoderKL) with INT
             frobenius_threshold=None,
             cosine_threshold=None,
         )
@@ -115,7 +115,7 @@ def test_vae_vgf_quant():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=1.0,  # TODO: MLETORCH-990 Reduce tolerance of vae(AutoencoderKL) with INT
+            atol=0.1,  # TODO: MLETORCH-990 Reduce tolerance of vae(AutoencoderKL) with INT
             quantize=True,
         )
         pipeline.run()

backends/arm/test/models/test_T5ForConditionalGeneration_arm.py

@@ -114,7 +114,6 @@ def test_t5_for_conditional_generation_tosa_INT():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=14,  # TODO: MLETORCH-1703: Reduce the tolerance of quantized T5ForConditionalGeneration
             frobenius_threshold=0.3,
         )
         pipeline.change_args(
@@ -162,7 +161,6 @@ def test_t5_for_conditional_generation_vgf_quant():
             aten_op=[],
             exir_op=[],
             use_to_edge_transform_and_lower=True,
-            atol=14,  # TODO: MLETORCH-1703: Reduce the tolerance of quantized T5ForConditionalGeneration
             quantize=True,
         )
         pipeline.change_args(