diff --git a/backends/arm/test/models/test_nss.py b/backends/arm/test/models/test_nss.py index aada338b9c6..e5e381cfe66 100644 --- a/backends/arm/test/models/test_nss.py +++ b/backends/arm/test/models/test_nss.py @@ -110,7 +110,7 @@ def test_nss_u85_INT(): reason="[MLETORCH-1430]: Double types are not supported in buffers in MSL" ) @common.SkipIfNoModelConverter -def test_nss_vgf_no_quant(): +def test_nss_vgf_FP(): pipeline = VgfPipeline[input_t]( nss().eval(), example_inputs(), @@ -119,12 +119,14 @@ def test_nss_vgf_no_quant(): use_to_edge_transform_and_lower=True, run_on_vulkan_runtime=True, quantize=False, + # Override tosa version to test FP-only path + tosa_version="TOSA-1.0+FP", ) pipeline.run() @common.SkipIfNoModelConverter -def test_nss_vgf_quant(): +def test_nss_vgf_INT(): pipeline = VgfPipeline[input_t]( nss().eval(), example_inputs(), @@ -134,6 +136,8 @@ def test_nss_vgf_quant(): use_to_edge_transform_and_lower=True, run_on_vulkan_runtime=True, quantize=True, + # Override tosa version to test INT-only path + tosa_version="TOSA-1.0+INT", ) pipeline.run() diff --git a/backends/arm/test/ops/test_hardsigmoid.py b/backends/arm/test/ops/test_hardsigmoid.py index 568eb069f8b..eb10e5a79e4 100644 --- a/backends/arm/test/ops/test_hardsigmoid.py +++ b/backends/arm/test/ops/test_hardsigmoid.py @@ -90,21 +90,25 @@ def test_hardsigmoid_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_hardsigmoid_vgf_FP(test_data: torch.Tensor): +def test_hardsigmoid_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( - Hardsigmoid(), (test_data(),), aten_op, exir_op=[], tosa_version="TOSA-1.0+FP" + Hardsigmoid(), + (test_data(),), + aten_op, + exir_op=[], + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_hardsigmoid_vgf_INT(test_data: torch.Tensor): +def test_hardsigmoid_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Hardsigmoid(), (test_data(),), aten_op, exir_op=[], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_hardswish.py b/backends/arm/test/ops/test_hardswish.py index 760293ec492..68cd249861a 100644 --- a/backends/arm/test/ops/test_hardswish.py +++ b/backends/arm/test/ops/test_hardswish.py @@ -80,21 +80,25 @@ def test_hardswish_u85_INT(test_data): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_hardswish_vgf_FP(test_data): +def test_hardswish_vgf_no_quant(test_data): pipeline = VgfPipeline[input_t1]( - Hardswish(), (test_data(),), aten_op, exir_op, tosa_version="TOSA-1.0+FP" + Hardswish(), + (test_data(),), + aten_op, + exir_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_hardswish_vgf_INT(test_data): +def test_hardswish_vgf_quant(test_data): pipeline = VgfPipeline[input_t1]( Hardswish(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_hardtanh.py b/backends/arm/test/ops/test_hardtanh.py index 3bb8e212cc9..a13e70d74d0 100644 --- a/backends/arm/test/ops/test_hardtanh.py +++ b/backends/arm/test/ops/test_hardtanh.py @@ -89,21 +89,25 @@ def test_hardtanh_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_hardtanh_vgf_FP(test_data: torch.Tensor): +def test_hardtanh_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t]( - HardTanh(), (test_data(),), aten_op, exir_op, tosa_version="TOSA-1.0+FP" + HardTanh(), + (test_data(),), + aten_op, + exir_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_hardtanh_vgf_INT(test_data: torch.Tensor): +def test_hardtanh_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t]( HardTanh(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_index_select.py b/backends/arm/test/ops/test_index_select.py index 6d2a6d73b70..239c27a8af6 100644 --- a/backends/arm/test/ops/test_index_select.py +++ b/backends/arm/test/ops/test_index_select.py @@ -137,41 +137,41 @@ def test_index_select_u55_INT_not_delegated(test_data: input_params): @pytest.mark.parametrize("test_data", list(test_data.values())) @common.SkipIfNoModelConverter -def test_index_select_vgf_FP(test_data: input_params): +def test_index_select_vgf_no_quant(test_data: input_params): op, inp = test_data pipeline = VgfPipeline[input_params]( op, inp, op.aten_op, op.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @pytest.mark.parametrize("test_data", list(test_data.values())[:-1]) @common.SkipIfNoModelConverter -def test_index_select_vgf_INT(test_data: input_params): +def test_index_select_vgf_quant(test_data: input_params): op, inp = test_data pipeline = VgfPipeline[input_params]( op, inp, op.aten_op, op.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @pytest.mark.parametrize("test_data", list(test_data.values())[-1:]) @common.SkipIfNoModelConverter -def test_index_select_vgf_INT_rand(test_data: input_params): +def test_index_select_vgf_quant_rand(test_data: input_params): op, inp = test_data pipeline = VgfPipeline[input_params]( op, inp, op.aten_op, op.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_layer_norm.py b/backends/arm/test/ops/test_layer_norm.py index f3f11959d4d..3b6db9f644c 100644 --- a/backends/arm/test/ops/test_layer_norm.py +++ b/backends/arm/test/ops/test_layer_norm.py @@ -115,26 +115,26 @@ def test_native_layer_norm_u85_INT(test_data): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_native_layer_norm_vgf_FP(test_data): +def test_native_layer_norm_vgf_no_quant(test_data): test_input, model = test_data() pipeline = VgfPipeline[input_t]( model, test_input, "torch.ops.aten.layer_norm.default", - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_native_layer_norm_vgf_INT(test_data): +def test_native_layer_norm_vgf_quant(test_data): test_input, model = test_data() pipeline = VgfPipeline[input_t]( model, test_input, "torch.ops.aten.sub.Tensor", - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_le.py b/backends/arm/test/ops/test_le.py index cc8ddfc4da2..3d4cc836038 100644 --- a/backends/arm/test/ops/test_le.py +++ b/backends/arm/test/ops/test_le.py @@ -246,51 +246,51 @@ def test_le_scalar_16a8w_u85_INT16(test_module): @common.parametrize("test_module", test_data_tensor) @common.SkipIfNoModelConverter -def test_le_tensor_vgf_FP(test_module): +def test_le_tensor_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessEqual.aten_op_tensor, LessEqual.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_data_tensor) @common.SkipIfNoModelConverter -def test_le_tensor_vgf_INT(test_module): +def test_le_tensor_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessEqual.aten_op_tensor, LessEqual.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_module", test_data_scalar) @common.SkipIfNoModelConverter -def test_le_scalar_vgf_FP(test_module): +def test_le_scalar_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessEqual.aten_op_scalar, LessEqual.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_data_scalar) @common.SkipIfNoModelConverter -def test_le_scalar_vgf_INT(test_module): +def test_le_scalar_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessEqual.aten_op_tensor, LessEqual.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_leaky_relu.py b/backends/arm/test/ops/test_leaky_relu.py index a7ae4cb8564..9be24857563 100644 --- a/backends/arm/test/ops/test_leaky_relu.py +++ b/backends/arm/test/ops/test_leaky_relu.py @@ -95,14 +95,14 @@ def test_leaky_relu_u85_INT(test_data): @common.parametrize("test_data", LeakyReLU.test_data) @common.SkipIfNoModelConverter -def test_leaky_relu_vgf_FP(test_data): +def test_leaky_relu_vgf_no_quant(test_data): data, slope = test_data() pipeline = VgfPipeline[input_t1]( LeakyReLU(slope), data, [], use_to_edge_transform_and_lower=True, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.add_stage_after( "to_edge_transform_and_lower", pipeline.tester.check_not, [aten_op] @@ -112,14 +112,14 @@ def test_leaky_relu_vgf_FP(test_data): @common.parametrize("test_data", LeakyReLU.test_data) @common.SkipIfNoModelConverter -def test_leaky_relu_vgf_INT(test_data): +def test_leaky_relu_vgf_quant(test_data): data, slope = test_data() pipeline = VgfPipeline[input_t1]( LeakyReLU(slope), data, [], use_to_edge_transform_and_lower=True, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.add_stage_after("quantize", pipeline.tester.check_not, [aten_op]) pipeline.run() diff --git a/backends/arm/test/ops/test_linalg_vector_norm.py b/backends/arm/test/ops/test_linalg_vector_norm.py index df3bef38cc1..2723479869e 100644 --- a/backends/arm/test/ops/test_linalg_vector_norm.py +++ b/backends/arm/test/ops/test_linalg_vector_norm.py @@ -128,7 +128,7 @@ def test_vector_norm_u85_INT_fvp(test_module): @common.parametrize("test_module", test_modules) @common.SkipIfNoModelConverter -def test_vector_norm_vgf_FP(test_module): +def test_vector_norm_vgf_no_quant(test_module): model, input_tensor = test_module # FP VGF aten_op = "torch.ops.aten.linalg_vector_norm.default" @@ -138,14 +138,14 @@ def test_vector_norm_vgf_FP(test_module): input_tensor, aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_modules) @common.SkipIfNoModelConverter -def test_vector_norm_vgf_INT(test_module): +def test_vector_norm_vgf_quant(test_module): model, input_tensor = test_module # Should not found this op exir_op = "executorch_exir_dialects_edge__ops_aten_linalg_vector_norm_default" @@ -155,6 +155,6 @@ def test_vector_norm_vgf_INT(test_module): input_tensor, aten_op_q_decomposed_q, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_linear.py b/backends/arm/test/ops/test_linear.py index b63d7c10b34..7f79f7c586b 100644 --- a/backends/arm/test/ops/test_linear.py +++ b/backends/arm/test/ops/test_linear.py @@ -209,39 +209,31 @@ def test_linear_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_rank1_FP | test_data_rank4_FP) @common.SkipIfNoModelConverter -def test_linear_vgf_FP(test_data: torch.Tensor): +def test_linear_vgf_no_quant(test_data: torch.Tensor): test_data, out_features, has_bias = test_data() in_features = test_data.shape[-1] pipeline = VgfPipeline[input_t1]( - Linear( - in_features=in_features, - out_features=out_features, - bias=has_bias, - ), + Linear(in_features=in_features, out_features=out_features, bias=has_bias), (test_data,), aten_op=aten_op, exir_op=[], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_rank1_INT | test_data_rank4_INT) @common.SkipIfNoModelConverter -def test_linear_vgf_INT(test_data: torch.Tensor): +def test_linear_vgf_quant(test_data: torch.Tensor): test_data, out_features, has_bias, per_channel_quantization = test_data() in_features = test_data.shape[-1] pipeline = VgfPipeline[input_t1]( - Linear( - in_features=in_features, - out_features=out_features, - bias=has_bias, - ), + Linear(in_features=in_features, out_features=out_features, bias=has_bias), (test_data,), aten_op=aten_op, exir_op=[], - tosa_version="TOSA-1.0+INT", per_channel_quantization=per_channel_quantization, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_log.py b/backends/arm/test/ops/test_log.py index 44811715407..3f4bfcdb17f 100644 --- a/backends/arm/test/ops/test_log.py +++ b/backends/arm/test/ops/test_log.py @@ -76,25 +76,25 @@ def test_log_u85_INT(test_data: input_t1): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_log_vgf_FP(test_data: input_t1): +def test_log_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Log(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_log_vgf_INT(test_data: input_t1): +def test_log_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Log(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_logical.py b/backends/arm/test/ops/test_logical.py index 8c290c28908..a2a82793170 100644 --- a/backends/arm/test/ops/test_logical.py +++ b/backends/arm/test/ops/test_logical.py @@ -144,26 +144,26 @@ def test_logical_and_u85_INT(test_data: input_t2): @common.parametrize("test_data", And().test_data) @common.SkipIfNoModelConverter -def test_logical_and_vgf_FP(test_data: input_t2): +def test_logical_and_vgf_no_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( And(), test_data(), And().aten_op, And().exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", And().test_data) @common.SkipIfNoModelConverter -def test_logical_and_vgf_INT(test_data: input_t2): +def test_logical_and_vgf_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( And(), test_data(), And().aten_op, And().exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -231,26 +231,26 @@ def test_logical_xor_u85_INT(test_data: input_t2): @common.parametrize("test_data", Xor().test_data) @common.SkipIfNoModelConverter -def test_logical_xor_vgf_FP(test_data: input_t2): +def test_logical_xor_vgf_no_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( Xor(), test_data(), Xor().aten_op, Xor().exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Xor().test_data) @common.SkipIfNoModelConverter -def test_logical_xor_vgf_INT(test_data: input_t2): +def test_logical_xor_vgf_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( Xor(), test_data(), Xor().aten_op, Xor().exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -318,26 +318,26 @@ def test_logical_or_u85_INT(test_data: input_t2): @common.parametrize("test_data", Or().test_data) @common.SkipIfNoModelConverter -def test_logical_or_vgf_FP(test_data: input_t2): +def test_logical_or_vgf_no_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( Or(), test_data(), Or().aten_op, Or().exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Or().test_data) @common.SkipIfNoModelConverter -def test_logical_or_vgf_INT(test_data: input_t2): +def test_logical_or_vgf_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( Or(), test_data(), Or().aten_op, Or().exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -405,25 +405,25 @@ def test_logical_not_u85_INT(test_data: input_t2): @common.parametrize("test_data", Not().test_data) @common.SkipIfNoModelConverter -def test_logical_not_vgf_FP(test_data: input_t2): +def test_logical_not_vgf_no_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( Not(), test_data(), Not().aten_op, Not().exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Not().test_data) @common.SkipIfNoModelConverter -def test_logical_not_vgf_INT(test_data: input_t2): +def test_logical_not_vgf_quant(test_data: input_t2): pipeline = VgfPipeline[input_t2]( Not(), test_data(), Not().aten_op, Not().exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_logit.py b/backends/arm/test/ops/test_logit.py index 8915c151bb9..b628504c716 100644 --- a/backends/arm/test/ops/test_logit.py +++ b/backends/arm/test/ops/test_logit.py @@ -92,13 +92,13 @@ def test_logit_u85_INT(test_data: Tuple): test_data_suite, ) @common.SkipIfNoModelConverter -def test_logit_vgf_FP(test_data: input_t1): +def test_logit_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Logit(), (*test_data,), [], [], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -108,12 +108,12 @@ def test_logit_vgf_FP(test_data: input_t1): test_data_suite, ) @common.SkipIfNoModelConverter -def test_logit_vgf_INT(test_data: input_t1): +def test_logit_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Logit(), (*test_data,), [], [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_logsoftmax.py b/backends/arm/test/ops/test_logsoftmax.py index f0411847dd3..8d090b660ae 100644 --- a/backends/arm/test/ops/test_logsoftmax.py +++ b/backends/arm/test/ops/test_logsoftmax.py @@ -94,10 +94,14 @@ def test_log_softmax_u85_INT(test_data): @common.parametrize("test_data", LogSoftmax.test_data) @common.SkipIfNoModelConverter -def test_log_softmax_vgf_FP(test_data): +def test_log_softmax_vgf_no_quant(test_data): data, dim = test_data() pipeline = VgfPipeline[input_t1]( - LogSoftmax(dim), data, [], [], tosa_version="TOSA-1.0+FP" + LogSoftmax(dim), + data, + [], + [], + quantize=False, ) pipeline.add_stage_after( "to_edge_transform_and_lower", pipeline.tester.check_not, [aten_op] @@ -107,14 +111,14 @@ def test_log_softmax_vgf_FP(test_data): @common.parametrize("test_data", LogSoftmax.test_data) @common.SkipIfNoModelConverter -def test_log_softmax_vgf_INT(test_data): +def test_log_softmax_vgf_quant(test_data): data, dim = test_data() pipeline = VgfPipeline[input_t1]( LogSoftmax(dim), data, [], [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.add_stage_after("quantize", pipeline.tester.check_not, [aten_op]) pipeline.run() diff --git a/backends/arm/test/ops/test_lshift.py b/backends/arm/test/ops/test_lshift.py index 1d4224a8efe..878b18fc805 100644 --- a/backends/arm/test/ops/test_lshift.py +++ b/backends/arm/test/ops/test_lshift.py @@ -120,26 +120,26 @@ def test_bitwise_left_shift_tensor_u85_INT_scalar(test_data): @common.parametrize("test_data", LshiftScalar.test_data) @common.SkipIfNoModelConverter -def test_bitwise_left_shift_scalar_vgf_FP_scalar(test_data: scalar_input_t): +def test_bitwise_left_shift_scalar_scalar_vgf_no_quant(test_data: scalar_input_t): pipeline = VgfPipeline[scalar_input_t]( LshiftScalar(), test_data, LshiftScalar.torch_op_FP, LshiftScalar.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", LshiftScalar.test_data) @common.SkipIfNoModelConverter -def test_bitwise_left_shift_tensor_vgf_INT_scalar(test_data: scalar_input_t): +def test_bitwise_left_shift_tensor_scalar_vgf_quant(test_data: scalar_input_t): pipeline = VgfPipeline[scalar_input_t]( LshiftScalar(), test_data, LshiftScalar.torch_op_INT, LshiftScalar.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -196,25 +196,25 @@ def test_bitwise_left_shift_tensor_u85_INT(test_data): @common.parametrize("test_data", LshiftTensor.test_data) @common.SkipIfNoModelConverter -def test_bitwise_left_shift_tensor_vgf_FP(test_data: tensor_input_t): +def test_bitwise_left_shift_tensor_vgf_no_quant(test_data: tensor_input_t): pipeline = VgfPipeline[tensor_input_t]( LshiftTensor(), test_data, LshiftTensor.torch_op, LshiftTensor.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", LshiftTensor.test_data) @common.SkipIfNoModelConverter -def test_bitwise_left_shift_tensor_vgf_INT(test_data: tensor_input_t): +def test_bitwise_left_shift_tensor_vgf_quant(test_data: tensor_input_t): pipeline = VgfPipeline[tensor_input_t]( LshiftTensor(), test_data, LshiftTensor.torch_op, LshiftTensor.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_lt.py b/backends/arm/test/ops/test_lt.py index 22958208bcd..e260cd5b75d 100644 --- a/backends/arm/test/ops/test_lt.py +++ b/backends/arm/test/ops/test_lt.py @@ -243,51 +243,51 @@ def test_lt_scalar_16a8w_u85_INT16(test_module): @common.parametrize("test_module", test_data_tensor) @common.SkipIfNoModelConverter -def test_lt_tensor_vgf_FP(test_module): +def test_lt_tensor_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessThan.aten_op_tensor, LessThan.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_data_scalar) @common.SkipIfNoModelConverter -def test_lt_scalar_vgf_FP(test_module): +def test_lt_scalar_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessThan.aten_op_scalar, LessThan.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_data_tensor) @common.SkipIfNoModelConverter -def test_lt_tensor_vgf_INT(test_module): +def test_lt_tensor_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessThan.aten_op_tensor, LessThan.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_module", test_data_scalar) @common.SkipIfNoModelConverter -def test_lt_scalar_vgf_INT(test_module): +def test_lt_scalar_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), LessThan.aten_op_tensor, LessThan.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_masked_fill.py b/backends/arm/test/ops/test_masked_fill.py index 3aab19925ec..2704fa53257 100644 --- a/backends/arm/test/ops/test_masked_fill.py +++ b/backends/arm/test/ops/test_masked_fill.py @@ -147,19 +147,25 @@ def test_masked_fill_scalar_u85_INT(test_module): @common.parametrize("test_module", test_modules) @common.SkipIfNoModelConverter -def test_masked_fill_scalar_vgf_FP(test_module): +def test_masked_fill_scalar_vgf_no_quant(test_module): module, inputs = test_module() pipeline = VgfPipeline[input_t]( - module, inputs, aten_op=[], tosa_version="TOSA-1.0+FP" + module, + inputs, + aten_op=[], + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_modules) @common.SkipIfNoModelConverter -def test_masked_fill_scalar_vgf_INT(test_module): +def test_masked_fill_scalar_vgf_quant(test_module): module, inputs = test_module() pipeline = VgfPipeline[input_t]( - module, inputs, aten_op=[], tosa_version="TOSA-1.0+INT" + module, + inputs, + aten_op=[], + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_matmul.py b/backends/arm/test/ops/test_matmul.py index 0baf609ce45..489078a6dfa 100644 --- a/backends/arm/test/ops/test_matmul.py +++ b/backends/arm/test/ops/test_matmul.py @@ -225,69 +225,77 @@ def test_matmul_combo_u85_INT(test_data: input_t1): @common.parametrize("test_data", MatMul.test_data_generators) @common.SkipIfNoModelConverter -def test_matmul_vgf_FP(test_data: input_t1): +def test_matmul_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( - MatMul(), test_data(), aten_op_mm, exir_op_mm, tosa_version="TOSA-1.0+FP" + MatMul(), + test_data(), + aten_op_mm, + exir_op_mm, + quantize=False, ) pipeline.run() @common.parametrize("test_data", MatMulSingleInput.test_data_generators) @common.SkipIfNoModelConverter -def test_matmul_single_input_vgf_FP(test_data: input_t1): +def test_matmul_single_input_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( MatMulSingleInput(), test_data(), aten_op_mm, exir_op_mm, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", MatMulCombo.test_data_generators) @common.SkipIfNoModelConverter -def test_matmul_combo_vgf_FP(test_data: input_t1): +def test_matmul_combo_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( - MatMulCombo(), test_data(), aten_op_mm, exir_op_mm, tosa_version="TOSA-1.0+FP" + MatMulCombo(), + test_data(), + aten_op_mm, + exir_op_mm, + quantize=False, ) pipeline.run() @common.parametrize("test_data", MatMul.test_data_generators) @common.SkipIfNoModelConverter -def test_matmul_vgf_INT(test_data: input_t1): +def test_matmul_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( MatMul(), test_data(), aten_op_mm, exir_op_mm, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", MatMulSingleInput.test_data_generators) @common.SkipIfNoModelConverter -def test_matmul_single_input_vgf_INT(test_data: input_t1): +def test_matmul_single_input_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( MatMulSingleInput(), test_data(), aten_op_mm, exir_op_mm, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", MatMulCombo.test_data_generators) @common.SkipIfNoModelConverter -def test_matmul_combo_vgf_INT(test_data: input_t1): +def test_matmul_combo_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( MatMulCombo(), test_data(), aten_op_mm, exir_op_mm, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_max_pool.py b/backends/arm/test/ops/test_max_pool.py index 21619afa7a3..5c780a9bcb1 100644 --- a/backends/arm/test/ops/test_max_pool.py +++ b/backends/arm/test/ops/test_max_pool.py @@ -269,35 +269,35 @@ def test_max_pool2d_tosa_INT_dilation(test_data): # VGF tests @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_max_pool2d_vgf_FP(test_data: torch.Tensor): +def test_max_pool2d_vgf_no_quant(test_data: torch.Tensor): test_data, model_params = test_data() pipeline = VgfPipeline[input_t1]( MaxPool2d(*model_params), (test_data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_max_pool2d_vgf_INT(test_data: torch.Tensor): +def test_max_pool2d_vgf_quant(test_data: torch.Tensor): test_data, model_params = test_data() pipeline = VgfPipeline[input_t1]( MaxPool2d(*model_params), (test_data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", dilation_test_data) @common.SkipIfNoModelConverter -def test_max_pool2d_vgf_FP_dilation(test_data: torch.Tensor): +def test_max_pool2d_dilation_vgf_no_quant(test_data: torch.Tensor): """ VGF FP pipeline with dilation > 1 (and dilation=1 sanity cases). """ @@ -307,14 +307,14 @@ def test_max_pool2d_vgf_FP_dilation(test_data: torch.Tensor): (test_data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", dilation_test_data) @common.SkipIfNoModelConverter -def test_max_pool2d_vgf_INT_dilation(test_data: torch.Tensor): +def test_max_pool2d_dilation_vgf_quant(test_data: torch.Tensor): """ VGF INT pipeline with dilation > 1 (and dilation=1 sanity cases). """ @@ -324,6 +324,6 @@ def test_max_pool2d_vgf_INT_dilation(test_data: torch.Tensor): (test_data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_maximum.py b/backends/arm/test/ops/test_maximum.py index ed3a5247d3d..e213842494f 100644 --- a/backends/arm/test/ops/test_maximum.py +++ b/backends/arm/test/ops/test_maximum.py @@ -76,23 +76,23 @@ def test_maximum_u85_INT(test_data: Tuple): @common.parametrize("test_data", Maximum.test_parameters) @common.SkipIfNoModelConverter -def test_maximum_vgf_FP(test_data: Tuple): +def test_maximum_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[test_t]( Maximum(), test_data(), aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Maximum.test_parameters) @common.SkipIfNoModelConverter -def test_maximum_vgf_INT(test_data: Tuple): +def test_maximum_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[test_t]( Maximum(), test_data(), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_mean_dim.py b/backends/arm/test/ops/test_mean_dim.py index babfb7d10da..5195d955a1a 100644 --- a/backends/arm/test/ops/test_mean_dim.py +++ b/backends/arm/test/ops/test_mean_dim.py @@ -85,27 +85,27 @@ def test_adaptive_avg_pool2d_u85_INT(test_data): @common.parametrize("test_data", AdaptiveAveragePool2d.test_data_suite) @common.SkipIfNoModelConverter -def test_adaptive_avg_pool2d_vgf_FP(test_data): +def test_adaptive_avg_pool2d_vgf_no_quant(test_data): pipeline = VgfPipeline[input_t]( AdaptiveAveragePool2d(), test_data(), AdaptiveAveragePool2d.aten_op, AdaptiveAveragePool2d.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", AdaptiveAveragePool2d.test_data_suite) @common.SkipIfNoModelConverter -def test_adaptive_avg_pool2d_vgf_INT(test_data): +def test_adaptive_avg_pool2d_vgf_quant(test_data): pipeline = VgfPipeline[input_t]( AdaptiveAveragePool2d(), test_data(), AdaptiveAveragePool2d.aten_op, AdaptiveAveragePool2d.exir_op, symmetric_io_quantization=True, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -322,28 +322,28 @@ def test_mean_dim_u85_INT(test_data): @common.parametrize("test_data", MeanDim.test_data_suite) @common.SkipIfNoModelConverter -def test_mean_dim_vgf_FP(test_data): +def test_mean_dim_vgf_no_quant(test_data): test_data_val, dim, keep_dim = test_data() pipeline = VgfPipeline[input_t]( MeanDim(dim, keep_dim), (test_data_val,), MeanDim.torch_op, MeanDim.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", MeanDim.test_data_suite) @common.SkipIfNoModelConverter -def test_mean_dim_vgf_INT(test_data): +def test_mean_dim_vgf_quant(test_data): test_data_val, dim, keep_dim = test_data() pipeline = VgfPipeline[input_t]( MeanDim(dim, keep_dim), (test_data_val,), - [], # Might be sum, avgpool, or both + [], symmetric_io_quantization=True, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_minimum.py b/backends/arm/test/ops/test_minimum.py index 3e87e64acbd..ff706f7261e 100644 --- a/backends/arm/test/ops/test_minimum.py +++ b/backends/arm/test/ops/test_minimum.py @@ -76,18 +76,23 @@ def test_minimum_u85_INT(test_data: Tuple): @common.parametrize("test_data", Minimum.test_parameters) @common.SkipIfNoModelConverter -def test_minimum_vgf_FP(test_data: test_t): - pipeline = VgfPipeline[test_t](Minimum(), test_data(), aten_op) +def test_minimum_vgf_no_quant(test_data: test_t): + pipeline = VgfPipeline[test_t]( + Minimum(), + test_data(), + aten_op, + quantize=False, + ) pipeline.run() @common.parametrize("test_data", Minimum.test_parameters) @common.SkipIfNoModelConverter -def test_minimum_vgf_INT(test_data: test_t): +def test_minimum_vgf_quant(test_data: test_t): pipeline = VgfPipeline[test_t]( Minimum(), test_data(), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_mm.py b/backends/arm/test/ops/test_mm.py index afb7a6d7d30..6d026888027 100644 --- a/backends/arm/test/ops/test_mm.py +++ b/backends/arm/test/ops/test_mm.py @@ -69,21 +69,25 @@ def test_mm_u85_INT(test_data: Tuple): @common.parametrize("test_data", MM.test_data_generators) @common.SkipIfNoModelConverter -def test_mm_vgf_FP(test_data: Tuple): +def test_mm_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[test_t]( - MM(), test_data(), MM.aten_op, MM.exir_op, tosa_version="TOSA-1.0+FP" + MM(), + test_data(), + MM.aten_op, + MM.exir_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", MM.test_data_generators) @common.SkipIfNoModelConverter -def test_mm_vgf_INT(test_data: Tuple): +def test_mm_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[test_t]( MM(), test_data(), MM.aten_op, MM.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_mul.py b/backends/arm/test/ops/test_mul.py index 2e40a244983..0cff1bb1d92 100644 --- a/backends/arm/test/ops/test_mul.py +++ b/backends/arm/test/ops/test_mul.py @@ -248,39 +248,39 @@ def test_mul_tensor_u85_INT_int32(test_data: torch.Tensor): test_data_suite | test_data_suite_2 | test_data_int32_without_broadcasting, ) @common.SkipIfNoModelConverter -def test_mul_tensor_vgf_FP(test_data: torch.Tensor): +def test_mul_tensor_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Mul(), test_data(), aten_op, exir_op=[], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite | test_data_suite_2) @common.SkipIfNoModelConverter -def test_mul_tensor_vgf_INT(test_data: torch.Tensor): +def test_mul_tensor_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Mul(), test_data(), aten_op, exir_op=[], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", test_data_suite_int32) @common.SkipIfNoModelConverter -def test_mul_tensor_vgf_INT_int32(test_data: torch.Tensor): +def test_mul_tensor_int32_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Mul(), test_data(), aten_op, exir_op=[], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_multihead_attention.py b/backends/arm/test/ops/test_multihead_attention.py index cbc2ccb32f4..50dcaae4635 100644 --- a/backends/arm/test/ops/test_multihead_attention.py +++ b/backends/arm/test/ops/test_multihead_attention.py @@ -108,14 +108,14 @@ def test_multihead_attention_u85_INT(test_data: input_t1): test_suite, ) @common.SkipIfNoModelConverter -def test_multihead_attention_vgf_FP(test_data: input_t1): +def test_multihead_attention_vgf_no_quant(test_data: input_t1): test_data_vals, module = test_data() pipeline = VgfPipeline[input_t1]( module, (*test_data_vals, *test_data_vals, *test_data_vals), [], [], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -125,15 +125,14 @@ def test_multihead_attention_vgf_FP(test_data: input_t1): test_suite, ) @common.SkipIfNoModelConverter -def test_multihead_attention_vgf_INT(test_data: input_t1): +def test_multihead_attention_vgf_quant(test_data: input_t1): test_data_vals, module = test_data() pipeline = VgfPipeline[input_t1]( module, (*test_data_vals, *test_data_vals, *test_data_vals), [], [], - tosa_version="TOSA-1.0+INT", - # TODO: Per-channel quantization is broken (MLETORCH-1144) per_channel_quantization=False, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_ne.py b/backends/arm/test/ops/test_ne.py index 69f9440d549..9fa1b1d96eb 100644 --- a/backends/arm/test/ops/test_ne.py +++ b/backends/arm/test/ops/test_ne.py @@ -193,51 +193,51 @@ def test_ne_scalar_u85_INT(test_module): @common.parametrize("test_module", test_data_tensor) @common.SkipIfNoModelConverter -def test_ne_tensor_vgf_FP(test_module): +def test_ne_tensor_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module, test_module.get_inputs(), NotEqual.aten_op_Tensor, NotEqual.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_data_tensor) @common.SkipIfNoModelConverter -def test_ne_tensor_vgf_INT(test_module): +def test_ne_tensor_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module, test_module.get_inputs(), NotEqual.decomposed_ops, NotEqual.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_module", test_data_scalar) @common.SkipIfNoModelConverter -def test_ne_scalar_vgf_FP(test_module): +def test_ne_scalar_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module, test_module.get_inputs(), NotEqual.aten_op_Scalar, NotEqual.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_data_scalar) @common.SkipIfNoModelConverter -def test_ne_scalar_vgf_INT(test_module): +def test_ne_scalar_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module, test_module.get_inputs(), NotEqual.decomposed_ops, NotEqual.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_neg.py b/backends/arm/test/ops/test_neg.py index f0afe7bd23b..11d1153a171 100644 --- a/backends/arm/test/ops/test_neg.py +++ b/backends/arm/test/ops/test_neg.py @@ -75,21 +75,25 @@ def test_neg_u85_INT(test_data: input_t1): @common.parametrize("test_data", Neg.test_data) @common.SkipIfNoModelConverter -def test_neg_vgf_FP(test_data: input_t1): +def test_neg_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( - Neg(), test_data, Neg.aten_op, Neg.exir_op, tosa_version="TOSA-1.0+FP" + Neg(), + test_data, + Neg.aten_op, + Neg.exir_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", Neg.test_data) @common.SkipIfNoModelConverter -def test_neg_vgf_INT(test_data: input_t1): +def test_neg_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Neg(), test_data, Neg.aten_op, Neg.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_ones.py b/backends/arm/test/ops/test_ones.py index 53351bfff53..48c75906579 100644 --- a/backends/arm/test/ops/test_ones.py +++ b/backends/arm/test/ops/test_ones.py @@ -124,23 +124,26 @@ def test_ones_tosa_INT_not_delegated(test_data: test_data_t): @common.parametrize("test_data", OnesAdd.test_data) @common.SkipIfNoModelConverter -def test_ones_vgf_FP(test_data: test_data_t): +def test_ones_vgf_no_quant(test_data: test_data_t): input_data, init_data = test_data pipeline = VgfPipeline[input_t]( - OnesAdd(*init_data), input_data(), OnesAdd.aten_op, tosa_version="TOSA-1.0+FP" + OnesAdd(*init_data), + input_data(), + OnesAdd.aten_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", OnesAdd.test_data) @common.SkipIfNoModelConverter -def test_ones_vgf_INT(test_data: test_data_t): +def test_ones_vgf_quant(test_data: test_data_t): input_data, init_data = test_data pipeline = VgfPipeline[input_t]( OnesAdd(*init_data), input_data(), OnesAdd.aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) # Pop the quantization check stage if it exists as no # quantization nodes will be present for int + fp inputs. diff --git a/backends/arm/test/ops/test_permute.py b/backends/arm/test/ops/test_permute.py index a1e97241d68..b507155c8f2 100644 --- a/backends/arm/test/ops/test_permute.py +++ b/backends/arm/test/ops/test_permute.py @@ -130,28 +130,28 @@ def test_permute_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_permute_vgf_FP(test_data): +def test_permute_vgf_no_quant(test_data): test_data, dims = test_data() pipeline = VgfPipeline[input_t1]( SimplePermute(dims=dims), (test_data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_permute_vgf_INT(test_data): +def test_permute_vgf_quant(test_data): test_data, dims = test_data() pipeline = VgfPipeline[input_t1]( SimplePermute(dims=dims), (test_data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_pixel_shuffling.py b/backends/arm/test/ops/test_pixel_shuffling.py index dd9dfbe62cc..0c3436da87e 100644 --- a/backends/arm/test/ops/test_pixel_shuffling.py +++ b/backends/arm/test/ops/test_pixel_shuffling.py @@ -123,56 +123,56 @@ def test_pixel_shuffle_tosa_INT(test_data: input_t1): @common.parametrize("test_data", PixelUnShuffle.test_data_generators) @common.SkipIfNoModelConverter -def test_pixel_unshuffle_vgf_FP(test_data: input_t1): +def test_pixel_unshuffle_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( PixelUnShuffle(), test_data(), aten_op_pixel_unshuffle, exir_op_pixel_unshuffle, - tosa_version="TOSA-1.0+FP", run_on_vulkan_runtime=True, + quantize=False, ) pipeline.run() @common.parametrize("test_data", PixelUnShuffle.test_data_generators) @common.SkipIfNoModelConverter -def test_pixel_unshuffle_vgf_INT(test_data: input_t1): +def test_pixel_unshuffle_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( PixelUnShuffle(), test_data(), aten_op_pixel_unshuffle, exir_op_pixel_unshuffle, - tosa_version="TOSA-1.0+INT", run_on_vulkan_runtime=True, + quantize=True, ) pipeline.run() @common.parametrize("test_data", PixelShuffle.test_data_generators) @common.SkipIfNoModelConverter -def test_pixel_shuffle_vgf_FP(test_data: input_t1): +def test_pixel_shuffle_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( PixelShuffle(), test_data(), aten_op_pixel_shuffle, exir_op_pixel_shuffle, - tosa_version="TOSA-1.0+FP", run_on_vulkan_runtime=True, + quantize=False, ) pipeline.run() @common.parametrize("test_data", PixelShuffle.test_data_generators) @common.SkipIfNoModelConverter -def test_pixel_shuffle_vgf_INT(test_data: input_t1): +def test_pixel_shuffle_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( PixelShuffle(), test_data(), aten_op_pixel_shuffle, exir_op_pixel_shuffle, - tosa_version="TOSA-1.0+INT", run_on_vulkan_runtime=True, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_pow.py b/backends/arm/test/ops/test_pow.py index c11bc985101..1955ff43587 100644 --- a/backends/arm/test/ops/test_pow.py +++ b/backends/arm/test/ops/test_pow.py @@ -105,13 +105,13 @@ def test_pow_tensor_tensor_tosa_FP(test_data: Pow_TensorTensor.input_t): @common.parametrize("test_data", Pow_TensorTensor.test_data, x_fail, strict=False) @common.SkipIfNoModelConverter -def test_pow_tensor_tensor_vgf_FP(test_data: Pow_TensorTensor.input_t): +def test_pow_tensor_tensor_vgf_no_quant(test_data: Pow_TensorTensor.input_t): pipeline = VgfPipeline[Pow_TensorTensor.input_t]( Pow_TensorTensor(), test_data(), Pow_TensorTensor.aten_op, Pow_TensorTensor.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -175,14 +175,14 @@ def test_pow_tensor_scalar_u85_INT(test_data: Pow_TensorScalar.input_t): @common.parametrize("test_data", Pow_TensorScalar.test_data, x_fail, strict=False) @common.SkipIfNoModelConverter -def test_pow_tensor_scalar_vgf_FP(test_data: Pow_TensorScalar.input_t): +def test_pow_tensor_scalar_vgf_no_quant(test_data: Pow_TensorScalar.input_t): base, exp = test_data() pipeline = VgfPipeline[Pow_TensorScalar.input_t]( Pow_TensorScalar(exp), (base,), Pow_TensorScalar.aten_op, Pow_TensorScalar.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -192,13 +192,13 @@ def test_pow_tensor_scalar_vgf_FP(test_data: Pow_TensorScalar.input_t): Pow_TensorScalar.test_data, ) @common.SkipIfNoModelConverter -def test_pow_tensor_scalar_vgf_INT(test_data: Pow_TensorScalar.input_t): +def test_pow_tensor_scalar_vgf_quant(test_data: Pow_TensorScalar.input_t): base, exp = test_data() pipeline = VgfPipeline[Pow_TensorScalar.input_t]( Pow_TensorScalar(exp), (base,), Pow_TensorScalar.aten_op, Pow_TensorScalar.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_reciprocal.py b/backends/arm/test/ops/test_reciprocal.py index 3e4d7c18b40..5d09dfd9268 100644 --- a/backends/arm/test/ops/test_reciprocal.py +++ b/backends/arm/test/ops/test_reciprocal.py @@ -90,23 +90,23 @@ def test_reciprocal_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_reciprocal_vgf_FP(test_data: torch.Tensor): +def test_reciprocal_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Reciprocal(), (test_data(),), aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_reciprocal_vgf_INT(test_data: torch.Tensor): +def test_reciprocal_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Reciprocal(), (test_data(),), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_relu.py b/backends/arm/test/ops/test_relu.py index fad6e7a9162..f659a3c86cb 100644 --- a/backends/arm/test/ops/test_relu.py +++ b/backends/arm/test/ops/test_relu.py @@ -142,25 +142,25 @@ def test_relu_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_relu_vgf_FP(test_data: torch.Tensor): +def test_relu_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Relu(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_relu_vgf_INT(test_data: torch.Tensor): +def test_relu_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Relu(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_remainder.py b/backends/arm/test/ops/test_remainder.py index 336d4db1a7b..d1874d15fdb 100644 --- a/backends/arm/test/ops/test_remainder.py +++ b/backends/arm/test/ops/test_remainder.py @@ -149,51 +149,51 @@ def test_remainder_scalar_u85_INT(test_data): @common.parametrize("test_data", Remainder.test_cases_tensor) @common.SkipIfNoModelConverter -def test_remainder_tensor_vgf_FP(test_data): +def test_remainder_tensor_vgf_no_quant(test_data): data = test_data() pipeline = VgfPipeline[Remainder.input_t]( Remainder(), data, Remainder.aten_op_tensor, Remainder.exir_op_tensor, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Remainder.test_cases_scalar) @common.SkipIfNoModelConverter -def test_remainder_scalar_vgf_FP(test_data): +def test_remainder_scalar_vgf_no_quant(test_data): data = test_data() pipeline = VgfPipeline[Remainder.input_t]( Remainder(), data, Remainder.aten_op_scalar, Remainder.exir_op_scalar, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Remainder.test_cases_tensor) @common.SkipIfNoModelConverter -def test_remainder_tensor_vgf_INT(test_data): +def test_remainder_tensor_vgf_quant(test_data): pipeline = VgfPipeline[Remainder.input_t]( Remainder(), test_data(), [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", Remainder.test_cases_scalar) @common.SkipIfNoModelConverter -def test_remainder_scalar_vgf_INT(test_data): +def test_remainder_scalar_vgf_quant(test_data): pipeline = VgfPipeline[Remainder.input_t]( Remainder(), test_data(), [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_repeat.py b/backends/arm/test/ops/test_repeat.py index c9a2b90f1c8..0b3de3b72df 100644 --- a/backends/arm/test/ops/test_repeat.py +++ b/backends/arm/test/ops/test_repeat.py @@ -138,25 +138,25 @@ def test_repeat_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_repeat_vgf_FP(test_data: Tuple): +def test_repeat_vgf_no_quant(test_data: Tuple): module, args = test_data() pipeline = VgfPipeline[input_t1]( module, args, module.aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_repeat_vgf_INT(test_data: Tuple): +def test_repeat_vgf_quant(test_data: Tuple): module, args = test_data() pipeline = VgfPipeline[input_t1]( module, args, module.aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_round.py b/backends/arm/test/ops/test_round.py index a4fea455e4f..572163c250a 100644 --- a/backends/arm/test/ops/test_round.py +++ b/backends/arm/test/ops/test_round.py @@ -87,25 +87,25 @@ def test_round_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_round_vgf_FP(test_data: torch.Tensor): +def test_round_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Round(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_round_vgf_INT(test_data: torch.Tensor): +def test_round_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Round(), (test_data(),), [], exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_rshift.py b/backends/arm/test/ops/test_rshift.py index 40258907b1e..ea7cd4092fa 100644 --- a/backends/arm/test/ops/test_rshift.py +++ b/backends/arm/test/ops/test_rshift.py @@ -123,26 +123,26 @@ def test_bitwise_right_shift_tensor_u85_INT_scalar(test_data): @common.parametrize("test_data", RshiftScalar.test_data) @common.SkipIfNoModelConverter -def test_bitwise_right_shift_scalar_vgf_FP_scalar(test_data): +def test_bitwise_right_shift_tensor_vgf_no_quant_scalar(test_data): pipeline = VgfPipeline[scalar_input_t]( RshiftScalar(), test_data(), RshiftScalar.torch_op_FP, RshiftScalar.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", RshiftScalar.test_data) @common.SkipIfNoModelConverter -def test_bitwise_right_shift_tensor_vgf_INT_scalar(test_data): +def test_bitwise_right_shift_tensor_vgf_quant_scalar(test_data): pipeline = VgfPipeline[scalar_input_t]( RshiftScalar(), test_data(), RshiftScalar.torch_op_INT, RshiftScalar.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -202,25 +202,25 @@ def test_bitwise_right_shift_tensor_u85_INT(test_data): @common.parametrize("test_data", RshiftTensor.test_data) @common.SkipIfNoModelConverter -def test_bitwise_right_shift_tensor_vgf_FP(test_data): +def test_bitwise_right_shift_tensor_vgf_no_quant(test_data): pipeline = VgfPipeline[tensor_input_t]( RshiftTensor(), test_data(), RshiftTensor.torch_op, RshiftTensor.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", RshiftTensor.test_data) @common.SkipIfNoModelConverter -def test_bitwise_right_shift_tensor_vgf_INT(test_data): +def test_bitwise_right_shift_tensor_vgf_quant(test_data): pipeline = VgfPipeline[tensor_input_t]( RshiftTensor(), test_data(), RshiftTensor.torch_op, RshiftTensor.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_rsqrt.py b/backends/arm/test/ops/test_rsqrt.py index 0fea7ba2ec0..8c3ee914758 100644 --- a/backends/arm/test/ops/test_rsqrt.py +++ b/backends/arm/test/ops/test_rsqrt.py @@ -85,24 +85,24 @@ def test_rsqrt_u85_INT(test_tensor: torch.Tensor): @common.parametrize("test_tensor", Rsqrt.test_parameters) @common.SkipIfNoModelConverter -def test_rsqrt_vgf_FP(test_tensor: torch.Tensor): +def test_rsqrt_vgf_no_quant(test_tensor: torch.Tensor): pipeline = VgfPipeline[input_t1]( Rsqrt(), test_tensor(), aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_tensor", Rsqrt.test_parameters) @common.SkipIfNoModelConverter -def test_rsqrt_vgf_INT(test_tensor: torch.Tensor): +def test_rsqrt_vgf_quant(test_tensor: torch.Tensor): pipeline = VgfPipeline[input_t1]( Rsqrt(), test_tensor(), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_rsub.py b/backends/arm/test/ops/test_rsub.py index 3dcab25efe9..1872521c1d7 100644 --- a/backends/arm/test/ops/test_rsub.py +++ b/backends/arm/test/ops/test_rsub.py @@ -98,29 +98,29 @@ def test_rsub_scalar_u85_INT(test_data): @common.parametrize("test_data", rsub_test_data) @common.SkipIfNoModelConverter -def test_rsub_scalar_vgf_FP(test_data: Tuple[torch.Tensor]): +def test_rsub_scalar_vgf_no_quant(test_data: Tuple[torch.Tensor]): """Test Subtraction (VGF FP)""" pipeline = VgfPipeline[input_t1]( Rsub(), test_data(), Rsub.aten_op, Rsub.exir_op, - tosa_version="TOSA-1.0+FP", use_to_edge_transform_and_lower=False, + quantize=False, ) pipeline.run() @common.parametrize("test_data", rsub_test_data) @common.SkipIfNoModelConverter -def test_rsub_scalar_vgf_INT(test_data: Tuple[torch.Tensor]): +def test_rsub_scalar_vgf_quant(test_data: Tuple[torch.Tensor]): """Test Subtraction (VGF INT)""" pipeline = VgfPipeline[input_t1]( Rsub(), test_data(), aten_op="torch.ops.aten.sub.Tensor", exir_op=Rsub.exir_op, - tosa_version="TOSA-1.0+INT", use_to_edge_transform_and_lower=False, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_scalar_tensor.py b/backends/arm/test/ops/test_scalar_tensor.py index 356bcf508b7..bc265077f58 100644 --- a/backends/arm/test/ops/test_scalar_tensor.py +++ b/backends/arm/test/ops/test_scalar_tensor.py @@ -104,13 +104,13 @@ def test_scalar_tensor_u85_INT(test_data): @common.parametrize("test_data", float_test_data_suite) @common.SkipIfNoModelConverter -def test_scalar_tensor_vgf_FP(test_data): +def test_scalar_tensor_vgf_no_quant(test_data): scalar, dtype, data = test_data() pipeline = VgfPipeline( ScalarTensor(scalar, dtype), tuple(data), ScalarTensor.aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -120,13 +120,13 @@ def test_scalar_tensor_vgf_FP(test_data): int_test_data_suite, ) @common.SkipIfNoModelConverter -def test_scalar_tensor_vgf_INT(test_data): +def test_scalar_tensor_vgf_quant(test_data): scalar, dtype, data = test_data() pipeline = VgfPipeline( ScalarTensor(scalar, dtype), tuple(data), ScalarTensor.aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) # Pop the quantization check stage if it exists as no # quantization nodes will be present for int + fp inputs. diff --git a/backends/arm/test/ops/test_sdpa.py b/backends/arm/test/ops/test_sdpa.py index 009e4b2ad70..201d80acaf1 100644 --- a/backends/arm/test/ops/test_sdpa.py +++ b/backends/arm/test/ops/test_sdpa.py @@ -48,22 +48,26 @@ def test_sdpa_tosa_INT(): @common.SkipIfNoModelConverter -def test_sdpa_vgf_FP(): +def test_sdpa_vgf_no_quant(): test_input = tuple(torch.randn(1, 3, 197, 64) for _ in range(3)) pipeline = VgfPipeline[input_t]( - SDPA(), test_input, [], [], tosa_version="TOSA-1.0+FP" + SDPA(), + test_input, + [], + [], + quantize=False, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sdpa_vgf_INT(): +def test_sdpa_vgf_quant(): test_input = tuple(torch.randn(1, 3, 197, 64) for _ in range(3)) pipeline = VgfPipeline[input_t]( SDPA(), test_input, [], [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_select.py b/backends/arm/test/ops/test_select.py index 23046c34fe4..6a82300f252 100644 --- a/backends/arm/test/ops/test_select.py +++ b/backends/arm/test/ops/test_select.py @@ -169,43 +169,51 @@ def test_select_int_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_select_int_vgf_FP_copy(test_data: Tuple): +def test_select_int_copy_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( - SelectCopy(), test_data(), aten_op_copy, [], tosa_version="TOSA-1.0+FP" + SelectCopy(), + test_data(), + aten_op_copy, + [], + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_select_int_vgf_FP(test_data: Tuple): +def test_select_int_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( - SelectInt(), test_data(), aten_op_int, [], tosa_version="TOSA-1.0+FP" + SelectInt(), + test_data(), + aten_op_int, + [], + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_select_int_vgf_INT_copy(test_data: Tuple): +def test_select_int_copy_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SelectCopy(), test_data(), aten_op_copy, [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_select_int_vgf_INT(test_data: Tuple): +def test_select_int_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SelectInt(), test_data(), aten_op_int, [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_select_scatter.py b/backends/arm/test/ops/test_select_scatter.py index 94bfc518b22..b4df8d4ab9d 100644 --- a/backends/arm/test/ops/test_select_scatter.py +++ b/backends/arm/test/ops/test_select_scatter.py @@ -149,25 +149,25 @@ def test_select_scatter_u85_INT(test_module: input_t): @common.SkipIfNoModelConverter @common.parametrize("test_module", test_data_suite) -def test_select_scatter_vgf_FP(test_module: input_t): +def test_select_scatter_vgf_no_quant(test_module: input_t): pipeline = VgfPipeline[input_t]( SelectScatter(), test_module(), aten_op=SelectScatter.fp_aten_op, exir_op=SelectScatter.fp_exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.SkipIfNoModelConverter @common.parametrize("test_module", test_data_suite) -def test_select_scatter_vgf_INT(test_module: input_t): +def test_select_scatter_vgf_quant(test_module: input_t): pipeline = VgfPipeline[input_t]( SelectScatter(), test_module(), aten_op=SelectScatter.int_aten_ops, exir_op=SelectScatter.int_exir_ops, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sigmoid.py b/backends/arm/test/ops/test_sigmoid.py index a9e3802f75b..bac6e376cee 100644 --- a/backends/arm/test/ops/test_sigmoid.py +++ b/backends/arm/test/ops/test_sigmoid.py @@ -167,98 +167,98 @@ def test_sigmoid_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sigmoid_vgf_FP(test_data: Tuple): +def test_sigmoid_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Sigmoid(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sigmoid_vgf_INT(test_data: Tuple): +def test_sigmoid_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Sigmoid(), (test_data(),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sigmoid_vgf_FP_add(): +def test_sigmoid_add_vgf_no_quant(): pipeline = VgfPipeline[input_t1]( AddSigmoid(), (test_data_suite["zeros"](),), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sigmoid_vgf_INT_add(): +def test_sigmoid_add_vgf_quant(): pipeline = VgfPipeline[input_t1]( AddSigmoid(), (test_data_suite["ramp"](),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sigmoid_vgf_FP_add_2(): +def test_sigmoid_add_2_vgf_no_quant(): pipeline = VgfPipeline[input_t1]( SigmoidAdd(), (test_data_suite["zeros"](),), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sigmoid_vgf_INT_add_2(): +def test_sigmoid_add_2_vgf_quant(): pipeline = VgfPipeline[input_t1]( SigmoidAdd(), (test_data_suite["zeros"](),), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sigmoid_vgf_FP_add_3(): +def test_sigmoid_add_3_vgf_no_quant(): pipeline = VgfPipeline[input_t1]( SigmoidAddSigmoid(), (test_data_suite["randn_neg"](), test_data_suite["randn_pos"]()), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.SkipIfNoModelConverter -def test_sigmoid_vgf_INT_add_3(): +def test_sigmoid_add_3_vgf_quant(): pipeline = VgfPipeline[input_t1]( SigmoidAddSigmoid(), (test_data_suite["randn_neg"](), test_data_suite["randn_pos"]()), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sign.py b/backends/arm/test/ops/test_sign.py index 35ea9fc3e45..dd4f28981de 100644 --- a/backends/arm/test/ops/test_sign.py +++ b/backends/arm/test/ops/test_sign.py @@ -89,25 +89,25 @@ def test_sign_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sign_vgf_FP(test_data: Tuple): +def test_sign_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Sign(), (test_data,), aten_op=aten_op, exir_op=exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sign_vgf_INT(test_data: Tuple): +def test_sign_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Sign(), (test_data,), aten_op=[], exir_op=exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_silu.py b/backends/arm/test/ops/test_silu.py index 362358d0813..62e960b750e 100644 --- a/backends/arm/test/ops/test_silu.py +++ b/backends/arm/test/ops/test_silu.py @@ -124,45 +124,51 @@ def test_silu_u85_INT_inplace(test_data: input_t): @common.parametrize("test_data", Silu.test_data) @common.SkipIfNoModelConverter -def test_silu_vgf_FP(test_data: input_t): +def test_silu_vgf_no_quant(test_data: input_t): silu_data = (test_data(), False) pipeline = VgfPipeline[input_t]( - Silu(), silu_data, Silu.aten_op_FP, tosa_version="TOSA-1.0+FP" + Silu(), + silu_data, + Silu.aten_op_FP, + quantize=False, ) pipeline.run() @common.parametrize("test_data", Silu.test_data) @common.SkipIfNoModelConverter -def test_silu_vgf_FP_inplace(test_data: input_t): +def test_silu_inplace_vgf_no_quant(test_data: input_t): silu_data = (test_data(), True) pipeline = VgfPipeline[input_t]( - Silu(), silu_data, Silu.aten_op_inplace_FP, tosa_version="TOSA-1.0+FP" + Silu(), + silu_data, + Silu.aten_op_inplace_FP, + quantize=False, ) pipeline.run() @common.parametrize("test_data", Silu.test_data) @common.SkipIfNoModelConverter -def test_silu_vgf_INT(test_data: input_t): +def test_silu_vgf_quant(test_data: input_t): silu_data = (test_data(), False) pipeline = VgfPipeline[input_t]( Silu(), silu_data, Silu.aten_op_INT, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", Silu.test_data) @common.SkipIfNoModelConverter -def test_silu_vgf_INT_inplace(test_data: input_t): +def test_silu_inplace_vgf_quant(test_data: input_t): silu_data = (test_data(), True) pipeline = VgfPipeline[input_t]( Silu(), silu_data, Silu.aten_op_INT, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sin.py b/backends/arm/test/ops/test_sin.py index 06d06e3b11d..05cc8f5534d 100644 --- a/backends/arm/test/ops/test_sin.py +++ b/backends/arm/test/ops/test_sin.py @@ -86,20 +86,23 @@ def test_sin_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sin_vgf_FP(test_data: Tuple): +def test_sin_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( - Sin(), (test_data,), aten_op, tosa_version="TOSA-1.0+FP" + Sin(), + (test_data,), + aten_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sin_vgf_INT(test_data: Tuple): +def test_sin_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Sin(), (test_data,), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sinh.py b/backends/arm/test/ops/test_sinh.py index a059ce0ad26..703d3e52011 100644 --- a/backends/arm/test/ops/test_sinh.py +++ b/backends/arm/test/ops/test_sinh.py @@ -81,20 +81,23 @@ def test_sinh_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sinh_vgf_FP(test_data: Tuple): +def test_sinh_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( - Sinh(), (test_data,), aten_op, tosa_version="TOSA-1.0+FP" + Sinh(), + (test_data,), + aten_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_sinh_vgf_INT(test_data: Tuple): +def test_sinh_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Sinh(), (test_data,), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_slice.py b/backends/arm/test/ops/test_slice.py index c3de789689e..ab5bafdef32 100644 --- a/backends/arm/test/ops/test_slice.py +++ b/backends/arm/test/ops/test_slice.py @@ -102,26 +102,26 @@ def test_slice_tensor_u85_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_slice_tensor_vgf_FP(test_data: torch.Tensor): +def test_slice_tensor_vgf_no_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Slice(), test_data(), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_slice_tensor_vgf_INT(test_data: torch.Tensor): +def test_slice_tensor_vgf_quant(test_data: torch.Tensor): pipeline = VgfPipeline[input_t1]( Slice(), test_data(), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_softmax.py b/backends/arm/test/ops/test_softmax.py index 22bd919fccd..0b2af23d10c 100644 --- a/backends/arm/test/ops/test_softmax.py +++ b/backends/arm/test/ops/test_softmax.py @@ -91,13 +91,13 @@ def test_softmax_u85_INT(test_data): @common.parametrize("test_data", Softmax.test_data) @common.SkipIfNoModelConverter -def test_softmax_vgf_FP(test_data): +def test_softmax_vgf_no_quant(test_data): data, dim = test_data() pipeline = VgfPipeline[input_t1]( Softmax(dim), data, [], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.add_stage_after( "to_edge_transform_and_lower", pipeline.tester.check_not, [exir_op] @@ -107,13 +107,13 @@ def test_softmax_vgf_FP(test_data): @common.parametrize("test_data", Softmax.test_data) @common.SkipIfNoModelConverter -def test_softmax_vgf_INT(test_data): +def test_softmax_vgf_quant(test_data): data, dim = test_data() pipeline = VgfPipeline[input_t1]( Softmax(dim), data, [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.add_stage_after("quantize", pipeline.tester.check_not, [aten_op]) # TODO: MLETORCH-1136 Change args of run_method_and_compare_outputs of the vgf tests diff --git a/backends/arm/test/ops/test_split.py b/backends/arm/test/ops/test_split.py index afe83b04466..a1028fd07ef 100644 --- a/backends/arm/test/ops/test_split.py +++ b/backends/arm/test/ops/test_split.py @@ -168,26 +168,26 @@ def test_split_with_sizes_u85_INT(test_data: input_t1): (Split.test_data | Split.test_data_list), ) @common.SkipIfNoModelConverter -def test_split_with_sizes_vgf_FP(test_data: input_t1): +def test_split_with_sizes_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Split(), test_data(), aten_op=[], exir_op=exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Split.test_data_list) @common.SkipIfNoModelConverter -def test_split_with_sizes_vgf_FP_2(test_data: input_t1): +def test_split_with_sizes_2_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( SplitWithSizes(), test_data(), aten_op=[], exir_op=exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -197,13 +197,13 @@ def test_split_with_sizes_vgf_FP_2(test_data: input_t1): (Split.test_data | Split.test_data_list), ) @common.SkipIfNoModelConverter -def test_split_with_sizes_vgf_FP_one_out(test_data: input_t1): +def test_split_with_sizes_one_out_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( SplitSingleOut(), test_data(), aten_op=[], exir_op=exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -213,13 +213,13 @@ def test_split_with_sizes_vgf_FP_one_out(test_data: input_t1): (Split.test_data | Split.test_data_list), ) @common.SkipIfNoModelConverter -def test_split_with_sizes_vgf_FP_two_out(test_data: input_t1): +def test_split_with_sizes_two_out_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( SplitTwoOut(), test_data(), aten_op=[], exir_op=exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @@ -229,13 +229,13 @@ def test_split_with_sizes_vgf_FP_two_out(test_data: input_t1): (Split.test_data | Split.test_data_list), ) @common.SkipIfNoModelConverter -def test_split_with_sizes_vgf_INT(test_data: input_t1): +def test_split_with_sizes_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Split(), test_data(), aten_op=[], exir_op=exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -288,25 +288,25 @@ def test_split_tensor_u85_INT(test_data: Tuple): @common.parametrize("test_data", Split.test_data) @common.SkipIfNoModelConverter -def test_split_tensor_vgf_FP(test_data: Tuple): +def test_split_tensor_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SplitCopy(), test_data(), aten_op=SplitCopy.aten_op, exir_op=SplitCopy.exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Split.test_data) @common.SkipIfNoModelConverter -def test_split_tensor_vgf_INT(test_data: Tuple): +def test_split_tensor_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SplitCopy(), test_data(), aten_op=SplitCopy.aten_op, exir_op=SplitCopy.exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sqrt.py b/backends/arm/test/ops/test_sqrt.py index 13a2366b17c..c3d1aae0883 100644 --- a/backends/arm/test/ops/test_sqrt.py +++ b/backends/arm/test/ops/test_sqrt.py @@ -88,25 +88,25 @@ def test_sqrt_u85_INT(test_data: Sqrt.input_t): @common.parametrize("test_data", Sqrt.test_data) @common.SkipIfNoModelConverter -def test_sqrt_vgf_FP(test_data: Sqrt.input_t): +def test_sqrt_vgf_no_quant(test_data: Sqrt.input_t): pipeline = VgfPipeline[Sqrt.input_t]( Sqrt(), test_data(), Sqrt.aten_op_FP, Sqrt.exir_op_FP, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Sqrt.test_data) @common.SkipIfNoModelConverter -def test_sqrt_vgf_INT(test_data: Sqrt.input_t): +def test_sqrt_vgf_quant(test_data: Sqrt.input_t): pipeline = VgfPipeline[Sqrt.input_t]( Sqrt(), test_data(), Sqrt.aten_op_INT, Sqrt.exir_op_INT, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_squeeze.py b/backends/arm/test/ops/test_squeeze.py index 3c2014cdcda..696c677b057 100644 --- a/backends/arm/test/ops/test_squeeze.py +++ b/backends/arm/test/ops/test_squeeze.py @@ -113,26 +113,26 @@ def test_squeeze_dim_u85_INT(test_data: Tuple): @common.parametrize("test_data", Squeeze.test_parameters) @common.SkipIfNoModelConverter -def test_squeeze_dim_vgf_FP(test_data: Tuple): +def test_squeeze_dim_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Squeeze(), test_data(), "torch.ops.aten.squeeze.default", [], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Squeeze.test_parameters) @common.SkipIfNoModelConverter -def test_squeeze_dim_vgf_INT(test_data: Tuple): +def test_squeeze_dim_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Squeeze(), test_data(), "torch.ops.aten.squeeze.default", [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -190,26 +190,26 @@ def test_squeeze_dim_u85_INT_2(test_data: Tuple): @common.parametrize("test_data", SqueezeDim.test_parameters) @common.SkipIfNoModelConverter -def test_squeeze_dim_vgf_FP_2(test_data: Tuple): +def test_squeeze_dim_2_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SqueezeDim(), test_data(), "torch.ops.aten.squeeze.dim", [], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", SqueezeDim.test_parameters) @common.SkipIfNoModelConverter -def test_squeeze_dim_vgf_INT_2(test_data: Tuple): +def test_squeeze_dim_2_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SqueezeDim(), test_data(), "torch.ops.aten.squeeze.dim", [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -267,25 +267,25 @@ def test_squeeze_dims_u85_INT(test_data: Tuple): @common.parametrize("test_data", SqueezeDims.test_parameters) @common.SkipIfNoModelConverter -def test_squeeze_dims_vgf_FP(test_data: Tuple): +def test_squeeze_dims_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SqueezeDims(), test_data(), "torch.ops.aten.squeeze.dims", [], - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", SqueezeDims.test_parameters) @common.SkipIfNoModelConverter -def test_squeeze_dims_vgf_INT(test_data: Tuple): +def test_squeeze_dims_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( SqueezeDims(), test_data(), "torch.ops.aten.squeeze.dims", [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_stack.py b/backends/arm/test/ops/test_stack.py index 873a599992a..a3911a62b01 100644 --- a/backends/arm/test/ops/test_stack.py +++ b/backends/arm/test/ops/test_stack.py @@ -122,29 +122,29 @@ def test_stack_u85_INT(test_module: input_t1): @common.SkipIfNoModelConverter @common.parametrize("test_module", test_data_suite) -def test_stack_vgf_FP(test_module: input_t1): +def test_stack_vgf_no_quant(test_module: input_t1): test_data = test_module() pipeline = VgfPipeline[input_t1]( Stack(), test_data, aten_op=Stack.aten_op, exir_op=Stack.exir_op, - tosa_version="TOSA-1.0+FP", use_to_edge_transform_and_lower=False, + quantize=False, ) pipeline.run() @common.SkipIfNoModelConverter @common.parametrize("test_module", test_data_suite) -def test_stack_vgf_INT(test_module: input_t1): +def test_stack_vgf_quant(test_module: input_t1): test_data = test_module() pipeline = VgfPipeline[input_t1]( Stack(), test_data, aten_op=Stack.aten_op, exir_op=Stack.exir_op, - tosa_version="TOSA-1.0+INT", use_to_edge_transform_and_lower=False, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sub.py b/backends/arm/test/ops/test_sub.py index b01e277fd3b..f18f6525d27 100644 --- a/backends/arm/test/ops/test_sub.py +++ b/backends/arm/test/ops/test_sub.py @@ -219,56 +219,56 @@ def test_sub_tensor_u85_INT(test_data: Tuple[torch.Tensor, torch.Tensor]): @common.parametrize("test_data", sub_test_data) @common.SkipIfNoModelConverter -def test_sub_tensor_vgf_FP(test_data: Tuple[torch.Tensor]): +def test_sub_tensor_vgf_no_quant(test_data: Tuple[torch.Tensor]): """Test Subtraction (VGF FP)""" pipeline = VgfPipeline[input_t1]( Sub(), test_data(), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", sub2_test_data) @common.SkipIfNoModelConverter -def test_sub_tensor_vgf_FP_2(test_data: Tuple[torch.Tensor, torch.Tensor]): +def test_sub_tensor_2_vgf_no_quant(test_data: Tuple[torch.Tensor, torch.Tensor]): """Test Two-Operand Subtraction (VGF FP)""" pipeline = VgfPipeline[input_t2]( Sub2(), test_data(), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", sub_test_data) @common.SkipIfNoModelConverter -def test_sub_tensor_vgf_INT(test_data: Tuple[torch.Tensor]): +def test_sub_tensor_vgf_quant(test_data: Tuple[torch.Tensor]): """Test Subtraction (VGF INT)""" pipeline = VgfPipeline[input_t1]( Sub(), test_data(), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @common.parametrize("test_data", sub2_test_data) @common.SkipIfNoModelConverter -def test_sub_tensor_vgf_INT_2(test_data: Tuple[torch.Tensor, torch.Tensor]): +def test_sub_tensor_2_vgf_quant(test_data: Tuple[torch.Tensor, torch.Tensor]): """Test Two-Operand Subtraction (VGF INT)""" pipeline = VgfPipeline[input_t2]( Sub2(), test_data(), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_sum.py b/backends/arm/test/ops/test_sum.py index 050a50e7251..14a6eeea14c 100644 --- a/backends/arm/test/ops/test_sum.py +++ b/backends/arm/test/ops/test_sum.py @@ -90,26 +90,26 @@ def test_view_u85_INT_1_0(test_data: Tuple): @common.parametrize("test_data", Sum.test_parameters) @common.SkipIfNoModelConverter -def test_sum_dim_intlist_vgf_FP(test_data: input_t1): +def test_sum_dim_intlist_vgf_no_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Sum(), test_data(), aten_op, - tosa_version="TOSA-1.0+FP", run_on_vulkan_runtime=True, + quantize=False, ) pipeline.run() @common.parametrize("test_data", Sum.test_parameters) @common.SkipIfNoModelConverter -def test_sum_dim_intlist_vgf_INT(test_data: input_t1): +def test_sum_dim_intlist_vgf_quant(test_data: input_t1): pipeline = VgfPipeline[input_t1]( Sum(), test_data(), aten_op, - tosa_version="TOSA-1.0+INT", run_on_vulkan_runtime=True, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_t_copy.py b/backends/arm/test/ops/test_t_copy.py index 2b254f2c922..705e812cd6d 100644 --- a/backends/arm/test/ops/test_t_copy.py +++ b/backends/arm/test/ops/test_t_copy.py @@ -89,27 +89,27 @@ def test_t_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_t_vgf_FP(test_data: Tuple): +def test_t_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( TCopy(), test_data(), aten_op=TCopy.aten_op, exir_op=TCopy.exir_op, - tosa_version="TOSA-1.0+FP", use_to_edge_transform_and_lower=False, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_t_vgf_INT(test_data: Tuple): +def test_t_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( TCopy(), test_data(), aten_op=TCopy.aten_op, exir_op=TCopy.exir_op, - tosa_version="TOSA-1.0+INT", use_to_edge_transform_and_lower=False, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_tanh.py b/backends/arm/test/ops/test_tanh.py index 0e4e96ca18e..d03fe03452b 100644 --- a/backends/arm/test/ops/test_tanh.py +++ b/backends/arm/test/ops/test_tanh.py @@ -89,21 +89,24 @@ def test_tanh_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_tanh_vgf_FP(test_data: Tuple): +def test_tanh_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( - Tanh(), (test_data(),), aten_op, tosa_version="TOSA-1.0+FP" + Tanh(), + (test_data(),), + aten_op, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_tanh_vgf_INT(test_data: Tuple): +def test_tanh_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( Tanh(), (test_data(),), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_to_copy.py b/backends/arm/test/ops/test_to_copy.py index b3b7fab5318..17db2c3f226 100644 --- a/backends/arm/test/ops/test_to_copy.py +++ b/backends/arm/test/ops/test_to_copy.py @@ -95,24 +95,15 @@ def test_to_tosa_FP(test_data: Tuple): @common.parametrize("test_data", _TO_COPY_TEST_DATA_FP) @common.SkipIfNoModelConverter -def test_to_vgf_FP(test_data: Tuple): +def test_to_vgf_no_quant(test_data: Tuple): test_tensor, new_dtype = test_data() pipeline = VgfPipeline[input_t1]( Cast(new_dtype), (test_tensor,), aten_op=[], exir_op=[], - tosa_version="TOSA-1.0+FP", + quantize=False, ) - # int to int cast is not supported in TOSA+FP profile - if not new_dtype.is_floating_point and not torch.is_floating_point(test_tensor): - pipeline.change_args( - "check_count.exir", - { - "torch.ops.higher_order.executorch_call_delegate": 0, - "executorch_exir_dialects_edge__ops_dim_order_ops__to_dim_order_copy_default": 1, - }, - ) pipeline.run() @@ -164,7 +155,7 @@ def test_to_tosa_INT_not_delegated(test_data: Tuple): @common.parametrize("test_data", _TO_COPY_TEST_DATA_INT) @common.SkipIfNoModelConverter -def test_to_vgf_INT(test_data: Tuple): +def test_to_vgf_quant(test_data: Tuple): # Op not supported pass diff --git a/backends/arm/test/ops/test_transpose_copy.py b/backends/arm/test/ops/test_transpose_copy.py index 58016020466..fb521eda1db 100644 --- a/backends/arm/test/ops/test_transpose_copy.py +++ b/backends/arm/test/ops/test_transpose_copy.py @@ -88,27 +88,27 @@ def test_transpose_int_u85_INT(test_data: Tuple): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_transpose_int_vgf_FP(test_data: Tuple): +def test_transpose_int_vgf_no_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( TransposeCopy(), test_data(), aten_op=TransposeCopy.aten_op, exir_op=TransposeCopy.exir_op, - tosa_version="TOSA-1.0+FP", use_to_edge_transform_and_lower=False, + quantize=False, ) pipeline.run() @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_transpose_int_vgf_INT(test_data: Tuple): +def test_transpose_int_vgf_quant(test_data: Tuple): pipeline = VgfPipeline[input_t1]( TransposeCopy(), test_data(), aten_op=TransposeCopy.aten_op, exir_op=TransposeCopy.exir_op, - tosa_version="TOSA-1.0+INT", use_to_edge_transform_and_lower=False, + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_unary_combos.py b/backends/arm/test/ops/test_unary_combos.py index bfeb9b59e80..312350ea8d3 100644 --- a/backends/arm/test/ops/test_unary_combos.py +++ b/backends/arm/test/ops/test_unary_combos.py @@ -132,9 +132,13 @@ def test_unary_combos_u85_INT(model_cls): @common.SkipIfNoModelConverter @pytest.mark.parametrize("model_cls", MODELS, ids=lambda c: c.__name__) -def test_unary_combos_vgf_INT(model_cls): +def test_unary_combos_vgf_quant(model_cls): m, inputs, exir = _build(model_cls) p = VgfPipeline[Tensor1]( - m, inputs, aten_op=[], exir_op=exir, tosa_version="TOSA-1.0+INT" + m, + inputs, + aten_op=[], + exir_op=exir, + quantize=True, ) p.run() diff --git a/backends/arm/test/ops/test_unbind.py b/backends/arm/test/ops/test_unbind.py index cd33f8217df..ce3f769cd06 100644 --- a/backends/arm/test/ops/test_unbind.py +++ b/backends/arm/test/ops/test_unbind.py @@ -58,25 +58,25 @@ def test_unbind_int_tosa_INT(test_data: test_data_t): @common.parametrize("test_data", Unbind.test_data) @common.SkipIfNoModelConverter -def test_unbind_int_vgf_FP(test_data: test_data_t): +def test_unbind_int_vgf_no_quant(test_data: test_data_t): input_data, init_data = test_data pipeline = VgfPipeline[input_t]( Unbind(*init_data), input_data(), Unbind.aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Unbind.test_data) @common.SkipIfNoModelConverter -def test_unbind_int_vgf_INT(test_data: test_data_t): +def test_unbind_int_vgf_quant(test_data: test_data_t): input_data, init_data = test_data pipeline = VgfPipeline[input_t]( Unbind(*init_data), input_data(), Unbind.aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_unflatten.py b/backends/arm/test/ops/test_unflatten.py index 35b264a4bb0..d4730ac6dc2 100644 --- a/backends/arm/test/ops/test_unflatten.py +++ b/backends/arm/test/ops/test_unflatten.py @@ -83,25 +83,25 @@ def test_unflatten_int_u85_INT(test_data: test_data_t): @common.parametrize("test_data", Unflatten.test_data) @common.SkipIfNoModelConverter -def test_unflatten_int_vgf_FP(test_data: test_data_t): +def test_unflatten_int_vgf_no_quant(test_data: test_data_t): module, inputs = test_data() pipeline = VgfPipeline[input_t]( module, inputs, Unflatten.aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", Unflatten.test_data) @common.SkipIfNoModelConverter -def test_unflatten_int_vgf_INT(test_data: test_data_t): +def test_unflatten_int_vgf_quant(test_data: test_data_t): module, inputs = test_data() pipeline = VgfPipeline[input_t]( module, inputs, Unflatten.aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_unsqueeze.py b/backends/arm/test/ops/test_unsqueeze.py index c76c1236ab3..0c29d3b588c 100644 --- a/backends/arm/test/ops/test_unsqueeze.py +++ b/backends/arm/test/ops/test_unsqueeze.py @@ -83,22 +83,25 @@ def test_unsqueeze_u85_INT(test_tensor: torch.Tensor): @common.parametrize("test_tensor", Unsqueeze.test_parameters) @common.SkipIfNoModelConverter -def test_unsqueeze_vgf_FP(test_tensor: torch.Tensor): +def test_unsqueeze_vgf_no_quant(test_tensor: torch.Tensor): for i in range(-test_tensor[0].dim() - 1, test_tensor[0].dim() + 1): pipeline = VgfPipeline[input_t1]( - Unsqueeze(), (*test_tensor, i), aten_op, tosa_version="TOSA-1.0+FP" + Unsqueeze(), + (*test_tensor, i), + aten_op, + quantize=False, ) pipeline.run() @common.parametrize("test_tensor", Unsqueeze.test_parameters) @common.SkipIfNoModelConverter -def test_unsqueeze_vgf_INT(test_tensor: torch.Tensor): +def test_unsqueeze_vgf_quant(test_tensor: torch.Tensor): for i in range(-test_tensor[0].dim() - 1, test_tensor[0].dim() + 1): pipeline = VgfPipeline[input_t1]( Unsqueeze(), (*test_tensor, i), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_upsample_bilinear2d.py b/backends/arm/test/ops/test_upsample_bilinear2d.py index db440fcb3d4..edac736981a 100644 --- a/backends/arm/test/ops/test_upsample_bilinear2d.py +++ b/backends/arm/test/ops/test_upsample_bilinear2d.py @@ -345,14 +345,16 @@ def test_upsample_bilinear2d_vec_U85_INT_a16w8( @common.parametrize("test_data", test_data_suite_tosa) @common.SkipIfNoModelConverter -def test_upsample_bilinear2d_vgf_FP_UpsamplingBilinear2d(test_data: torch.Tensor): +def test_upsample_bilinear2d_UpsamplingBilinear2d_vgf_no_quant( + test_data: torch.Tensor, +): data, size, scale_factor, compare = test_data pipeline = VgfPipeline[input_t1]( UpsamplingBilinear2d(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) if not compare: pipeline.pop_stage(-1) @@ -361,14 +363,14 @@ def test_upsample_bilinear2d_vgf_FP_UpsamplingBilinear2d(test_data: torch.Tensor @common.parametrize("test_data", test_data_suite_tosa) @common.SkipIfNoModelConverter -def test_upsample_bilinear2d_vgf_FP_Upsample(test_data: torch.Tensor): +def test_upsample_bilinear2d_Upsample_vgf_no_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data pipeline = VgfPipeline[input_t1]( Upsample(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) if not compare: pipeline.pop_stage(-1) @@ -377,14 +379,14 @@ def test_upsample_bilinear2d_vgf_FP_Upsample(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite_tosa) @common.SkipIfNoModelConverter -def test_upsample_bilinear2d_vgf_FP_Interpolate(test_data: torch.Tensor): +def test_upsample_bilinear2d_Interpolate_vgf_no_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data pipeline = VgfPipeline[input_t1]( Interpolate(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) if not compare: pipeline.pop_stage(-1) @@ -393,14 +395,16 @@ def test_upsample_bilinear2d_vgf_FP_Interpolate(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite_tosa) @common.SkipIfNoModelConverter -def test_upsample_bilinear2d_vgf_INT_UpsamplingBilinear2d(test_data: torch.Tensor): +def test_upsample_bilinear2d_UpsamplingBilinear2d_vgf_quant( + test_data: torch.Tensor, +): data, size, scale_factor, compare = test_data pipeline = VgfPipeline[input_t1]( UpsamplingBilinear2d(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) if not compare: pipeline.pop_stage(-1) @@ -409,14 +413,14 @@ def test_upsample_bilinear2d_vgf_INT_UpsamplingBilinear2d(test_data: torch.Tenso @common.parametrize("test_data", test_data_suite_tosa) @common.SkipIfNoModelConverter -def test_upsample_bilinear2d_vgf_INT_Upsample(test_data: torch.Tensor): +def test_upsample_bilinear2d_Upsample_vgf_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data pipeline = VgfPipeline[input_t1]( Upsample(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) if not compare: pipeline.pop_stage(-1) @@ -425,14 +429,14 @@ def test_upsample_bilinear2d_vgf_INT_Upsample(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite_tosa) @common.SkipIfNoModelConverter -def test_upsample_bilinear2d_vgf_INT_Interpolate(test_data: torch.Tensor): +def test_upsample_bilinear2d_Interpolate_vgf_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data pipeline = VgfPipeline[input_t1]( Interpolate(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) if not compare: pipeline.pop_stage(-1) diff --git a/backends/arm/test/ops/test_upsample_nearest2d.py b/backends/arm/test/ops/test_upsample_nearest2d.py index e7da0643d0e..5da590398f4 100644 --- a/backends/arm/test/ops/test_upsample_nearest2d.py +++ b/backends/arm/test/ops/test_upsample_nearest2d.py @@ -213,14 +213,14 @@ def test_upsample_nearest2d_vec_tosa_INT_a16w8(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_upsample_nearest2d_vgf_FP(test_data: torch.Tensor): +def test_upsample_nearest2d_vgf_no_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data() pipeline = VgfPipeline[input_t1]( UpsamplingNearest2d(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) if not compare: pipeline.pop_stage(-1) @@ -229,14 +229,14 @@ def test_upsample_nearest2d_vgf_FP(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_upsample_nearest2d_vgf_FP_nearest(test_data: torch.Tensor): +def test_upsample_nearest2d_nearest_vgf_no_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data() pipeline = VgfPipeline[input_t1]( Upsample(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) if not compare: pipeline.pop_stage(-1) @@ -245,13 +245,15 @@ def test_upsample_nearest2d_vgf_FP_nearest(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_upsample_nearest2d_vgf_FP_interpolate(test_data: torch.Tensor): +def test_upsample_nearest2d_interpolate_vgf_FP(test_data: torch.Tensor): data, size, scale_factor, compare = test_data() pipeline = VgfPipeline[input_t1]( Interpolate(size, scale_factor), (data,), aten_op, exir_op, + quantize=False, + # Override tosa version to test FP-only path tosa_version="TOSA-1.0+FP", ) if not compare: @@ -261,14 +263,14 @@ def test_upsample_nearest2d_vgf_FP_interpolate(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_upsample_nearest2d_vgf_INT(test_data: torch.Tensor): +def test_upsample_nearest2d_vgf_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data() pipeline = VgfPipeline[input_t1]( UpsamplingNearest2d(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) if not compare: pipeline.pop_stage(-1) @@ -277,14 +279,14 @@ def test_upsample_nearest2d_vgf_INT(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_upsample_nearest2d_vgf_INT_nearest(test_data: torch.Tensor): +def test_upsample_nearest2d_nearest_vgf_quant(test_data: torch.Tensor): data, size, scale_factor, compare = test_data() pipeline = VgfPipeline[input_t1]( Upsample(size, scale_factor), (data,), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) if not compare: pipeline.pop_stage(-1) @@ -293,13 +295,15 @@ def test_upsample_nearest2d_vgf_INT_nearest(test_data: torch.Tensor): @common.parametrize("test_data", test_data_suite) @common.SkipIfNoModelConverter -def test_upsample_nearest2d_vgf_INT_interpolate(test_data: torch.Tensor): +def test_upsample_nearest2d_interpolate_vgf_INT(test_data: torch.Tensor): data, size, scale_factor, compare = test_data() pipeline = VgfPipeline[input_t1]( Interpolate(size, scale_factor), (data,), aten_op, exir_op, + quantize=True, + # Override tosa version to test INT-only path tosa_version="TOSA-1.0+INT", ) if not compare: diff --git a/backends/arm/test/ops/test_var.py b/backends/arm/test/ops/test_var.py index 282c3a4455d..73bf2165b23 100644 --- a/backends/arm/test/ops/test_var.py +++ b/backends/arm/test/ops/test_var.py @@ -213,24 +213,28 @@ def test_var_dim_u85_INT_no_dim(test_data: Tuple): @common.parametrize("test_data", Var.test_parameters) @common.SkipIfNoModelConverter -def test_var_dim_vgf_FP_no_dim(test_data: Tuple): +def test_var_dim_no_dim_vgf_no_quant(test_data: Tuple): data, keepdim, correction = test_data() pipeline = VgfPipeline[input_t1]( - Var(keepdim, correction), (data,), [], [], tosa_version="TOSA-1.0+FP" + Var(keepdim, correction), + (data,), + [], + [], + quantize=False, ) pipeline.run() @common.parametrize("test_data", Var.test_parameters) @common.SkipIfNoModelConverter -def test_var_dim_vgf_INT_no_dim(test_data: Tuple): +def test_var_dim_no_dim_vgf_quant(test_data: Tuple): data, keepdim, correction = test_data() pipeline = VgfPipeline[input_t1]( Var(keepdim, correction), (data,), [], [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -293,24 +297,28 @@ def test_var_dim_u85_INT(test_data: Tuple): @common.parametrize("test_data", VarDim.test_parameters) @common.SkipIfNoModelConverter -def test_var_dim_vgf_FP(test_data: Tuple): +def test_var_dim_vgf_no_quant(test_data: Tuple): data, dim, keepdim, unbiased = test_data() pipeline = VgfPipeline[input_t1]( - VarDim(dim, keepdim, unbiased), (data,), [], [], tosa_version="TOSA-1.0+FP" + VarDim(dim, keepdim, unbiased), + (data,), + [], + [], + quantize=False, ) pipeline.run() @common.parametrize("test_data", VarDim.test_parameters) @common.SkipIfNoModelConverter -def test_var_dim_vgf_INT(test_data: Tuple): +def test_var_dim_vgf_quant(test_data: Tuple): data, dim, keepdim, unbiased = test_data() pipeline = VgfPipeline[input_t1]( VarDim(dim, keepdim, unbiased), (data,), [], [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() @@ -382,23 +390,27 @@ def test_var_dim_u85_INT_correction(test_data: Tuple): @common.parametrize("test_data", VarCorrection.test_parameters) @common.SkipIfNoModelConverter -def test_var_dim_vgf_FP_correction(test_data: Tuple): +def test_var_dim_correction_vgf_no_quant(test_data: Tuple): data, dim, keepdim, corr = test_data() pipeline = VgfPipeline[input_t1]( - VarCorrection(dim, keepdim, corr), (data,), [], [], tosa_version="TOSA-1.0+FP" + VarCorrection(dim, keepdim, corr), + (data,), + [], + [], + quantize=False, ) pipeline.run() @common.parametrize("test_data", VarCorrection.test_parameters) @common.SkipIfNoModelConverter -def test_var_dim_vgf_INT_correction(test_data: Tuple): +def test_var_dim_correction_vgf_quant(test_data: Tuple): data, dim, keepdim, corr = test_data() pipeline = VgfPipeline[input_t1]( VarCorrection(dim, keepdim, corr), (data,), [], [], - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_view.py b/backends/arm/test/ops/test_view.py index 3e706ae1cac..99df4f2f2f7 100644 --- a/backends/arm/test/ops/test_view.py +++ b/backends/arm/test/ops/test_view.py @@ -107,26 +107,26 @@ def test_view_u55_INT(test_data: Tuple): @common.parametrize("test_data", View.needs_transpose_tests) @common.SkipIfNoModelConverter -def test_view_vgf_FP(test_data: Tuple): +def test_view_vgf_no_quant(test_data: Tuple): test_tensor, new_shape = test_data() pipeline = VgfPipeline[input_t1]( View(new_shape), (test_tensor,), aten_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_data", View.needs_transpose_tests) @common.SkipIfNoModelConverter -def test_view_vgf_INT(test_data: Tuple): +def test_view_vgf_quant(test_data: Tuple): test_tensor, new_shape = test_data() pipeline = VgfPipeline[input_t1]( View(new_shape), (test_tensor,), aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_where.py b/backends/arm/test/ops/test_where.py index e05aa2ffa5a..50a7aef657d 100644 --- a/backends/arm/test/ops/test_where.py +++ b/backends/arm/test/ops/test_where.py @@ -261,25 +261,25 @@ def test_where_self_u85_INT(test_module): @common.parametrize("test_module", test_modules_FP) @common.SkipIfNoModelConverter -def test_where_self_vgf_FP(test_module): +def test_where_self_vgf_no_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), aten_op, exir_op, - tosa_version="TOSA-1.0+FP", + quantize=False, ) pipeline.run() @common.parametrize("test_module", test_modules_INT) @common.SkipIfNoModelConverter -def test_where_self_vgf_INT(test_module): +def test_where_self_vgf_quant(test_module): pipeline = VgfPipeline[input_t]( test_module(), test_module().get_inputs(), aten_op, exir_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) pipeline.run() diff --git a/backends/arm/test/ops/test_zeros.py b/backends/arm/test/ops/test_zeros.py index d9a885620d9..7e1609e8976 100644 --- a/backends/arm/test/ops/test_zeros.py +++ b/backends/arm/test/ops/test_zeros.py @@ -127,10 +127,13 @@ def test_zeros_tosa_INT_not_delegated(test_data: test_data_t): ZerosAdd.test_data, ) @common.SkipIfNoModelConverter -def test_zeros_vgf_FP(test_data: test_data_t): +def test_zeros_vgf_no_quant(test_data: test_data_t): input_data, init_data = test_data pipeline = VgfPipeline[input_t]( - ZerosAdd(*init_data), input_data(), ZerosAdd.aten_op, tosa_version="TOSA-1.0+FP" + ZerosAdd(*init_data), + input_data(), + ZerosAdd.aten_op, + quantize=False, ) pipeline.run() @@ -140,13 +143,13 @@ def test_zeros_vgf_FP(test_data: test_data_t): ZerosAdd.test_data, ) @common.SkipIfNoModelConverter -def test_zeros_vgf_INT(test_data: test_data_t): +def test_zeros_vgf_quant(test_data: test_data_t): input_data, init_data = test_data pipeline = VgfPipeline[input_t]( ZerosAdd(*init_data), input_data(), ZerosAdd.aten_op, - tosa_version="TOSA-1.0+INT", + quantize=True, ) # Pop the quantization check stage if it exists as no # quantization nodes will be present for int + fp inputs. diff --git a/backends/arm/test/tester/test_pipeline.py b/backends/arm/test/tester/test_pipeline.py index b09fd9c8c0c..829fc01f893 100644 --- a/backends/arm/test/tester/test_pipeline.py +++ b/backends/arm/test/tester/test_pipeline.py @@ -1088,7 +1088,7 @@ def __init__( transform_passes=transform_passes, ) - if quantize and tosa_spec.support_integer(): + if quantize: quantizer = VgfQuantizer(compile_spec) quantization_config = get_symmetric_quantization_config( is_per_channel=per_channel_quantization