- Sort Score
- Result 10 results
- Languages All
Results 41 - 50 of 80 for Convolution (0.16 sec)
-
tensorflow/compiler/mlir/quantization/stablehlo/cc/pass_pipeline.h
// Processes tensors with NCHW format (== (batch, channel, height, weight)) by // converting them to NHWC formats along with extra optimizations such as // constant folding the transpose->convolution pattern. This is useful when // downstream pipeline (e.g. XLA) is more optimized when accepting NHWC formats. void AddProcessNchwTensorPasses(OpPassManager& pm);
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Mon Apr 15 12:53:33 UTC 2024 - 3.6K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/passes/quantization_patterns.cc
// usually has the following pattern. In the example below, // the input operand would be stablehlo.convolution op, and return value would // be stablehlo.add op. // // ``` // %0 = stablehlo.constant dense<3> // %1 = stablehlo.constant dense<4> // %2 = stablehlo.constant dense<2> // %3 = stablehlo.convolution(%%arg0, %%arg1) : // (tensor<?x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<?x3x4x2xf32>
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri May 03 06:04:36 UTC 2024 - 41.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/lite/transforms/decompose_hybrid_quantization.cc
} // If all quantized or floating point then types are consistent. // Int is valid in combination with both quantized and floating point. // This occurs when doing qi16 convolution, as bias is passed as a // non-quantized int64 if (allTypesFp || allTypesQuantizedOrInt) return failure(); Location loc = op->getLoc(); SmallVector<Value> newOperands;
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 5.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/passes/nchw_convolution_to_nhwc.cc
class NchwConvolutionToNhwcPass : public impl::NchwConvolutionToNhwcPassBase<NchwConvolutionToNhwcPass> { private: void runOnOperation() override; }; // Rewrites NCHW convolution to NHWC. // * Src dimension numbers: [b, f, 0, 1]x[o, i, 0, 1]->[b, f, 0, 1] // * Dst dimension numbers: [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f] class RewriteNchwConvolutionToNhwc
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 8.2K bytes - Viewed (0) -
tensorflow/compiler/mlir/lite/stablehlo/tests/legalize_hlo.mlir
// CHECK: return %[[VAL_14]] : tensor<16x32x256xbf16> // CHECK: } func.func @convert_conv1d(%arg0: tensor<16x32x256xbf16>, %arg1: tensor<1x256x256xbf16>) -> tensor<16x32x256xbf16> { %0 = "mhlo.convolution"(%arg0, %arg1) { batch_group_count = 1 : i64, dimension_numbers = #mhlo.conv<[b, 0, f]x[0, i, o]->[b, 0, f]>, feature_group_count = 1 : i64, lhs_dilation = dense<1> : tensor<1xi64>,
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed May 29 07:26:59 UTC 2024 - 340.2K bytes - Viewed (0) -
tensorflow/compiler/mlir/tfr/examples/mnist/mnist_train.py
'b4': tf.Variable(tf.zeros([num_classes])), } @tf.function def __call__(self, data): """The Model definition.""" x = tf.reshape(data, [-1, 28, 28, 1]) # 2D convolution, with 'SAME' padding (i.e. the output feature map has # the same size as the input). # NOTE: The data/x/input is always specified in floating point precision. # output shape: [-1, 28, 28, 32]
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Oct 20 03:05:18 UTC 2021 - 6.5K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/tests/layout_optimization_layout_assignment_gpu_cc_70.mlir
} // CHECK-LABEL: func @transposeConv2D_1x1_f32 func.func @transposeConv2D_1x1_f32(%input: tensor<1x64x28x28xf32>, %filter: tensor<1x1x64x64xf32>) -> tensor<1x64x28x28xf32> { // 1x1 convolution can be computed as a GEMM in NHWC data format. // CHECK: "tf.Conv2D"(%[[INPUT_TRANSPOSE:[0-9]*]], %arg1) // CHECK-SAME: data_format = "NHWC" %0 = "tf.Conv2D"(%input, %filter) { data_format = "NCHW",
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Jun 21 08:41:18 UTC 2022 - 8.5K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/tests/legalize-tf-quant.mlir
// CHECK: %[[CONST:.*]] = mhlo.constant() // CHECK-SAME{LITERAL} value = dense<127> : tensor<2x3x3x2xi8> // CHECK-SAME: tensor<2x3x3x2x!quant.uniform<i8:f32, 1.000000e+00:3>> // CHECK: mhlo.convolution(%arg0, %[[CONST]]) // CHECK-SAME{LITERAL}: dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f] // CHECK-SAME{LITERAL}: window = {stride = [1, 2], pad = [[0, 0], [0, 0]], lhs_dilate = [1, 1], rhs_dilate = [2, 2]}
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu May 09 01:25:29 UTC 2024 - 37.3K bytes - Viewed (0) -
tensorflow/compiler/mlir/lite/transforms/prepare_tf.cc
return failure(); } // Check if the given op is based on grouped convolution. // Dim size zero will be verified by the tf.Conv2D operator verification. if (input_type.getDimSize(3) % filter_type.getDimSize(2) != 0) { return failure(); } // TensorFlow convolution op only has two inputs, while the TFLite one has
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue May 28 21:49:50 UTC 2024 - 64.6K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/passes/defer_activation_transpose.cc
class DeferActivationTransposeForAddOp : public OpRewritePattern<AddOp> { public: using OpRewritePattern<AddOp>::OpRewritePattern; LogicalResult match(AddOp op) const override { // Only supports the case for 2D convolution. const Value lhs = op.getOperand(0); if (!HasRankOf(lhs, /*rank=*/4)) return failure(); const Value rhs = op.getOperand(1); Operation* rhs_op = rhs.getDefiningOp();
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 11.5K bytes - Viewed (0)