- Sort Score
- Result 10 results
- Languages All
Results 71 - 80 of 178 for dequantize (0.26 sec)
-
tensorflow/compiler/mlir/lite/stablehlo/transforms/uniform_quantized_stablehlo_to_tfl_pass.cc
} }; // stablehlo.uniform_dequantize -> tfl.dequantize class RewriteUniformDequantizeOp : public OpRewritePattern<stablehlo::UniformDequantizeOp> { using OpRewritePattern<stablehlo::UniformDequantizeOp>::OpRewritePattern; // Determines whether the input and output types are compatible with // `tfl.dequantize`. See the definition for the `DEQUANTIZE` kernel for the // detailed limitations
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Mon Apr 22 09:00:19 UTC 2024 - 99.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/tensorflow/passes/quantize_composite_functions.cc
call_op, result_types, args, FlatSymbolRefAttr::get(new_quant_func_name)); return success(); } // For composite functions followed by Dequantize ops, merges the Dequantize // op into the functions by creating quantized functions with float output. LogicalResult mergeDequantizeOpFollowingQuantizedFunction( TF::PartitionedCallOp call_op, const SmallVector<Value, 4>& args,
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 54.5K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/passes/quantization_patterns.cc
} void rewrite(quantfork::DequantizeCastOp op, PatternRewriter& rewriter) const final { // Rewrite the floating-point ops to the quantized version, by fusing // preceding dequantize ops and succeding quantize ops. for (Operation* op_with_region : op.getResult().getUsers()) { // Collect all the quantized inputs and "clone" the matched op by these // inputs. SmallVector<Value, 4> inputs;
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/quantization/stablehlo/passes/bridge/legalize_tf_quant_test.cc
%min_range = "tf.Const"() { value = dense<1.0> : tensor<f32> } : () -> tensor<f32> %max_range = "tf.Const"() { value = dense<5.0> : tensor<f32> } : () -> tensor<f32> %0 = "tf.Dequantize"(%arg0, %min_range, %max_range) : (tensor<1x!tf_type.qint8>, tensor<f32>, tensor<f32>) -> tensor<1xf32> func.return %0 : tensor<1xf32> } })mlir"; std::vector<tensorflow::TensorShape> arg_shapes = {{1}};
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Feb 29 18:43:55 UTC 2024 - 7.2K bytes - Viewed (0) -
tensorflow/compiler/mlir/lite/tests/optimize_batch_matmul.mlir
%0 = arith.constant dense<[[1.0], [2.0]]> : tensor<2x1xf32> %1 = "tfl.quantize"(%0) {qtype = tensor<2x1x!quant.uniform<i8:f32, 0.024986599940879671:92>>} : (tensor<2x1xf32>) -> tensor<2x1x!quant.uniform<i8:f32, 0.024986599940879671:92>> %2 = "tfl.dequantize"(%1) : (tensor<2x1x!quant.uniform<i8:f32, 0.024986599940879671:92>>) -> tensor<2x1xf32>
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu May 02 09:41:17 UTC 2024 - 9K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/tensorflow/ops/tf_quantize_op.cc
func_name, rewriter, quant_type, val_to_dequantize, result_type, LogicsForUniformDequanization); return dequant_op; } } // namespace // Generate quantize and dequantize functions with uniform quantization. std::optional<TF::PartitionedCallOp> ApplyUniformQuantization( PatternRewriter& rewriter, TF::ConstOp op, tensorflow::quantization::QuantizationComponentSpec& weight_spec) {
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 11K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/tensorflow/calibrator/calibration_algorithm.py
find the quant_min and quant_max that best describe this distribution. To do this, we quantize hist_mids using quant_min and quant_max and dequantize them again. Then the difference between hist_mids and dequantized hist_mids equates to quantization error when using quant_min and quant_max. Args:
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Mon Mar 11 19:29:56 UTC 2024 - 14.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/common/quantization_lib/quantization_config.h
// weights but will dequantize them back at runtime which is useful for // memory bound case without kernel support available in lower precisions. // Used in MLIR dynamic range quantizer. bool weight_only_quantization = false; // The minimum number of elements in a weights array required to apply // quantization. This is especially useful not to quantize small tensors as
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Mar 13 10:16:19 UTC 2024 - 10.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/ops/stablehlo_op_quant_spec_test.cc
func.return %add : tensor<3x2xf32> } } )mlir"; // Quantizable ops: XlaCallModule op with "fully_quantizable" attribute and // same-scale StableHLO ops // Non-quantizable ops: quantize/dequantize ops constexpr absl::string_view kModuleCompositeSameScale = R"mlir( module { func.func @same_scale_after_composite() -> tensor<3x1xf32> {
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 04 07:19:09 UTC 2024 - 14.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/quantization/stablehlo/quantization_config.proto
// hardware performs better with integer ops. // Default value: true optional bool unpack_quantized_types = 1; // When set to True, requantize op in the quantized fusion will merge with the // subsequent dequantize op if present. // Default value: false // TODO: b/321729008 - re-consider default value after testing on prod model. bool merge_fusion_with_dequantize = 2; }
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri May 17 03:36:50 UTC 2024 - 14.3K bytes - Viewed (0)