cell_clip = 1.000000e+01 : f32, fused_activation_function = "TANH", kernel_type = #tfl<lstm_kernel_type_attr FULL>, proj_clip = 0.01 : f32

      .Case("RELU", tflite::ActivationFunctionType_RELU)
      .Case("RELU_N1_TO_1", tflite::ActivationFunctionType_RELU_N1_TO_1)
      .Case("RELU6", tflite::ActivationFunctionType_RELU6)
      .Case("TANH", tflite::ActivationFunctionType_TANH)
      .Case("SIGN_BIT", tflite::ActivationFunctionType_SIGN_BIT);
}

static tflite::TensorType ConvertDerivedTFLiteTypeAttrForOptionWriter(

func.func private @gelu_decomp_2(%arg0: tensor<5x10xf32>) -> tensor<5x10xf32>
func.func @gelu_aten_approximate(%arg0: tensor<5x10xf32>) -> (tensor<*xf32>) {
  %0 = mhlo.composite "aten.gelu.default" %arg0 {composite_attributes = {approximate = "tanh"}, decomposition = @gelu_decomp_2} : (tensor<5x10xf32>) -> tensor<5x10xf32>
  %1 = "tf.Identity"(%0) {device = ""} : (tensor<5x10xf32>) -> tensor<*xf32>
  %2 = "tf.Identity"(%1) {device = ""} : (tensor<*xf32>) -> tensor<*xf32>

int64_t LogisticOp::GetArithmeticCount(Operation* op) {
  int64_t count;
  // As a very rough ballpark, the cost of evaluating a math function
  // such as tanh or logistic is about 32 multiplications, and about as
  // many additions/subtractions. (Just a power-of-two order-of-magnitude
  // from looking at actual implementations that we use in runtime/code).

}

// CHECK-LABEL:   func @tanh(
// CHECK-SAME:               %[[VAL_0:.*]]: tensor<2xf32>) -> tensor<2xf32> {
// CHECK:           %[[VAL_1:.*]] = "tf.Tanh"(%[[VAL_0]]) : (tensor<2xf32>) -> tensor<2xf32>
// CHECK:           return %[[VAL_1]] : tensor<2xf32>
// CHECK:         }
func.func @tanh(%arg0: tensor<2xf32>) -> tensor<2xf32> {
  %0 = "mhlo.tanh"(%arg0) : (tensor<2xf32>) -> tensor<2xf32>

      %cst_7, %cst_7, %cst_7,
      %cst_7, %cst_7, %cst_7, %cst_7,
      %cst_3, %cst_2,
      %recurrent_stats, %cell_stats,
      %cst_2, %cst_2, %cst_2, %cst_2) {cell_clip = 1.000000e+01 : f32, fused_activation_function = "TANH", proj_clip = 0.000000e+00 : f32, time_major = false}
    : ( tensor<1x28x28xf32>,
        tensor<20x20xf32>, tensor<20x20xf32>, tensor<20x20xf32>, tensor<20x20xf32>,

  %3 = "tfl.mul"(%2, %cst) {fused_activation_function = "NONE"} : (tensor<3xf32>, tensor<f32>) -> tensor<3xf32>
  %4 = "tfl.tanh"(%3) : (tensor<3xf32>) -> tensor<3xf32>
  %5 = "tfl.add"(%4, %cst_1) {fused_activation_function = "NONE"} : (tensor<3xf32>, tensor<f32>) -> tensor<3xf32>

// -----

// CHECK-LABEL: func @tanh
func.func @tanh(%arg0: tensor<2xf32>) -> tensor<2xf32> {
  // CHECK:  mhlo.tanh %arg0 : tensor<2xf32>
  %0 = "tf.Tanh"(%arg0) : (tensor<2xf32>) -> tensor<2xf32>
  func.return %0 : tensor<2xf32>
}

// -----

// CHECK-LABEL: func @tanh_dynamic
func.func @tanh_dynamic(%arg0: tensor<?xf32>) -> tensor<?xf32> {
  // CHECK:  mhlo.tanh %arg0 : tensor<?xf32>

---
hide:
  - navigation
---

# Tính năng

## Tính năng của FastAPI

**FastAPI** cho bạn những tính năng sau:

### Dựa trên những tiêu chuẩn mở

            "Atan", "Atanh", "Ceil", "Cos", "Cosh", "Sin", "Exp", "Expm1",
            "Floor", "IsFinite", "IsInf", "IsNan", "Inv", "Reciprocal", "Log",
            "Log1p", "Invert", "LogicalNot", "Ndtri", "Neg", "Rint", "Round",
            "Rsqrt", "Sigmoid", "Sign", "Sinh", "Softplus", "Softsign", "Sqrt",
            "Square", "Tan", "Tanh", "Real", "Imag", "Erf", "Erfc", "Erfinv",