Fixing cargo doc

dfdx/src/lib.rs

@@ -5,10 +5,10 @@
 //! that a model is stored on.
 //!
 //! For example, a linear model has a couple pieces:
-//! 1. The architecture configuration type: [LinearConfig]
-//! 2. The actual built type that contains the parameters: [Linear]
+//! 1. The architecture configuration type: [nn::LinearConfig]
+//! 2. The actual built type that contains the parameters: [nn::Linear]
 //!
-//! There's a third piece for convenience: [LinearConstConfig], which let's you specify dimensions at compile time.
+//! There's a third piece for convenience: [nn::LinearConstConfig], which let's you specify dimensions at compile time.
 //!
 //! For specifying architecture, you just need the dimensions for the linear, but not the device/dtype:
 //! ```rust
@@ -22,11 +22,11 @@
 //! ```
 //! **Note** that we don't have any idea on what device or what dtype this will be.
 //!
-//! When we build this configuration into a [Linear] object, it will be placed on a device and have a certain dtype.
+//! When we build this configuration into a [nn::Linear] object, it will be placed on a device and have a certain dtype.
 //!
 //! # Building a model from an architecture
 //!
-//! We will use [BuildModuleExt::build_module()], an extension trait on devices, to actually construct a model.
+//! We will use [nn::BuildModuleExt::build_module()], an extension trait on devices, to actually construct a model.
 //!
 //! ```rust
 //! # use dfdx::prelude::*;
@@ -40,7 +40,7 @@
 //!
 //! # Using a model
 //!
-//! There are many things you can do with models. The main action is calling [Module::forward()] and [Module::forward_mut()]
+//! There are many things you can do with models. The main action is calling [nn::Module::forward()] and [nn::Module::forward_mut()]
 //! during inference and training.
 //!
 //! ```rust
@@ -63,9 +63,9 @@
 //!
 //! Under the hood, the code generated for Sequential vs tuples are identical.
 //!
-//! ## Deriving [Sequential]
+//! ## Deriving [nn::Sequential]
 //!
-//! See [Sequential] for more detailed information.
+//! See [nn::Sequential] for more detailed information.
 //!
 //! ```rust
 //! # use dfdx::prelude::*;
@@ -114,16 +114,16 @@
 //!
 //! # Optimizers and Gradients
 //!
-//! *See [optim] for more information*
+//! *See [nn::optim] for more information*
 //!
 //! dfdx-nn supports a number of the standard optimizers:
 //!
 //! | Optimizer | dfdx | pytorch |
 //! | --- | --- | --- |
-//! | SGD | [optim::Sgd] | `torch.optim.SGD` |
-//! | Adam | [optim::Adam] | torch.optim.Adam` |
-//! | AdamW | [optim::Adam] with [optim::WeightDecay::Decoupled] | `torch.optim.AdamW` |
-//! | RMSprop | [optim::RMSprop] | `torch.optim.RMSprop` |
+//! | SGD | [nn::optim::Sgd] | `torch.optim.SGD` |
+//! | Adam | [nn::optim::Adam] | torch.optim.Adam` |
+//! | AdamW | [nn::optim::Adam] with [nn::optim::WeightDecay::Decoupled] | `torch.optim.AdamW` |
+//! | RMSprop | [nn::optim::RMSprop] | `torch.optim.RMSprop` |
 //!
 //! You can use optimizers to optimize neural networks (or even tensors!). Here's
 //! a simple example of how to do this:

dfdx/src/nn/layers/abs.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::abs()]
+/// Calls [crate::tensor_ops::abs()]
 #[derive(Default, Debug, Clone, Copy, crate::nn::CustomModule)]
 pub struct Abs;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> crate::nn::Module<Tensor<S, E, D, T>>

dfdx/src/nn/layers/batch_norm1d.rs

@@ -12,8 +12,8 @@ use crate::prelude::*;
 /// # Training vs Inference
 ///
 /// BatchNorm1D supports the following cases (see sections below for more details):
-/// 1. **Training**: [crate::Module::forward_mut()] and [OwnedTape] on the input tensor
-/// 2. **Inference**: [crate::Module::forward()] and [NoneTape] on the input tensor.
+/// 1. **Training**: [crate::nn::Module::forward_mut()] and [OwnedTape] on the input tensor
+/// 2. **Inference**: [crate::nn::Module::forward()] and [NoneTape] on the input tensor.
 ///
 /// Examples:
 /// ```rust

dfdx/src/nn/layers/batch_norm2d.rs

@@ -12,8 +12,8 @@ use crate::prelude::*;
 /// # Training vs Inference
 ///
 /// BatchNorm2D supports the following cases (see sections below for more details):
-/// 1. **Training**: [crate::Module::forward_mut()] and [OwnedTape] on the input tensor
-/// 2. **Inference**: [crate::Module::forward()] and [NoneTape] on the input tensor.
+/// 1. **Training**: [crate::nn::Module::forward_mut()] and [OwnedTape] on the input tensor
+/// 2. **Inference**: [crate::nn::Module::forward()] and [NoneTape] on the input tensor.
 ///
 /// *NOTE: ModuleMut/NoneTape, and Module/OwnedTape will fail to compile.*
 ///

dfdx/src/nn/layers/conv2d.rs

@@ -19,7 +19,7 @@ use crate::prelude::*;
 /// };
 /// ```
 ///
-/// To create a biased conv, combine with [crate::Bias2D].
+/// To create a biased conv, combine with [crate::nn::Bias2D].
 ///
 /// Generics:
 /// - `InChan`: The number of input channels in an image.

dfdx/src/nn/layers/conv_trans2d.rs

@@ -4,7 +4,7 @@ use crate::prelude::*;
 ///
 /// **Pytorch Equivalent**: `torch.nn.ConvTranspose2d(..., bias=False)`
 ///
-/// To create a biased conv, combine with [crate::Bias2D].
+/// To create a biased conv, combine with [crate::nn::Bias2D].
 ///
 /// Generics:
 /// - `InChan`: The number of input channels in an image.

dfdx/src/nn/layers/cos.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::cos()].
+/// Calls [crate::tensor_ops::cos()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Cos;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Cos {

dfdx/src/nn/layers/dropout.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::dropout()] with `p = 1.0 / N` in [Module::forward_mut()], and does nothing in  [Module::forward()].
+/// Calls [crate::tensor_ops::dropout()] with `p = 1.0 / N` in [Module::forward_mut()], and does nothing in  [Module::forward()].
 ///
 /// Generics:
 /// - `N`: p is set as `1.0 / N`
@@ -44,12 +44,11 @@ impl<const N: usize, S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Ten
     }
 }
 
-/// Calls [dfdx::tensor_ops::dropout()] in [Module::forward_mut()], and does nothing in  [Module::forward()].
+/// Calls [crate::tensor_ops::dropout()] in [crate::nn::Module::forward_mut()], and does nothing in  [crate::nn::Module::forward()].
 ///
 /// Examples:
 /// ```rust
 /// # use dfdx::prelude::*;
-/// # use dfdx::*;
 /// # let dev: Cpu = Default::default();
 /// let mut dropout = Dropout { p: 0.5 };
 /// let grads = dropout.alloc_grads();

dfdx/src/nn/layers/exp.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::exp()].
+/// Calls [crate::tensor_ops::exp()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Exp;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Exp {

dfdx/src/nn/layers/gelu.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::fast_gelu()].
+/// Calls [crate::tensor_ops::fast_gelu()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct FastGeLU;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for FastGeLU {
@@ -11,7 +11,7 @@ impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>>
     }
 }
 
-/// Calls [dfdx::tensor_ops::accurate_gelu()].
+/// Calls [crate::tensor_ops::accurate_gelu()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct AccurateGeLU;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for AccurateGeLU {

dfdx/src/nn/layers/ln.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::ln()].
+/// Calls [crate::tensor_ops::ln()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Ln;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Ln {

dfdx/src/nn/layers/log_softmax.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::log_softmax()].
+/// Calls [crate::tensor_ops::log_softmax()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct LogSoftmax;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for LogSoftmax {

dfdx/src/nn/layers/prelu.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::prelu()] with learnable value.
+/// Calls [crate::tensor_ops::prelu()] with learnable value.
 #[derive(Debug, Clone, Copy)]
 pub struct PReLUConfig(pub f64);
 

dfdx/src/nn/layers/prelu1d.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::prelu()] with learnable values along second dimension.
+/// Calls [crate::tensor_ops::prelu()] with learnable values along second dimension.
 #[derive(Debug, Clone, Copy)]
 pub struct PReLU1DConfig<C: Dim> {
     pub a: f64,

dfdx/src/nn/layers/relu.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::relu()].
+/// Calls [crate::tensor_ops::relu()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct ReLU;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for ReLU {

dfdx/src/nn/layers/sigmoid.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::sigmoid()].
+/// Calls [crate::tensor_ops::sigmoid()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Sigmoid;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Sigmoid {

dfdx/src/nn/layers/sin.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::sin()].
+/// Calls [crate::tensor_ops::sin()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Sin;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Sin {

dfdx/src/nn/layers/softmax.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::softmax()] on the last axis of the input.
+/// Calls [crate::tensor_ops::softmax()] on the last axis of the input.
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Softmax;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Softmax {

dfdx/src/nn/layers/sqrt.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::sqrt()].
+/// Calls [crate::tensor_ops::sqrt()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Sqrt;
 

dfdx/src/nn/layers/square.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::square()].
+/// Calls [crate::tensor_ops::square()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Square;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Square {

dfdx/src/nn/layers/tanh.rs

@@ -1,6 +1,6 @@
 use crate::prelude::*;
 
-/// Calls [dfdx::tensor_ops::tanh()].
+/// Calls [crate::tensor_ops::tanh()].
 #[derive(Default, Debug, Clone, Copy, CustomModule)]
 pub struct Tanh;
 impl<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>> Module<Tensor<S, E, D, T>> for Tanh {

dfdx/src/nn/optim/adam.rs

@@ -23,7 +23,7 @@ use crate::{
 /// });
 /// ```
 ///
-/// See module level documentation at [crate::optim] for examples of how to actually use an optimizer.
+/// See module level documentation at [crate::nn::optim] for examples of how to actually use an optimizer.
 #[derive(Debug, Clone)]
 pub struct Adam<M, E: Dtype, D: Storage<E>> {
     /// Hyperparameter configuration

dfdx/src/nn/optim/mod.rs

@@ -10,8 +10,8 @@
 //!
 //! # Updating network parameters
 //!
-//! This is done via [crate::Optimizer::update()], where you pass in a mutable [crate::Module], and
-//! the [dfdx::tensor::Gradients]:
+//! This is done via [crate::nn::Optimizer::update()], where you pass in a mutable [crate::nn::Module], and
+//! the [crate::tensor::Gradients]:
 //!
 //! ```rust
 //! # use dfdx::prelude::*;

dfdx/src/nn/optim/sgd.rs

@@ -27,7 +27,7 @@ use crate::{
 /// });
 /// ```
 ///
-/// See module level documentation at [crate::optim] for examples of how to actually use an optimizer.
+/// See module level documentation at [crate::nn::optim] for examples of how to actually use an optimizer.
 #[derive(Debug, Clone)]
 pub struct Sgd<M, E: Dtype, D: Storage<E>> {
     pub cfg: SgdConfig,