Functions — github.com/TuckerBMorgan/can-t

hub / github.com/TuckerBMorgan/can-t / functions

Functions825 in github.com/TuckerBMorgan/can-t

⨍ Functions 825◇ Types & classes 48

↓ 2 callersMethodones

Allocates a new tensor with provided shape, all with 1s # Arugments 'shape' - The shape of the allocated tensor

src/central/tensor.rs:214

↓ 2 callersFunctionpreprocess_images

Convert u8 pixel values to normalized f32 tensors

examples/autoencoder.rs:29

↓ 2 callersFunctionpreprocess_labels

Convert digit labels to one-hot encoded vectors

examples/autoencoder.rs:39

↓ 2 callersFunctionread_lines

(filename: &str)

src/lib.rs:142

↓ 2 callersMethodshould_broadcast

(&self, other: Shape)

src/central/shape.rs:373

↓ 2 callersFunctionsoftmax_internal

Internal function for calculation the softmax of a array along a particular axis Breaking it out into its own function, as both the forward and backwa

src/central/softmax_op.rs:9

↓ 2 callersMethodswap_axes

Helper function swap around two axis

src/central/equation.rs:361

↓ 2 callersFunctiontensor_add

Takes two tensors as flat buffers, adds them together at an elementwise level and returns the result # Arguments 'a' : The first vector 'b' : the seco

crates/cant_cpu/src/tensor_add.rs:5

↓ 2 callersFunctiontensor_matmul

( a: &[f32], a_shape: [usize; MAX_DIMS], b: &[f32], b_shape: [usize; MAX_DIMS], )

crates/cant_cpu/src/tensor_matmul.rs:44

↓ 2 callersFunctiontensor_mul

Takes two tensors as flat buffers, multiplies them together at an elementwise level and returns the result # Arguments 'a' : The first vector 'b' : th

crates/cant_cpu/src/tensor_mul.rs:5

↓ 2 callersMethodtopological_sort_util

a util function to preform topological sort on the tensors of the equation for the backward pass # Arugments 'node' - the id of the tensor that is goi

src/central/equation.rs:665

↓ 2 callersMethodzero_grad

Zeroes out the grad, important to call before calling backwards on a value

src/central/equation.rs:723

↓ 1 callersMethodadd_tensors

Takes two tensors as flat buffers, adds them together at an elementwise level and returns the result # Arguments 'a' : The first tensor 'b' : the seco

src/central/equation.rs:189

↓ 1 callersMethodallocate_from_element

Allocates a tensor, and returns the ID of it, so it can be looked up later, will be filled with element # Arguments 'shape' - the shape of the tensor.

src/central/equation.rs:134

↓ 1 callersMethodallocate_ones_tensor

Allocates a tensor, and returns the ID of it, so it can be looked up later, is all ones # Arguments 'shape' - the shape of the tensor. shape.total_siz

src/central/equation.rs:125

↓ 1 callersMethodallocate_random_tesnor

Allocates a tensor with random values # Arguments 'shape' - the shape of desired tensor

src/central/equation.rs:142

↓ 1 callersMethodallocate_tensor_id

Allocated a tensor id, this is unique id for each tensor, used to look in up later

src/central/equation.rs:164

↓ 1 callersFunctionbackward_for_add

Handles calculating and passing back the gradient of an add operation

src/central/add_op.rs:75

↓ 1 callersFunctionbackward_for_chunk

(packet: BackproagationPacket)

src/central/chunk_op.rs:58

↓ 1 callersFunctionbackward_for_clamp

(packet: BackproagationPacket)

src/central/clamp_op.rs:24

↓ 1 callersFunctionbackward_for_cos

Handles calculating and passing back the gradient of a cos operation The derivative of cos(x) is -sin(x)

src/central/cos_op.rs:25

↓ 1 callersFunctionbackward_for_gather

(packet: BackproagationPacket)

src/central/gather_op.rs:55

↓ 1 callersFunctionbackward_for_matmul

(backprop_backet: BackproagationPacket)

src/central/matmul_op.rs:38

↓ 1 callersFunctionbackward_for_mean

Handles calculating and passing back the gradient of a mean operation The gradient is distributed equally among all elements that contributed to the m

src/central/mean_op.rs:58

↓ 1 callersFunctionbackward_for_movedim

(packet: BackproagationPacket)

src/central/movedim_op.rs:46

↓ 1 callersFunctionbackward_for_mul

(backprop_backet: BackproagationPacket)

src/central/mul_op.rs:76

↓ 1 callersFunctionbackward_for_permute

(packet: BackproagationPacket)

src/central/permute.rs:112

↓ 1 callersFunctionbackward_for_pow

(backprop_backet: BackproagationPacket)

src/central/pow_op.rs:18

↓ 1 callersFunctionbackward_for_relu

(backprop_packet: BackproagationPacket)

src/central/relu_op.rs:21

↓ 1 callersFunctionbackward_for_reshape

Handles calculating and passing back the gardient of a reshape operation

src/central/reshape.rs:21

↓ 1 callersFunctionbackward_for_select

Handles calculating and passing back the gradient of a select operation For select operations, gradients are accumulated back to the source tensor at

src/central/select_op.rs:78

↓ 1 callersFunctionbackward_for_sigmoid

Handles calculating and passing back the gradient of a Sigmoid operation The derivative of sigmoid(x) is

src/central/sigmoid_op.rs:27

↓ 1 callersFunctionbackward_for_sin

Handles calculating and passing back the gradient of a sin operation The derivative of sin(x) is cos(x)

src/central/sin_op.rs:25

↓ 1 callersFunctionbackward_for_softmax

Handles calculating and passing back the gradient of a softmax operation The gradient computation involves the Jacobian of softmax For softmax S_i = e

src/central/softmax_op.rs:73

↓ 1 callersFunctionbackward_for_std

(backprop_packet: BackproagationPacket)

src/central/std_op.rs:84

↓ 1 callersFunctionbackward_for_sum

(backprop_backet: BackproagationPacket)

src/central/sum_op.rs:62

↓ 1 callersFunctionbackward_for_tanh

Handles calculating and passing back the gradient of a tanh operation The derivative of tanh(x) is 1 - tanh²(x)

src/central/tanh_op.rs:25

↓ 1 callersFunctionbackward_for_topk

(backprop_backet: BackproagationPacket)

src/central/topk_op.rs:155

↓ 1 callersFunctionbackward_for_unsqueeze

(packet: BackproagationPacket)

src/central/unsqueeze_op.rs:20

↓ 1 callersMethodbackward_for_value

Util function that handles passing back a single value in the equation # Arugments 'incoming_grad' - the tensor that we are passing back

src/central/equation.rs:513

↓ 1 callersFunctionbackwards_for_cat

(packet: BackproagationPacket)

src/central/cat_op.rs:74

↓ 1 callersFunctionbackwards_for_diagonal

(backprop_backet: BackproagationPacket)

src/central/diagonal_op.rs:122

↓ 1 callersFunctionbackwards_for_log

(backprop_packet: BackproagationPacket)

src/central/log.rs:19

↓ 1 callersFunctionbackwards_for_mask_fill

(packet: BackproagationPacket)

src/central/masked_fill.rs:34

↓ 1 callersFunctionbackwards_for_max

(packet: BackproagationPacket)

src/central/max_op.rs:69

↓ 1 callersFunctionbackwards_for_transpose

(packet: BackproagationPacket)

src/central/transpose_op.rs:76

↓ 1 callersMethodbuild_concentration_and_inv_freq

(&self)

src/nn/rotary_embedding.rs:68

↓ 1 callersMethodcan_matmul

Returns if two shapes can be matmuled together This only covers 2x2 cases or lower higher dimenions are considered "batched" matmul # Arguments 'other

src/central/shape.rs:180

↓ 1 callersFunctionclean_up_tensor_store

()

src/central/mod.rs:100

↓ 1 callersMethodclip_grad

(&mut self, max_norm: f32)

src/central/equation.rs:730

↓ 1 callersFunctioncreate_instance

(project, zone, instance_file, credentials_file=None)

utils/launch_nvidia_on_gcp.py:6

↓ 1 callersMethoddependencies

Returns which Tensors created this operation

src/central/tensor.rs:43

↓ 1 callersFunctiondraw_ui

(f: &mut Frame, loss_history: &[(f64, f64)], original_ascii: &str, recon_ascii: &str)

examples/autoencoder.rs:262

↓ 1 callersMethodend_clock_with_name

Ends a clock with a given name, giving it an offical length

src/utils/timing.rs:23

↓ 1 callersMethodend_timer

util function to end a time with a provided name # Aruguments 'name': the name of the timer

src/central/equation.rs:181

↓ 1 callersFunctionextract_batch

Extract batch from tensor by copying rows

examples/autoencoder.rs:51

↓ 1 callersMethodforward

(&mut self, input: Tensor)

src/optimizers/stochastic_gradient_descent.rs:78

↓ 1 callersMethodforward

(&mut self, input: Tensor)

src/optimizers/adamw.rs:147

↓ 1 callersMethodforward

(&mut self, input: Tensor)

src/optimizers/adam.rs:138

↓ 1 callersMethodforward

(&mut self, input: Tensor)

examples/autoencoder.rs:124

↓ 1 callersMethodget_item

(&self, index: usize)

src/utils/data_set.rs:47

↓ 1 callersMethodget_parameters

(&self)

src/nn/embedding.rs:28

↓ 1 callersMethodget_parameters

(&self)

src/nn/layer_norm.rs:72

↓ 1 callersMethodget_parameters

(&self)

src/nn/rmsnorm.rs:42

↓ 1 callersMethodget_parameters

(&mut self, model: &mut dyn Model)

src/optimizers/stochastic_gradient_descent.rs:31

↓ 1 callersMethodget_parameters

(&mut self, model: &mut dyn Model)

src/optimizers/adamw.rs:43

↓ 1 callersMethodget_parameters

(&mut self, model: &mut dyn Model)

src/optimizers/adam.rs:43

↓ 1 callersMethodget_random_batch

(&self, batch_size: usize)

src/utils/data_loader.rs:17

↓ 1 callersMethodget_tensor_dims

(&self, name: impl AsRef<str>)

src/utils/gguf.rs:40

↓ 1 callersMethodget_weight_for_tensor

(&self, name: impl AsRef<str>)

src/utils/gguf.rs:50

↓ 1 callersFunctionlerp_array

(start: f32, end: f32, steps: usize)

src/nn/sequential.rs:267

↓ 1 callersFunctionlerp_array

(start: f32, end: f32, steps: usize)

examples/autoencoder.rs:200

↓ 1 callersFunctionload_mnist_data

Helper function to load and preprocess MNIST data

examples/autoencoder.rs:84

↓ 1 callersFunctionloop_count

(shape: [usize; MAX_DIMS])

crates/cant_cpu/src/tensor_matmul.rs:39

↓ 1 callersFunctionloop_count

(shape: [usize; MAX_DIMS])

crates/cant_cuda/src/tensor_matmul.rs:42

↓ 1 callersFunctionloop_count

(shape: [usize; MAX_DIMS])

crates/cant_metal/src/tensor_matmul.rs:43

↓ 1 callersMethodmatmul_tensor

Takes two tensors are flat buffers and preforms matmul on them and returns the result # Arugments 'a' - the first tensor 'b' - The seconf tensor

src/central/equation.rs:203

↓ 1 callersMethodmul_tensors

Takes two tensors as flat buffers, multiplies them together at an elementwise level and returns the result # Arguments 'a' : The first tensor 'b' : th

src/central/equation.rs:295

↓ 1 callersMethodnumber_of_samples

(&self)

src/utils/data_loader.rs:38

↓ 1 callersMethodpartial_cmp

(&self, other: &Self)

src/central/topk_op.rs:25

↓ 1 callersFunctionproduct

(slice: &[usize])

crates/cant_cpu/src/tensor_matmul.rs:32

↓ 1 callersFunctionproduct

(slice: &[usize])

crates/cant_cuda/src/tensor_matmul.rs:35

↓ 1 callersFunctionproduct

(slice: &[usize])

crates/cant_metal/src/tensor_matmul.rs:36

↓ 1 callersMethodset_mask

(&mut self, mask: Tensor)

src/nn/multihead_attention.rs:155

↓ 1 callersMethodset_parameter

sets the weights of tensor Arguments 'param_id' : the Id of the Tensor we are updating 'new_parameters' : the new values

src/central/equation.rs:761

↓ 1 callersMethodset_single_value

(&mut self, tensor_id: TensorID, index: Indexable, value: f32)

src/central/equation.rs:480

↓ 1 callersMethodset_tensor_grad

Copies data into the grad of tensor_id # Arugments 'tensor_id' - Id of for the loopup on the tensor 'grad' - the grad we are copying in

src/central/equation.rs:442

↓ 1 callersFunctionshifted_batches

(data: &[u32], batch_size: usize)

src/nn/gpt2.rs:238

↓ 1 callersFunctionsingle_run

(input: Vec<f32>, expected_output: Vec<f32>, model: &mut GPT2, vocab_size: usize)

src/nn/gpt2.rs:260

↓ 1 callersMethodstart_clock_with_name

Starts a clock with a given name

src/utils/timing.rs:18

↓ 1 callersMethodstart_timer

util function to start a time with a provided name # Aruguments 'name': the name of the timer

src/central/equation.rs:174

↓ 1 callersMethodsub_vector

Takes two tensors as flat buffers, subs them together at an elementwise level and returns the result # Arguments 'a' : The first tensor 'b' : the seco

src/central/equation.rs:285

↓ 1 callersMethodswap_index

Returns a new shape with the dimensions changed to a new ones # Arguments 'index' - the 0base index we will be swapping 'new_dimension' the new dimen

src/central/shape.rs:141

↓ 1 callersFunctiontensor_sub

Takes two tensors as flat buffers, subs them together at an elementwise level and returns the result # Arguments 'a' : The first vector 'b' : the seco

crates/cant_cpu/src/tensor_sub.rs:5

↓ 1 callersMethodupdate

(&mut self)

src/optimizers/stochastic_gradient_descent.rs:35

↓ 1 callersMethodupdate

(&mut self)

src/optimizers/adamw.rs:58

↓ 1 callersMethodupdate

(&mut self)

src/optimizers/adam.rs:58

↓ 1 callersMethodupdate_parameters

Updates all parameters for all tensor that are marked for needed gradients(set_requires_grad) Arguments 'learning_rate': a singe learning rate applied

src/central/equation.rs:739

↓ 1 callersMethodupdate_single_parameter

Updates the paramaters of a single tensor Arugments 'parameter': the tensor that will have its weights updated 'learning_rate': the learning rate appl

src/central/equation.rs:784

↓ 1 callersMethodupdate_single_parameter_with_provided_gradient

Updates the paramaters of a single tensor using the provided gradient vector Arugments 'parameter': the tensor that will have its weights updated 'lea

src/central/equation.rs:819

← previous next →101–200 of 825, ranked by callers