lib/Peta/NN/Layer/Embed.pm

filter: Statementstmt 100% 20 / 20 Branchbran 100% 2 / 2 Conditioncond n/a MC/DCmcdc n/a Subroutinesub 100% 9 / 9 Podpod 100% 8 / 8 totaltotal 100% 39 / 39 CC 2 SCAR 1.8
CC
2
Cov
100%
init
2.0
BEGIN
1.0
new
1.0
CRAP 1.2
« lib/Peta/NN/Layer/Dense.pm ↑ Summary lib/Peta/NN/Backend/WebGPU.pm »
1 package Peta::NN::Layer::Embed;
2 # ABSTRACT: learned vectors for token indices
3  
4 # Embedding layer: the input is a fixed number of token indices per sample
5 # (the characters of a window, say), each looked up in a learned table and
6 # the rows concatenated. It can only be the first layer, so it never
7 # produces an input gradient.
8  
9 27 ▶use v5.36;
Subroutine BEGIN: called CC=1 SCAR=0.0
10  
11 27 our $VERSION = '0.2610090';
12  
13 27 my $INIT_SCALE = 0.1;
14  
15 116 ▶sub new ($class, %arg) {
Subroutine new: called CC=1 SCAR=0.0
Pod documented
16 116 return bless { dim => $arg{dim} }, $class;
17 }
18  
19 1691 ▶sub type ($self) { return 'embed' }
Subroutine type: called CC=1 SCAR=0.0
Pod documented
20  
21 # $in is { tokens => count per sample, vocab => table rows }.
22 116 ▶sub init ($self, $in, $rng, $backend) {
Subroutine init: called CC=2 SCAR=6.9
Pod documented
23 116 ▶ die "embed must be the first layer, on a token input\n" if ref $in ne 'HASH';
Branch: 1 branch
BranchTrueFalseTotal
if ref $in ne "HASH" 1 115 116
24 115 my $dim = $self->{dim};
25 115 $self->{backend} = $backend;
26 115 $self->{E} = $backend->tensor([ map { $rng->normal * $INIT_SCALE } 1 .. $in->{vocab} * $dim ], $dim);
27 115 return $in->{tokens} * $dim;
28 }
29  
30 14817 ▶sub param_names ($self) { return qw(E) }
Subroutine param_names: called CC=1 SCAR=0.0
Pod documented
31  
32 # Numbers per row of the table: one row per token.
33 794 ▶sub param_cols ($self, $name) { return $self->{dim} }
Subroutine param_cols: called CC=1 SCAR=0.0
Pod documented
34  
35 70245 ▶sub forward ($self, $tokens) {
Subroutine forward: called CC=1 SCAR=0.0
Pod documented
36 70245 $self->{tokens} = $tokens;
37 70245 return $self->{backend}->embed($self->{E}, $tokens);
38 }
39  
40 69142 ▶sub backward ($self, $dY) {
Subroutine backward: called CC=1 SCAR=0.0
Pod documented
41 69142 $self->{gE} = $self->{backend}->embed_grad($self->{E}, $self->{tokens}, $dY);
42 69142 return;
43 }
44  
45 29 ▶sub spec ($self) { return [ embed => $self->{dim} ] }
Subroutine spec: called CC=1 SCAR=0.0
Pod documented
46  
47 27 1;
48  
49 __END__
« lib/Peta/NN/Layer/Dense.pm ↑ Summary lib/Peta/NN/Backend/WebGPU.pm »