1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167

use std::{collections::HashMap, ops::Mul, sync::Arc};

use candle_core::{Module, Result, Shape, Tensor};
use candle_nn::{init, Dropout, Linear, VarBuilder};
use either::Either;

use crate::{
    frozenlinear::FrozenLinear, LinearLayerLike, LoraConfig, Merge, MergeError, MergeErrorOrError,
    Saveable,
};

#[derive(Debug, Clone)]
pub struct LoraLinear {
    old: Arc<FrozenLinear>,
    ff_a: Linear,
    ff_b: Linear,
    scale: Option<f64>,
    dropout: Option<Arc<Dropout>>,
    merged: bool,
    prefix: String,
    id: usize,
}

#[derive(Clone, Debug)]
/// Configuration for LoraLinear
pub struct LoraLinearConfig {
    in_features: usize,
    out_features: usize,
}

impl LoraLinearConfig {
    pub fn new(in_features: usize, out_features: usize) -> Self {
        LoraLinearConfig {
            in_features,
            out_features,
        }
    }
}

impl LoraLinear {
    pub fn new(
        old: &dyn LinearLayerLike,
        linear_config: &LoraLinearConfig,
        config: &LoraConfig,
        vb: &VarBuilder,
        id: usize,
    ) -> Result<Self> {
        let a = vb.pp(format!("a{id}")).get_with_hints(
            (config.rank, linear_config.in_features),
            "weight",
            init::DEFAULT_KAIMING_NORMAL,
        )?;
        let b = vb.pp(format!("b{id}")).get_with_hints(
            (linear_config.out_features, config.rank),
            "weight",
            init::ZERO,
        )?;

        Ok(LoraLinear {
            old: Arc::new(FrozenLinear::new_from_linear(old)?),
            ff_a: Linear::new(a, None),
            ff_b: Linear::new(b, None),
            scale: if config.rank > 0 {
                Some(config.alpha / config.rank as f64)
            } else {
                None
            },
            dropout: config.dropout.map(|x| Arc::new(Dropout::new(x))),
            merged: false,
            prefix: vb.prefix(),
            id,
        })
    }
}

impl Merge for LoraLinear {
    fn get_delta_weight(&self) -> std::result::Result<Tensor, MergeErrorOrError> {
        let result = self
            .ff_b
            .weight()
            .matmul(self.ff_a.weight())
            .map_err(Either::Right)?;
        Ok(match self.scale {
            Some(scale) => result.mul(scale).map_err(Either::Right)?,
            None => result,
        })
    }

    fn merge_weights(&mut self) -> std::result::Result<(), MergeErrorOrError> {
        if self.merged {
            Err(Either::Left(MergeError::AlreadyMerged))
        } else {
            self.old = Arc::new(
                FrozenLinear::new(
                    (self.old.weight() + self.get_delta_weight()?).map_err(Either::Right)?,
                    self.old.bias().cloned(),
                )
                .map_err(Either::Right)?,
            );
            self.merged = true;
            Ok(())
        }
    }

    fn unmerge_weights(&mut self) -> std::result::Result<(), MergeErrorOrError> {
        if !self.merged {
            Err(Either::Left(MergeError::NotMerged))
        } else {
            self.old = Arc::new(
                FrozenLinear::new(
                    (self.old.weight() - self.get_delta_weight()?).map_err(Either::Right)?,
                    self.old.bias().cloned(),
                )
                .map_err(Either::Right)?,
            );
            self.merged = false;
            Ok(())
        }
    }
}

impl Module for LoraLinear {
    fn forward(&self, input: &Tensor) -> Result<Tensor> {
        if self.merged {
            self.old.forward(input)
        } else {
            //No fan_in_fan_out so no weight.transpose(0,1)
            let mut result = self.old.forward(input)?;
            if let Some(scale) = self.scale {
                let input_new = if self.dropout.is_some() {
                    self.dropout.as_ref().unwrap().forward(input, true)?
                } else {
                    input.clone()
                };

                result =
                    (result + self.ff_b.forward(&self.ff_a.forward(&input_new)?))?.mul(scale)?;
            }
            Ok(result)
        }
    }
}

impl Saveable for LoraLinear {
    fn get_tensors(&self, accum: &mut HashMap<String, Tensor>) {
        accum.insert(
            self.prefix.clone() + &format!(".a{}.weight", self.id),
            self.ff_a.weight().clone(),
        );
        accum.insert(
            self.prefix.clone() + &format!(".b{}.weight", self.id),
            self.ff_b.weight().clone(),
        );
    }
}

impl LinearLayerLike for LoraLinear {
    fn bias(&self) -> Option<&Tensor> {
        self.old.bias()
    }
    fn weight(&self) -> &Tensor {
        self.old.weight()
    }
    fn shape(&self) -> &Shape {
        self.old.shape()
    }
}