numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
Documentation 
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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215

//! Backward implementation for RMS Normalization

use crate::autograd::GradFn;
use crate::autograd::var::Var;
use crate::autograd::var_ops::{var_mean, var_mul, var_sub, var_sum};
use crate::error::Result;
use crate::ops::{BinaryOps, ReduceOps, ScalarOps, TensorOps, UnaryOps};
use crate::runtime::{Runtime, RuntimeClient};
use crate::tensor::{Tensor, TensorId};
use std::sync::Arc;

/// Backward for RMS Normalization: y = x / rms(x) * weight
///
/// Where rms(x) = sqrt(mean(x^2, dim=-1) + eps)
///
/// Gradients:
/// - d_input = rstd * (grad_out * weight - x_norm * mean(grad_out * weight * x_norm, dim=-1))
/// - d_weight = sum(grad_out * x_norm, batch_dims)
///
/// Where rstd = 1/rms(x), x_norm = x * rstd
pub struct RmsNormBackward<R: Runtime> {
    input_ids: [TensorId; 2],
    saved_tensors: Vec<Tensor<R>>, // [input, weight]
    eps: f32,
    input_grad_fns: [Option<Arc<dyn GradFn<R>>>; 2],
}

impl<R: Runtime> RmsNormBackward<R> {
    /// Create a new RmsNormBackward
    pub fn new(
        input_id: TensorId,
        weight_id: TensorId,
        input: Tensor<R>,
        weight: Tensor<R>,
        eps: f32,
        input_grad_fn: Option<Arc<dyn GradFn<R>>>,
        weight_grad_fn: Option<Arc<dyn GradFn<R>>>,
    ) -> Self {
        Self {
            input_ids: [input_id, weight_id],
            saved_tensors: vec![input, weight],
            eps,
            input_grad_fns: [input_grad_fn, weight_grad_fn],
        }
    }
}

impl<R: Runtime> GradFn<R> for RmsNormBackward<R>
where
    R::Client: TensorOps<R> + ScalarOps<R> + BinaryOps<R> + ReduceOps<R> + UnaryOps<R>,
{
    fn backward(&self, grad_output: &Tensor<R>) -> Result<Vec<Option<Tensor<R>>>> {
        let client = R::default_client(grad_output.device());
        let saved_input = &self.saved_tensors[0];
        let saved_weight = &self.saved_tensors[1];
        let ndim = saved_input.ndim();
        let last_dim = ndim - 1;

        // Recompute rstd = 1 / sqrt(mean(x^2, dim=-1, keepdim=True) + eps)
        let x_sq = client.mul(saved_input, saved_input)?;
        let mean_x_sq = client.mean(&x_sq, &[last_dim], true)?;
        let variance_eps = client.add_scalar(&mean_x_sq, self.eps as f64)?;
        let rms = client.sqrt(&variance_eps)?;
        let rstd = client.recip(&rms)?;

        // x_norm = x * rstd
        let x_norm = client.mul(saved_input, &rstd)?;

        // d_input = rstd * (grad_output * weight - x_norm * mean(grad_output * weight * x_norm))
        let gw = client.mul(grad_output, saved_weight)?;
        let gw_xn = client.mul(&gw, &x_norm)?;
        let mean_gw_xn = client.mean(&gw_xn, &[last_dim], true)?;
        let correction = client.mul(&x_norm, &mean_gw_xn)?;
        let inner = client.sub(&gw, &correction)?;
        let d_input = client.mul(&inner, &rstd)?;

        // d_weight = sum(grad_output * x_norm, batch_dims)
        let g_xn = client.mul(grad_output, &x_norm)?;
        let batch_dims: Vec<usize> = (0..last_dim).collect();
        let d_weight = if batch_dims.is_empty() {
            g_xn
        } else {
            client.sum(&g_xn, &batch_dims, false)?
        };

        Ok(vec![Some(d_input), Some(d_weight)])
    }

    fn backward_var(&self, grad_output: &Var<R>) -> Result<Vec<Option<Var<R>>>>
    where
        R::Client: RuntimeClient<R>
            + TensorOps<R>
            + ScalarOps<R>
            + BinaryOps<R>
            + ReduceOps<R>
            + UnaryOps<R>,
    {
        let client = R::default_client(grad_output.tensor().device());
        let saved_input = &self.saved_tensors[0];
        let saved_weight = &self.saved_tensors[1];
        let ndim = saved_input.ndim();
        let last_dim = ndim - 1;

        // Recompute rstd and x_norm from saved tensors (treat as constants)
        let x_sq = client.mul(saved_input, saved_input)?;
        let mean_x_sq = client.mean(&x_sq, &[last_dim], true)?;
        let variance_eps = client.add_scalar(&mean_x_sq, self.eps as f64)?;
        let rms = client.sqrt(&variance_eps)?;
        let rstd = client.recip(&rms)?;
        let x_norm = client.mul(saved_input, &rstd)?;

        // Wrap as non-differentiable Vars (constants w.r.t. grad_output)
        let rstd_var = Var::new(rstd, false);
        let x_norm_var = Var::new(x_norm, false);
        let weight_var = Var::new(saved_weight.clone(), false);

        // d_input = rstd * (grad_output * weight - x_norm * mean(grad_output * weight * x_norm))
        let gw = var_mul(grad_output, &weight_var, &client)?;
        let gw_xn = var_mul(&gw, &x_norm_var, &client)?;
        let mean_gw_xn = var_mean(&gw_xn, &[last_dim], true, &client)?;
        let correction = var_mul(&x_norm_var, &mean_gw_xn, &client)?;
        let inner = var_sub(&gw, &correction, &client)?;
        let d_input = var_mul(&inner, &rstd_var, &client)?;

        // d_weight = sum(grad_output * x_norm, batch_dims)
        let g_xn = var_mul(grad_output, &x_norm_var, &client)?;
        let batch_dims: Vec<usize> = (0..last_dim).collect();
        let d_weight = if batch_dims.is_empty() {
            g_xn
        } else {
            var_sum(&g_xn, &batch_dims, false, &client)?
        };

        Ok(vec![Some(d_input), Some(d_weight)])
    }

    fn inputs(&self) -> &[TensorId] {
        &self.input_ids
    }

    fn input_grad_fns(&self) -> Vec<Option<Arc<dyn GradFn<R>>>> {
        self.input_grad_fns.to_vec()
    }

    fn saved_tensors(&self) -> &[Tensor<R>] {
        &self.saved_tensors
    }

    fn name(&self) -> &'static str {
        "RmsNormBackward"
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::dtype::DType;
    use crate::runtime::cpu::{CpuDevice, CpuRuntime};

    #[test]
    fn test_rms_norm_backward_uniform() {
        let device = CpuDevice::new();

        let input = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 1.0, 1.0, 1.0], &[1, 4], &device);
        let weight = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 1.0, 1.0, 1.0], &[4], &device);
        let eps = 1e-5f32;

        let grad_out = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 0.0, 0.0, 0.0], &[1, 4], &device);

        let backward = RmsNormBackward::<CpuRuntime>::new(
            input.id(),
            weight.id(),
            input,
            weight,
            eps,
            None,
            None,
        );
        let grads = backward.backward(&grad_out).unwrap();

        assert_eq!(grads.len(), 2);
        let d_input: Vec<f32> = grads[0].as_ref().unwrap().to_vec();
        let d_weight: Vec<f32> = grads[1].as_ref().unwrap().to_vec();

        assert!(d_input[0] > 0.0, "d_input[0] should be positive");
        assert!(d_input[1] < 0.0, "d_input[1] should be negative");
        assert!((d_weight[0] - 1.0).abs() < 0.01);
        assert!(d_weight[1].abs() < 1e-5);
    }

    #[test]
    fn test_rms_norm_backward_gradient_sum() {
        let device = CpuDevice::new();

        let input = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0], &[1, 3], &device);
        let weight = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 1.0, 1.0], &[3], &device);
        let grad_out = Tensor::<CpuRuntime>::ones(&[1, 3], DType::F32, &device);

        let backward = RmsNormBackward::<CpuRuntime>::new(
            input.id(),
            weight.id(),
            input,
            weight,
            1e-5,
            None,
            None,
        );
        let grads = backward.backward(&grad_out).unwrap();
        let d_input: Vec<f32> = grads[0].as_ref().unwrap().to_vec();

        for val in &d_input {
            assert!(val.is_finite(), "gradient should be finite");
        }
    }
}