mamba-rs 0.3.1

Mamba SSM and Mamba-3 SISO in Rust with optional CUDA GPU acceleration. Inference and training (BPTT through SSM state, AdamW), CPU + GPU paths, custom CUDA kernels, CUDA Graph capture, f32 / bf16 / f16. Batch-invariant bf16 inference — per-row output is bit-identical across batch sizes.
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

#![allow(clippy::needless_range_loop)]
//! Shared normalization operations for Mamba SSM and Mamba-3 SISO.
//!
//! - `rms_norm` — standard RMSNorm (scale only, no bias)
//! - `rms_norm_weighted` — RMSNorm with learnable weight
//! - `bcnorm` — per-group RMSNorm for B/C (Mamba-3)
//! - `rmsnorm_gated` — RMSNormGated: norm(y) * weight * SiLU(z) (Mamba-3)

use super::fast_math::fast_exp_scalar;

/// RMSNorm: `out[i] = x[i] / rms * weight[i]`
/// Returns inv_rms for backward.
#[inline]
pub fn rms_norm_weighted(out: &mut [f32], x: &[f32], weight: &[f32], eps: f32) -> f32 {
    let n = x.len();
    debug_assert_eq!(n, weight.len());
    debug_assert_eq!(n, out.len());

    let mut sum_sq = 0.0_f32;
    for &v in x {
        sum_sq += v * v;
    }
    let inv_rms = 1.0 / (sum_sq / n as f32 + eps).sqrt();

    for i in 0..n {
        out[i] = x[i] * inv_rms * weight[i];
    }
    inv_rms
}

/// RMSNorm in-place: `x[i] = x[i] / rms * weight[i]`
#[inline]
pub fn rms_norm_inplace(x: &mut [f32], weight: &[f32], eps: f32) -> f32 {
    let n = x.len();
    let mut sum_sq = 0.0_f32;
    for &v in x.iter() {
        sum_sq += v * v;
    }
    let inv_rms = 1.0 / (sum_sq / n as f32 + eps).sqrt();

    for i in 0..n {
        x[i] *= inv_rms * weight[i];
    }
    inv_rms
}

/// BCNorm: per-group RMSNorm(x) * weight for B/C in Mamba-3.
/// x is `[ngroups * d_state]`, weight is `[d_state]` (shared across groups).
/// Returns per-group inv_rms values.
#[inline]
pub fn bcnorm(
    out: &mut [f32],
    x: &[f32],
    weight: &[f32],
    ngroups: usize,
    d_state: usize,
    eps: f32,
    inv_rms_out: &mut [f32],
) {
    debug_assert_eq!(x.len(), ngroups * d_state);
    debug_assert_eq!(out.len(), ngroups * d_state);
    debug_assert_eq!(weight.len(), d_state);
    debug_assert!(inv_rms_out.len() >= ngroups);

    for g in 0..ngroups {
        let start = g * d_state;
        let end = start + d_state;
        let group = &x[start..end];

        let mut sum_sq = 0.0_f32;
        for &v in group {
            sum_sq += v * v;
        }
        let inv_rms = 1.0 / (sum_sq / d_state as f32 + eps).sqrt();
        inv_rms_out[g] = inv_rms;

        for i in 0..d_state {
            out[start + i] = group[i] * inv_rms * weight[i];
        }
    }
}

/// RMSNormGated: per-group RMSNorm(y) * weight * SiLU(z).
/// group_size = headdim (Mamba-3 convention).
#[inline]
pub fn rmsnorm_gated(
    out: &mut [f32],
    y: &[f32],
    z: &[f32],
    weight: &[f32],
    group_size: usize,
    eps: f32,
) {
    let dim = y.len();
    debug_assert_eq!(dim, z.len());
    debug_assert_eq!(dim, weight.len());
    debug_assert_eq!(dim, out.len());

    for g_start in (0..dim).step_by(group_size) {
        let g_end = (g_start + group_size).min(dim);
        let g_len = g_end - g_start;

        let mut sum_sq = 0.0_f32;
        for i in g_start..g_end {
            sum_sq += y[i] * y[i];
        }
        let inv_rms = 1.0 / (sum_sq / g_len as f32 + eps).sqrt();

        for i in g_start..g_end {
            let zi = z[i];
            let silu_z = zi / (1.0 + fast_exp_scalar(-zi));
            out[i] = y[i] * inv_rms * weight[i] * silu_z;
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_rms_norm_weighted_unit_weight() {
        let x = vec![3.0, 4.0];
        let w = vec![1.0, 1.0];
        let mut out = vec![0.0; 2];
        let inv_rms = rms_norm_weighted(&mut out, &x, &w, 1e-5);

        let rms = (12.5_f32).sqrt(); // (9+16)/2 = 12.5
        let expected_inv = 1.0 / rms;
        assert!((inv_rms - expected_inv).abs() < 1e-5);
        assert!((out[0] - 3.0 * expected_inv).abs() < 1e-5);
        assert!((out[1] - 4.0 * expected_inv).abs() < 1e-5);
    }

    #[test]
    fn test_bcnorm_per_group() {
        let x = vec![1.0, 2.0, 3.0, 4.0]; // 2 groups of 2
        let w = vec![1.0, 1.0];
        let mut out = vec![0.0; 4];
        let mut inv_rms = vec![0.0; 2];
        bcnorm(&mut out, &x, &w, 2, 2, 1e-5, &mut inv_rms);

        // Group 0: rms = sqrt((1+4)/2) = sqrt(2.5)
        // Group 1: rms = sqrt((9+16)/2) = sqrt(12.5)
        assert!(inv_rms[0] > 0.0);
        assert!(inv_rms[1] > 0.0);
        assert!(inv_rms[0] > inv_rms[1]); // smaller values → larger inv_rms
    }
}