libmir-cuda 0.1.0

CUDA inference backend for libmir
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

use mircuda::{DeviceBuffer, Stream, bf16};

use super::{CudaBackend, GatedActivation, NvFp4ExpertBank, view};
use crate::{
    Error, Result,
    kernels::{
        NvFp4MicroBanks, NvFp4MicroKernels, NvFp4MicroLaunch, NvFp4MicroSpec, NvFp4MicroWorkspace,
    },
};

/// Direct small-batch NVFP4 `MoE` candidate with packed activation residency.
#[derive(Debug)]
pub struct DirectNvFp4MoeBf16 {
    kernels: NvFp4MicroKernels,
    gate: NvFp4ExpertBank,
    up: NvFp4ExpertBank,
    down: NvFp4ExpertBank,
    gate_packed: DeviceBuffer<u8>,
    up_packed: DeviceBuffer<u8>,
    gate_scales: DeviceBuffer<u8>,
    up_scales: DeviceBuffer<u8>,
    intermediate_packed: DeviceBuffer<u8>,
    intermediate_scales: DeviceBuffer<u8>,
    stream: Stream,
    tokens: usize,
    selected: usize,
    hidden: usize,
}

impl CudaBackend {
    /// Prepares a direct W4A4 candidate for isolated numerical and latency
    /// gates.
    pub fn prepare_direct_nvfp4_moe_bf16(
        &self,
        tokens: usize,
        selected: usize,
        activation: GatedActivation,
        gate: NvFp4ExpertBank,
        up: NvFp4ExpertBank,
        down: NvFp4ExpertBank,
    ) -> Result<DirectNvFp4MoeBf16> {
        DirectNvFp4MoeBf16::new(self, tokens, selected, activation, gate, up, down)
    }
}

impl DirectNvFp4MoeBf16 {
    fn new(
        backend: &CudaBackend,
        tokens: usize,
        selected: usize,
        activation: GatedActivation,
        gate: NvFp4ExpertBank,
        up: NvFp4ExpertBank,
        down: NvFp4ExpertBank,
    ) -> Result<Self> {
        validate(&gate, &up, &down)?;
        let experts = gate.config.experts;
        let hidden = gate.config.input_features;
        let intermediate = gate.config.output_features;
        let spec = NvFp4MicroSpec {
            experts,
            selected,
            hidden,
            intermediate,
            tokens,
            activation: activation.into(),
        };
        let groups = tokens
            .checked_mul(selected)
            .ok_or(Error::InvalidNvFp4("direct micro expert count overflow"))?;
        let allocate = |elements| backend.inner.pool.allocate(&backend.inner.stream, elements);
        Ok(Self {
            kernels: NvFp4MicroKernels::compile(&backend.inner.compiler, spec)?,
            gate_packed: allocate(elements(groups, hidden / 2)?)?,
            up_packed: allocate(elements(groups, hidden / 2)?)?,
            gate_scales: allocate(elements(groups, hidden / 16)?)?,
            up_scales: allocate(elements(groups, hidden / 16)?)?,
            intermediate_packed: allocate(elements(groups, intermediate / 2)?)?,
            intermediate_scales: allocate(elements(groups, intermediate / 16)?)?,
            stream: backend.inner.stream.clone(),
            gate,
            up,
            down,
            tokens,
            selected,
            hidden,
        })
    }

    pub fn execute(
        &mut self,
        input: &DeviceBuffer<bf16>,
        selected: &DeviceBuffer<u32>,
        routing: &DeviceBuffer<bf16>,
        output: &mut DeviceBuffer<bf16>,
    ) -> Result<()> {
        self.kernels.execute(
            &self.stream,
            NvFp4MicroLaunch {
                input,
                selected,
                routing,
                banks: NvFp4MicroBanks {
                    gate: view(&self.gate),
                    up: view(&self.up),
                    down: view(&self.down),
                },
                workspace: NvFp4MicroWorkspace {
                    gate_packed: &mut self.gate_packed,
                    up_packed: &mut self.up_packed,
                    gate_scales: &mut self.gate_scales,
                    up_scales: &mut self.up_scales,
                    intermediate_packed: &mut self.intermediate_packed,
                    intermediate_scales: &mut self.intermediate_scales,
                },
                output,
            },
        )
    }

    #[must_use]
    pub fn output_elements(&self) -> usize {
        self.tokens.saturating_mul(self.hidden)
    }

    #[must_use]
    pub const fn selected_count(&self) -> usize {
        self.selected
    }
}

fn validate(gate: &NvFp4ExpertBank, up: &NvFp4ExpertBank, down: &NvFp4ExpertBank) -> Result<()> {
    let same_gate_up = gate.config == up.config;
    let transposed_down =
        (gate.config.input_features, gate.config.output_features, gate.config.experts)
            == (down.config.output_features, down.config.input_features, down.config.experts);
    let valid = same_gate_up && transposed_down;
    if valid {
        Ok(())
    } else {
        Err(Error::InvalidNvFp4("incompatible direct micro expert banks"))
    }
}

fn elements(rows: usize, columns: usize) -> Result<usize> {
    rows.checked_mul(columns)
        .ok_or(Error::InvalidNvFp4("direct micro workspace overflow"))
}