boostr 0.1.0

ML framework built on numr - attention, quantization, model architectures
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

//! Paged decode attention — S_q=1 specialized fast path.

use crate::error::{Error, Result};
use crate::ops::cuda::kernels::{self, PAGED_DECODE_ATTENTION_MODULE};
use cudarc::driver::PushKernelArg;
use cudarc::driver::safe::LaunchConfig;
use numr::dtype::DType;
use numr::runtime::Device;
use numr::runtime::cuda::{CudaClient, CudaRuntime};
use numr::tensor::Tensor;

/// Paged decode attention — S_q=1 specialized fast path
#[allow(clippy::too_many_arguments)]
pub(super) fn paged_decode_attention_fwd(
    client: &CudaClient,
    q: &Tensor<CudaRuntime>,
    k_blocks: &Tensor<CudaRuntime>,
    v_blocks: &Tensor<CudaRuntime>,
    block_table: &Tensor<CudaRuntime>,
    batch_size: usize,
    num_heads: usize,
    num_kv_heads: usize,
    seq_len_k: usize,
    head_dim: usize,
    block_size: usize,
) -> Result<(Tensor<CudaRuntime>, Tensor<CudaRuntime>)> {
    let kernel_name = format!("paged_decode_attention_{}_fp32", head_dim);
    let device = q.device();
    let device_index = device.id();

    let module = kernels::get_or_load_module(
        client.context(),
        device_index,
        PAGED_DECODE_ATTENTION_MODULE,
    )?;
    let func = kernels::get_kernel_function(&module, &kernel_name)?;

    let output =
        Tensor::<CudaRuntime>::empty(&[batch_size, num_heads, 1, head_dim], DType::F32, device);
    // LSE not computed by decode kernel (not needed for inference)
    let lse = Tensor::<CudaRuntime>::empty(&[batch_size, num_heads, 1], DType::F32, device);

    let max_num_blocks = block_table.shape()[1];
    let scale = (head_dim as f32).sqrt().recip();

    let cfg = LaunchConfig {
        grid_dim: ((batch_size * num_heads) as u32, 1, 1),
        block_dim: (head_dim as u32, 1, 1),
        shared_mem_bytes: 0,
    };

    let q_ptr = q.ptr();
    let kb_ptr = k_blocks.ptr();
    let vb_ptr = v_blocks.ptr();
    let bt_ptr = block_table.ptr();
    let o_ptr = output.ptr();
    let nh_i32 = num_heads as i32;
    let nkvh_i32 = num_kv_heads as i32;
    let sk_i32 = seq_len_k as i32;
    let mnb_i32 = max_num_blocks as i32;
    let bs_i32 = block_size as i32;

    unsafe {
        let mut builder = client.stream().launch_builder(&func);
        builder.arg(&q_ptr);
        builder.arg(&kb_ptr);
        builder.arg(&vb_ptr);
        builder.arg(&bt_ptr);
        builder.arg(&o_ptr);
        builder.arg(&nh_i32);
        builder.arg(&nkvh_i32);
        builder.arg(&sk_i32);
        builder.arg(&mnb_i32);
        builder.arg(&bs_i32);
        builder.arg(&scale);
        builder.launch(cfg).map_err(|e| Error::KernelError {
            reason: format!("Paged decode attention kernel launch failed: {:?}", e),
        })?;
    }

    Ok((output, lse))
}

/// Paged decode attention — graph-mode variant that reads seq_len_k from device memory.
///
/// `seq_len_k_ptr` is a device pointer to an i32. Updated before each graph replay
/// via `cuMemsetD32Async`.
#[allow(clippy::too_many_arguments)]
pub fn paged_decode_attention_fwd_graph(
    client: &CudaClient,
    q: &Tensor<CudaRuntime>,
    k_blocks: &Tensor<CudaRuntime>,
    v_blocks: &Tensor<CudaRuntime>,
    block_table: &Tensor<CudaRuntime>,
    output: &Tensor<CudaRuntime>,
    batch_size: usize,
    num_heads: usize,
    num_kv_heads: usize,
    seq_len_k_ptr: u64,
    head_dim: usize,
    block_size: usize,
    max_num_blocks: usize,
) -> Result<()> {
    let kernel_name = format!("paged_decode_attention_{}_fp32_graph", head_dim);
    let device = q.device();
    let device_index = device.id();

    let module = kernels::get_or_load_module(
        client.context(),
        device_index,
        PAGED_DECODE_ATTENTION_MODULE,
    )?;
    let func = kernels::get_kernel_function(&module, &kernel_name)?;

    let scale = (head_dim as f32).sqrt().recip();

    let cfg = LaunchConfig {
        grid_dim: ((batch_size * num_heads) as u32, 1, 1),
        block_dim: (head_dim as u32, 1, 1),
        shared_mem_bytes: 0,
    };

    let q_ptr = q.ptr();
    let kb_ptr = k_blocks.ptr();
    let vb_ptr = v_blocks.ptr();
    let bt_ptr = block_table.ptr();
    let o_ptr = output.ptr();
    let nh_i32 = num_heads as i32;
    let nkvh_i32 = num_kv_heads as i32;
    let mnb_i32 = max_num_blocks as i32;
    let bs_i32 = block_size as i32;

    unsafe {
        let mut builder = client.stream().launch_builder(&func);
        builder.arg(&q_ptr);
        builder.arg(&kb_ptr);
        builder.arg(&vb_ptr);
        builder.arg(&bt_ptr);
        builder.arg(&o_ptr);
        builder.arg(&nh_i32);
        builder.arg(&nkvh_i32);
        builder.arg(&seq_len_k_ptr); // device pointer to i32
        builder.arg(&mnb_i32);
        builder.arg(&bs_i32);
        builder.arg(&scale);
        builder.launch(cfg).map_err(|e| Error::KernelError {
            reason: format!("Paged decode attention graph kernel launch failed: {:?}", e),
        })?;
    }

    Ok(())
}