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
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
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262

//! Paged attention forward kernel launchers (standard and FP8).

use crate::error::{Error, Result};
use crate::ops::cuda::kernels::{self, PAGED_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;

use super::paged_attention::fwd_block_config;
use super::paged_decode::paged_decode_attention_fwd;

/// Standard (non-FP8) paged attention forward.
#[allow(clippy::too_many_arguments)]
pub(super) fn paged_attention_fwd_impl(
    client: &CudaClient,
    q: &Tensor<CudaRuntime>,
    k_blocks: &Tensor<CudaRuntime>,
    v_blocks: &Tensor<CudaRuntime>,
    block_table: &Tensor<CudaRuntime>,
    num_heads: usize,
    num_kv_heads: usize,
    seq_len_q: usize,
    seq_len_k: usize,
    head_dim: usize,
    block_size: usize,
    causal: bool,
) -> Result<(Tensor<CudaRuntime>, Tensor<CudaRuntime>)> {
    let q_shape = q.shape();
    if q_shape.len() != 4 {
        return Err(Error::InvalidArgument {
            arg: "q",
            reason: format!("expected 4D [B, H, S, D], got {}D", q_shape.len()),
        });
    }
    let batch_size = q_shape[0];
    let dtype = q.dtype();

    // Fast path: S_q=1 decode with specialized kernel (no shared memory tiling overhead)
    if seq_len_q == 1 && dtype == DType::F32 && (head_dim == 64 || head_dim == 128) {
        return paged_decode_attention_fwd(
            client,
            q,
            k_blocks,
            v_blocks,
            block_table,
            batch_size,
            num_heads,
            num_kv_heads,
            seq_len_k,
            head_dim,
            block_size,
        );
    }

    let dtype_suffix = match dtype {
        DType::F32 => "fp32",
        DType::F16 => "fp16",
        DType::BF16 => "bf16",
        _ => {
            return Err(Error::InvalidArgument {
                arg: "dtype",
                reason: format!(
                    "unsupported dtype {:?}. Use paged_attention_fwd_fp8 for FP8.",
                    dtype
                ),
            });
        }
    };

    let (block_m, block_n) = fwd_block_config(head_dim, dtype)?;
    let kernel_name = format!(
        "paged_flash_attention_fwd_{}_{}_small",
        head_dim, dtype_suffix
    );

    let device = q.device();
    let output =
        Tensor::<CudaRuntime>::empty(&[batch_size, num_heads, seq_len_q, head_dim], dtype, device);
    let lse = Tensor::<CudaRuntime>::empty(&[batch_size, num_heads, seq_len_q], DType::F32, device);

    let dtype_size = dtype.size_in_bytes();
    let smem_size = (block_m * head_dim + block_n * head_dim + block_n * head_dim) * dtype_size;

    let max_num_blocks = block_table.shape()[1];
    let device_index = device.id();
    let module =
        kernels::get_or_load_module(client.context(), device_index, PAGED_ATTENTION_MODULE)?;
    let func = kernels::get_kernel_function(&module, &kernel_name)?;

    let cfg = LaunchConfig {
        grid_dim: (
            (batch_size * num_heads) as u32,
            seq_len_q.div_ceil(block_m) as u32,
            1,
        ),
        block_dim: (block_m as u32, 1, 1),
        shared_mem_bytes: smem_size as u32,
    };

    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 l_ptr = lse.ptr();
    let scale = (head_dim as f32).sqrt().recip();
    let batch_i32 = batch_size as i32;
    let nh_i32 = num_heads as i32;
    let nkvh_i32 = num_kv_heads as i32;
    let sq_i32 = seq_len_q 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;
    let causal_i32 = if causal { 1i32 } else { 0i32 };

    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(&l_ptr);
        builder.arg(&batch_i32);
        builder.arg(&nh_i32);
        builder.arg(&nkvh_i32);
        builder.arg(&sq_i32);
        builder.arg(&sk_i32);
        builder.arg(&mnb_i32);
        builder.arg(&bs_i32);
        builder.arg(&scale);
        builder.arg(&causal_i32);
        builder.launch(cfg).map_err(|e| Error::KernelError {
            reason: format!("Paged attention fwd kernel launch failed: {:?}", e),
        })?;
    }

    Ok((output, lse))
}

/// FP8 paged attention forward.
#[allow(clippy::too_many_arguments)]
pub(super) fn paged_attention_fwd_fp8_impl(
    client: &CudaClient,
    q: &Tensor<CudaRuntime>,
    k_blocks: &Tensor<CudaRuntime>,
    v_blocks: &Tensor<CudaRuntime>,
    block_table: &Tensor<CudaRuntime>,
    num_heads: usize,
    num_kv_heads: usize,
    seq_len_q: usize,
    seq_len_k: usize,
    head_dim: usize,
    block_size: usize,
    causal: bool,
    q_scale: f32,
    k_scale: f32,
    v_scale: f32,
    o_scale: f32,
) -> Result<(Tensor<CudaRuntime>, Tensor<CudaRuntime>)> {
    let q_shape = q.shape();
    if q_shape.len() != 4 {
        return Err(Error::InvalidArgument {
            arg: "q",
            reason: format!("expected 4D [B, H, S, D], got {}D", q_shape.len()),
        });
    }
    let batch_size = q_shape[0];
    let dtype = q.dtype();

    let dtype_suffix = match dtype {
        DType::FP8E4M3 => "fp8_e4m3",
        DType::FP8E5M2 => "fp8_e5m2",
        _ => {
            return Err(Error::InvalidArgument {
                arg: "dtype",
                reason: format!(
                    "paged_attention_fwd_fp8 requires FP8 dtype, got {:?}",
                    dtype
                ),
            });
        }
    };

    let (block_m, _block_n) = fwd_block_config(head_dim, dtype)?;
    let kernel_name = format!(
        "paged_flash_attention_fwd_{}_{}_small",
        head_dim, dtype_suffix
    );

    let device = q.device();
    let output =
        Tensor::<CudaRuntime>::empty(&[batch_size, num_heads, seq_len_q, head_dim], dtype, device);
    let lse = Tensor::<CudaRuntime>::empty(&[batch_size, num_heads, seq_len_q], DType::F32, device);

    // FP8 uses FP32 smem: (BLOCK_M + 2*BLOCK_N) * HEAD_DIM * 4
    let (bm, bn) = fwd_block_config(head_dim, dtype)?;
    let smem_size = (bm * head_dim + 2 * bn * head_dim) * 4;

    let max_num_blocks = block_table.shape()[1];
    let device_index = device.id();
    let module =
        kernels::get_or_load_module(client.context(), device_index, PAGED_ATTENTION_MODULE)?;
    let func = kernels::get_kernel_function(&module, &kernel_name)?;

    let cfg = LaunchConfig {
        grid_dim: (
            (batch_size * num_heads) as u32,
            seq_len_q.div_ceil(block_m) as u32,
            1,
        ),
        block_dim: (block_m as u32, 1, 1),
        shared_mem_bytes: smem_size as u32,
    };

    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 l_ptr = lse.ptr();
    let attn_scale = (head_dim as f32).sqrt().recip();
    let batch_i32 = batch_size as i32;
    let nh_i32 = num_heads as i32;
    let nkvh_i32 = num_kv_heads as i32;
    let sq_i32 = seq_len_q 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;
    let causal_i32 = if causal { 1i32 } else { 0i32 };

    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(&l_ptr);
        builder.arg(&batch_i32);
        builder.arg(&nh_i32);
        builder.arg(&nkvh_i32);
        builder.arg(&sq_i32);
        builder.arg(&sk_i32);
        builder.arg(&mnb_i32);
        builder.arg(&bs_i32);
        builder.arg(&attn_scale);
        builder.arg(&q_scale);
        builder.arg(&k_scale);
        builder.arg(&v_scale);
        builder.arg(&o_scale);
        builder.arg(&causal_i32);
        builder.launch(cfg).map_err(|e| Error::KernelError {
            reason: format!("Paged attention FP8 fwd kernel launch failed: {:?}", e),
        })?;
    }

    Ok((output, lse))
}