bunsen 0.0.1

burn neural network extension library
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
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402

use crate::models::swin::v2::windowing::window_partition;
use burn::prelude::{Backend, Bool, Int, Tensor};

/// Apply an attention mask.
///
/// # Arguments
///
/// - `b_nw`: Batch size times number of windows.
/// - `n`: Number of elements in the input tensor, Wh*Ww.
/// - `num_heads`: Number of attention heads.
/// - `attn`: Attention logits tensor of shape (`b_nw`, `num_heads`, Wh*Ww, Wh*Ww).
/// - `mask`: Mask tensor of shape (`num_windows`, Wh*Ww, Wh*Ww).
///
/// # Returns
///
/// - Output tensor of shape (`b_nw`, `num_heads`, Wh*Ww, Wh*Ww).
#[inline(always)]
#[must_use]
pub fn apply_attention_mask<B: Backend>(
    b_nw: usize,
    n: usize,
    num_heads: usize,
    attn: Tensor<B, 4>,
    mask: Tensor<B, 3>,
) -> Tensor<B, 4> {
    // Attention mask
    let num_windows = mask.dims()[0];
    let b = b_nw / num_windows;

    let attn = attn.reshape([b, num_windows, num_heads, n, n]);

    let mask = mask.unsqueeze_dim::<4>(1).unsqueeze::<5>();
    // 1, num_windows, 1, Wh*Ww, Wh*Ww

    let attn: Tensor<B, 5> = attn + mask;
    // b, num_windows, num_heads, Wh*Ww, Wh*Ww

    attn.reshape([-1, num_heads as i32, n as i32, n as i32])
    // b_nw, num_heads, Wh*Ww, Wh*Ww
}

/// Internal function to create a shifted window image mask.
///
/// This function generates a mask for the shifted window attention mechanism.
///
/// # Arguments
///
/// - `input_shape`: The shape of the input tensor; must be divisible by the window size.
/// - `window_size`: The size of the window.
/// - `shift_size`: The size of the shift.
/// - `device`: The device on which the tensor will be created.
///
/// # Returns
///
/// A tensor representing the shifted window image mask.
#[must_use]
fn sw_img_mask<B: Backend>(
    input_shape: [usize; 2],
    window_size: usize,
    shift_size: usize,
    device: &B::Device,
) -> Tensor<B, 2, Int> {
    let [h, w] = input_shape;

    let mut img_mask = Tensor::<B, 2, Int>::zeros([h, w], device);

    let h = h as i32;
    let w = w as i32;

    let window_size = window_size as i32;
    let shift_size = shift_size as i32;

    assert_eq!(
        h % window_size,
        0,
        "Height {h} is not divisible by window size {window_size}"
    );
    assert_eq!(
        w % window_size,
        0,
        "Width {w} is not divisible by window size {window_size}"
    );

    let h_slices = [
        0..(h - window_size) as usize,
        (h - window_size) as usize..(h - shift_size) as usize,
        (h - shift_size) as usize..h as usize,
    ];
    let w_slices = [
        0..(w - window_size) as usize,
        (w - window_size) as usize..(w - shift_size) as usize,
        (w - shift_size) as usize..w as usize,
    ];

    let mut cnt = 0;
    for h in h_slices.iter() {
        for w in w_slices.iter() {
            let slice_shape = img_mask.clone().slice([h.clone(), w.clone()]).dims();
            let val: Tensor<B, 1, Int> = Tensor::from_data([cnt], device);
            let val = val.unsqueeze::<2>().expand(slice_shape);

            img_mask = img_mask.slice_assign([h.clone(), w.clone()], val);
            cnt += 1;
        }
    }

    img_mask
}

/// Create a shifted window attention mask.
///
/// This function generates a mask for the shifted window attention mechanism.
///
/// # Arguments
///
/// - `input_shape`: The shape of the input tensor.
/// - `window_size`: The size of the window.
/// - `shift_size`: The size of the shift.
/// - `device`: The device on which the tensor will be created.
///
/// # Returns
///
/// A tensor representing the shifted window attention mask.
#[must_use]
pub fn sw_attn_mask<B: Backend>(
    input_shape: [usize; 2],
    window_size: usize,
    shift_size: usize,
    device: &B::Device,
) -> Tensor<B, 3, Bool> {
    let img_mask = sw_img_mask(input_shape, window_size, shift_size, device);
    // ws, ws
    let img_mask = img_mask.unsqueeze_dim::<3>(2).unsqueeze::<4>();
    // b_nw=1, ws, ws, 1

    let mask_windows = window_partition(img_mask, window_size);
    // b_nw=1, ws, ws, 1

    let mask_windows = mask_windows.reshape([-1, (window_size * window_size) as i32]);
    // b_nw=nW, ws * ws

    let mask =
        mask_windows.clone().unsqueeze_dim::<3>(1) - mask_windows.clone().unsqueeze_dim::<3>(2);

    mask.not_equal_elem(0)
}

#[cfg(test)]
mod tests {
    use super::*;
    use burn::backend::NdArray;
    use burn::prelude::{Tensor, TensorData, s};

    #[should_panic(expected = "Height 5 is not divisible by window size 2")]
    #[test]
    fn test_sw_img_mask_height_not_divisible() {
        let device = Default::default();
        let _d = sw_img_mask::<NdArray>([5, 4], 2, 1, &device);
    }

    #[should_panic(expected = "Width 5 is not divisible by window size 2")]
    #[test]
    fn test_sw_img_mask_width_not_divisible() {
        let device = Default::default();
        let _d = sw_img_mask::<NdArray>([4, 5], 2, 1, &device);
    }

    #[test]
    fn test_apply_attention_mask() {
        let b = 2;
        let nw = 2;
        let b_nw = b * nw;
        let ws = 2;
        let n = ws * ws;
        let num_heads = 5;

        let device = Default::default();
        let attn = Tensor::<NdArray, 4>::zeros([b_nw, num_heads, n, n], &device);
        // (b*nw, num_heads, ws*ws, ws*ws)

        // (nw, ws*ws, ws*ws)
        let mask = Tensor::<NdArray, 3>::from_data(
            [
                [
                    [0.0, 0.25, 0.5, 0.75],
                    [1.0, 1.25, 1.5, 1.75],
                    [2.0, 2.25, 2.5, 2.75],
                    [3.0, 3.25, 3.5, 3.75],
                ],
                [
                    [1.0, 0.0, 0.0, 0.0],
                    [0.0, 1.0, 0.0, 0.0],
                    [0.0, 0.0, 1.0, 0.0],
                    [0.0, 0.0, 0.0, 1.0],
                ],
            ],
            &device,
        );

        let res = apply_attention_mask(b_nw, n, num_heads, attn, mask.clone());
        assert_eq!(res.dims(), [b_nw, num_heads, n, n]);

        let res = res.reshape([b, nw, num_heads, n, n]);

        for bi in 0..b {
            for wi in 0..nw {
                let window = res
                    .clone()
                    .slice(s![bi, wi, .., ..])
                    .squeeze_dim::<4>(0)
                    .squeeze_dim::<3>(0);

                let wmask: Tensor<NdArray, 2> =
                    mask.clone().slice(s![wi, .., ..]).squeeze_dim::<2>(0);

                for hi in 0..num_heads {
                    let h_attn = window.clone().slice(s![hi, .., ..]).squeeze_dim::<2>(0);

                    h_attn.to_data().assert_eq(&wmask.to_data(), true);
                }
            }
        }
    }

    #[test]
    fn test_attn_mask() {
        // let b_nw = 1;
        let device = Default::default();

        sw_attn_mask::<NdArray>([4, 4], 2, 1, &device)
            .to_data()
            .assert_eq(
                &TensorData::from([
                    [
                        [false, false, false, false],
                        [false, false, false, false],
                        [false, false, false, false],
                        [false, false, false, false],
                    ],
                    [
                        [false, true, false, true],
                        [true, false, true, false],
                        [false, true, false, true],
                        [true, false, true, false],
                    ],
                    [
                        [false, false, true, true],
                        [false, false, true, true],
                        [true, true, false, false],
                        [true, true, false, false],
                    ],
                    [
                        [false, true, true, true],
                        [true, false, true, true],
                        [true, true, false, true],
                        [true, true, true, false],
                    ],
                ]),
                true,
            );

        sw_attn_mask::<NdArray>([6, 6], 3, 1, &device)
            .to_data()
            .assert_eq(
                &TensorData::from([
                    [
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                    ],
                    [
                        [false, false, true, false, false, true, false, false, true],
                        [false, false, true, false, false, true, false, false, true],
                        [true, true, false, true, true, false, true, true, false],
                        [false, false, true, false, false, true, false, false, true],
                        [false, false, true, false, false, true, false, false, true],
                        [true, true, false, true, true, false, true, true, false],
                        [false, false, true, false, false, true, false, false, true],
                        [false, false, true, false, false, true, false, false, true],
                        [true, true, false, true, true, false, true, true, false],
                    ],
                    [
                        [false, false, false, false, false, false, true, true, true],
                        [false, false, false, false, false, false, true, true, true],
                        [false, false, false, false, false, false, true, true, true],
                        [false, false, false, false, false, false, true, true, true],
                        [false, false, false, false, false, false, true, true, true],
                        [false, false, false, false, false, false, true, true, true],
                        [true, true, true, true, true, true, false, false, false],
                        [true, true, true, true, true, true, false, false, false],
                        [true, true, true, true, true, true, false, false, false],
                    ],
                    [
                        [false, false, true, false, false, true, true, true, true],
                        [false, false, true, false, false, true, true, true, true],
                        [true, true, false, true, true, false, true, true, true],
                        [false, false, true, false, false, true, true, true, true],
                        [false, false, true, false, false, true, true, true, true],
                        [true, true, false, true, true, false, true, true, true],
                        [true, true, true, true, true, true, false, false, true],
                        [true, true, true, true, true, true, false, false, true],
                        [true, true, true, true, true, true, true, true, false],
                    ],
                ]),
                true,
            );

        sw_attn_mask::<NdArray>([6, 6], 3, 2, &device)
            .to_data()
            .assert_eq(
                &TensorData::from([
                    [
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                        [
                            false, false, false, false, false, false, false, false, false,
                        ],
                    ],
                    [
                        [false, true, true, false, true, true, false, true, true],
                        [true, false, false, true, false, false, true, false, false],
                        [true, false, false, true, false, false, true, false, false],
                        [false, true, true, false, true, true, false, true, true],
                        [true, false, false, true, false, false, true, false, false],
                        [true, false, false, true, false, false, true, false, false],
                        [false, true, true, false, true, true, false, true, true],
                        [true, false, false, true, false, false, true, false, false],
                        [true, false, false, true, false, false, true, false, false],
                    ],
                    [
                        [false, false, false, true, true, true, true, true, true],
                        [false, false, false, true, true, true, true, true, true],
                        [false, false, false, true, true, true, true, true, true],
                        [true, true, true, false, false, false, false, false, false],
                        [true, true, true, false, false, false, false, false, false],
                        [true, true, true, false, false, false, false, false, false],
                        [true, true, true, false, false, false, false, false, false],
                        [true, true, true, false, false, false, false, false, false],
                        [true, true, true, false, false, false, false, false, false],
                    ],
                    [
                        [false, true, true, true, true, true, true, true, true],
                        [true, false, false, true, true, true, true, true, true],
                        [true, false, false, true, true, true, true, true, true],
                        [true, true, true, false, true, true, false, true, true],
                        [true, true, true, true, false, false, true, false, false],
                        [true, true, true, true, false, false, true, false, false],
                        [true, true, true, false, true, true, false, true, true],
                        [true, true, true, true, false, false, true, false, false],
                        [true, true, true, true, false, false, true, false, false],
                    ],
                ]),
                true,
            );
    }
}