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
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

//! CPU fallback for variable-length (ragged) attention
//!
//! Unpacks sequences from the packed buffer, runs standard attention per-sequence,
//! and writes results back to the packed output buffer.

use crate::error::Result;
use crate::ops::traits::VarLenAttentionOps;
use numr::runtime::cpu::{CpuClient, CpuRuntime};
use numr::tensor::Tensor;

impl VarLenAttentionOps<CpuRuntime> for CpuClient {
    fn varlen_attention_fwd(
        &self,
        q: &Tensor<CpuRuntime>,
        k: &Tensor<CpuRuntime>,
        v: &Tensor<CpuRuntime>,
        cu_seqlens_q: &Tensor<CpuRuntime>,
        cu_seqlens_k: &Tensor<CpuRuntime>,
        batch_size: usize,
        num_heads: usize,
        _max_seqlen_q: usize,
        _max_seqlen_k: usize,
        head_dim: usize,
        causal: bool,
    ) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
        let total_tokens_q = q.shape()[0];
        let device = q.device();

        let q_data = q.to_vec::<f32>();
        let k_data = k.to_vec::<f32>();
        let v_data = v.to_vec::<f32>();
        let cu_q = cu_seqlens_q.to_vec::<i32>();
        let cu_k = cu_seqlens_k.to_vec::<i32>();

        let mut out = vec![0.0f32; total_tokens_q * num_heads * head_dim];
        let mut lse = vec![0.0f32; total_tokens_q * num_heads];

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

        for b in 0..batch_size {
            let sq_start = cu_q[b] as usize;
            let sq_end = cu_q[b + 1] as usize;
            let sk_start = cu_k[b] as usize;
            let sk_end = cu_k[b + 1] as usize;
            let seq_len_q = sq_end - sq_start;
            let seq_len_k = sk_end - sk_start;

            for h in 0..num_heads {
                for qi in 0..seq_len_q {
                    let q_offset = ((sq_start + qi) * num_heads + h) * head_dim;

                    let mut max_score = f32::NEG_INFINITY;
                    let mut scores = Vec::with_capacity(seq_len_k);

                    for ki in 0..seq_len_k {
                        if causal && qi < ki {
                            scores.push(f32::NEG_INFINITY);
                            continue;
                        }
                        let k_offset = ((sk_start + ki) * num_heads + h) * head_dim;
                        let mut dot = 0.0f32;
                        for d in 0..head_dim {
                            dot += q_data[q_offset + d] * k_data[k_offset + d];
                        }
                        let s = dot * scale;
                        max_score = max_score.max(s);
                        scores.push(s);
                    }

                    let mut sum_exp = 0.0f32;
                    let mut exp_scores = Vec::with_capacity(seq_len_k);
                    for &s in &scores {
                        let e = (s - max_score).exp();
                        sum_exp += e;
                        exp_scores.push(e);
                    }

                    let o_offset = ((sq_start + qi) * num_heads + h) * head_dim;
                    for (ki, &exp_s) in exp_scores.iter().enumerate() {
                        let weight = exp_s / sum_exp;
                        let v_offset = ((sk_start + ki) * num_heads + h) * head_dim;
                        for d in 0..head_dim {
                            out[o_offset + d] += weight * v_data[v_offset + d];
                        }
                    }

                    lse[(sq_start + qi) * num_heads + h] = max_score + sum_exp.ln();
                }
            }
        }

        let output =
            Tensor::<CpuRuntime>::from_slice(&out, &[total_tokens_q, num_heads, head_dim], device);
        let lse_tensor =
            Tensor::<CpuRuntime>::from_slice(&lse, &[total_tokens_q, num_heads], device);
        Ok((output, lse_tensor))
    }

    fn varlen_attention_bwd(
        &self,
        dout: &Tensor<CpuRuntime>,
        q: &Tensor<CpuRuntime>,
        k: &Tensor<CpuRuntime>,
        v: &Tensor<CpuRuntime>,
        output: &Tensor<CpuRuntime>,
        lse: &Tensor<CpuRuntime>,
        cu_seqlens_q: &Tensor<CpuRuntime>,
        cu_seqlens_k: &Tensor<CpuRuntime>,
        batch_size: usize,
        num_heads: usize,
        _max_seqlen_q: usize,
        _max_seqlen_k: usize,
        head_dim: usize,
        causal: bool,
    ) -> Result<(Tensor<CpuRuntime>, Tensor<CpuRuntime>, Tensor<CpuRuntime>)> {
        let total_tokens_q = q.shape()[0];
        let total_tokens_k = k.shape()[0];
        let device = q.device();

        let q_data = q.to_vec::<f32>();
        let k_data = k.to_vec::<f32>();
        let v_data = v.to_vec::<f32>();
        let o_data = output.to_vec::<f32>();
        let lse_data = lse.to_vec::<f32>();
        let do_data = dout.to_vec::<f32>();
        let cu_q = cu_seqlens_q.to_vec::<i32>();
        let cu_k = cu_seqlens_k.to_vec::<i32>();

        let mut dq = vec![0.0f32; total_tokens_q * num_heads * head_dim];
        let mut dk = vec![0.0f32; total_tokens_k * num_heads * head_dim];
        let mut dv = vec![0.0f32; total_tokens_k * num_heads * head_dim];

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

        for b in 0..batch_size {
            let sq_start = cu_q[b] as usize;
            let sq_end = cu_q[b + 1] as usize;
            let sk_start = cu_k[b] as usize;
            let sk_end = cu_k[b + 1] as usize;
            let seq_len_q = sq_end - sq_start;
            let seq_len_k = sk_end - sk_start;

            for h in 0..num_heads {
                for qi in 0..seq_len_q {
                    let q_off = ((sq_start + qi) * num_heads + h) * head_dim;
                    let l = lse_data[(sq_start + qi) * num_heads + h];

                    // D = sum(dO * O)
                    let o_off = ((sq_start + qi) * num_heads + h) * head_dim;
                    let mut d_val = 0.0f32;
                    for d in 0..head_dim {
                        d_val += do_data[o_off + d] * o_data[o_off + d];
                    }

                    for ki in 0..seq_len_k {
                        if causal && qi < ki {
                            continue;
                        }
                        let k_off = ((sk_start + ki) * num_heads + h) * head_dim;

                        // Recompute score and prob
                        let mut score = 0.0f32;
                        for d in 0..head_dim {
                            score += q_data[q_off + d] * k_data[k_off + d];
                        }
                        score *= scale;
                        let prob = (score - l).exp();

                        // grad_prob = V @ dO
                        let v_off = ((sk_start + ki) * num_heads + h) * head_dim;
                        let mut grad_prob = 0.0f32;
                        for d in 0..head_dim {
                            grad_prob += v_data[v_off + d] * do_data[o_off + d];
                        }

                        let grad_score = prob * (grad_prob - d_val);

                        // Accumulate gradients
                        for d in 0..head_dim {
                            dq[q_off + d] += scale * grad_score * k_data[k_off + d];
                            dk[k_off + d] += scale * grad_score * q_data[q_off + d];
                            dv[v_off + d] += prob * do_data[o_off + d];
                        }
                    }
                }
            }
        }

        let dq_t =
            Tensor::<CpuRuntime>::from_slice(&dq, &[total_tokens_q, num_heads, head_dim], device);
        let dk_t =
            Tensor::<CpuRuntime>::from_slice(&dk, &[total_tokens_k, num_heads, head_dim], device);
        let dv_t =
            Tensor::<CpuRuntime>::from_slice(&dv, &[total_tokens_k, num_heads, head_dim], device);
        Ok((dq_t, dk_t, dv_t))
    }
}

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

    #[test]
    fn test_varlen_fwd_shape() {
        let (client, dev) = cpu_setup();

        // 2 sequences: [3, 2] tokens, 2 heads, head_dim=4
        let total_q = 5;
        let num_heads = 2;
        let head_dim = 4;

        let q_data = vec![0.1f32; total_q * num_heads * head_dim];
        let k_data = vec![0.1f32; total_q * num_heads * head_dim];
        let v_data = vec![0.2f32; total_q * num_heads * head_dim];

        let q = Tensor::<CpuRuntime>::from_slice(&q_data, &[total_q, num_heads, head_dim], &dev);
        let k = Tensor::<CpuRuntime>::from_slice(&k_data, &[total_q, num_heads, head_dim], &dev);
        let v = Tensor::<CpuRuntime>::from_slice(&v_data, &[total_q, num_heads, head_dim], &dev);

        let cu_seqlens = vec![0i32, 3, 5];
        let cu = Tensor::<CpuRuntime>::from_slice(&cu_seqlens, &[3], &dev);

        let (out, lse) = client
            .varlen_attention_fwd(&q, &k, &v, &cu, &cu, 2, num_heads, 3, 3, head_dim, false)
            .unwrap();

        assert_eq!(out.shape(), &[total_q, num_heads, head_dim]);
        assert_eq!(lse.shape(), &[total_q, num_heads]);
    }

    #[test]
    fn test_varlen_fwd_causal() {
        let (client, dev) = cpu_setup();

        // Single sequence of 4 tokens, 1 head, head_dim=2
        let total_q = 4;
        let num_heads = 1;
        let head_dim = 2;

        let q_data: Vec<f32> = (0..total_q * num_heads * head_dim)
            .map(|i| (i as f32) * 0.1 + 0.1)
            .collect();
        let k_data = q_data.clone();
        let v_data: Vec<f32> = (0..total_q * num_heads * head_dim)
            .map(|i| (i as f32) * 0.05)
            .collect();

        let q = Tensor::<CpuRuntime>::from_slice(&q_data, &[total_q, num_heads, head_dim], &dev);
        let k = Tensor::<CpuRuntime>::from_slice(&k_data, &[total_q, num_heads, head_dim], &dev);
        let v = Tensor::<CpuRuntime>::from_slice(&v_data, &[total_q, num_heads, head_dim], &dev);

        let cu_seqlens = vec![0i32, 4];
        let cu = Tensor::<CpuRuntime>::from_slice(&cu_seqlens, &[2], &dev);

        let (out_causal, _) = client
            .varlen_attention_fwd(&q, &k, &v, &cu, &cu, 1, num_heads, 4, 4, head_dim, true)
            .unwrap();
        let (out_full, _) = client
            .varlen_attention_fwd(&q, &k, &v, &cu, &cu, 1, num_heads, 4, 4, head_dim, false)
            .unwrap();

        let causal_data = out_causal.to_vec::<f32>();
        let full_data = out_full.to_vec::<f32>();

        // Last token: causal sees all 4, non-causal sees all 4 → same
        let last_off = (total_q - 1) * num_heads * head_dim;
        for d in 0..head_dim {
            assert!(
                (causal_data[last_off + d] - full_data[last_off + d]).abs() < 1e-5,
                "Last token should match between causal and non-causal"
            );
        }

        // Second token (idx=1): causal sees [0,1], non-causal sees [0,1,2,3] → different
        let second_off = num_heads * head_dim;
        let differs = (0..head_dim)
            .any(|d| (causal_data[second_off + d] - full_data[second_off + d]).abs() > 1e-6);
        assert!(
            differs,
            "Middle tokens should differ between causal and non-causal"
        );
    }

    #[test]
    fn test_varlen_bwd_shapes() {
        let (client, dev) = cpu_setup();

        let total_q = 5;
        let num_heads = 2;
        let head_dim = 4;

        let n = total_q * num_heads * head_dim;
        let q_data: Vec<f32> = (0..n).map(|i| (i as f32 * 0.3).sin()).collect();
        let k_data: Vec<f32> = (0..n).map(|i| (i as f32 * 0.7).cos()).collect();
        let v_data: Vec<f32> = (0..n).map(|i| (i as f32 * 0.5 + 1.0).sin()).collect();

        let q = Tensor::<CpuRuntime>::from_slice(&q_data, &[total_q, num_heads, head_dim], &dev);
        let k = Tensor::<CpuRuntime>::from_slice(&k_data, &[total_q, num_heads, head_dim], &dev);
        let v = Tensor::<CpuRuntime>::from_slice(&v_data, &[total_q, num_heads, head_dim], &dev);

        let cu_seqlens = vec![0i32, 3, 5];
        let cu = Tensor::<CpuRuntime>::from_slice(&cu_seqlens, &[3], &dev);

        let (out, lse) = client
            .varlen_attention_fwd(&q, &k, &v, &cu, &cu, 2, num_heads, 3, 3, head_dim, false)
            .unwrap();

        let do_data: Vec<f32> = (0..n).map(|i| (i as f32 * 0.2).cos() * 0.1).collect();
        let dout =
            Tensor::<CpuRuntime>::from_slice(&do_data, &[total_q, num_heads, head_dim], &dev);

        let (dq, dk, dv) = client
            .varlen_attention_bwd(
                &dout, &q, &k, &v, &out, &lse, &cu, &cu, 2, num_heads, 3, 3, head_dim, false,
            )
            .unwrap();

        assert_eq!(dq.shape(), &[total_q, num_heads, head_dim]);
        assert_eq!(dk.shape(), &[total_q, num_heads, head_dim]);
        assert_eq!(dv.shape(), &[total_q, num_heads, head_dim]);

        // Gradients should be non-zero
        let dq_data = dq.to_vec::<f32>();
        let has_nonzero = dq_data.iter().any(|&x: &f32| x.abs() > 1e-10);
        assert!(has_nonzero, "dQ should have non-zero gradients");
    }
}