burn_dragon_core 0.5.0

burn dragon core model and utilities
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
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
use super::*;
use crate::model::bdh_support::LanguageMhcSplitBindings;
use crate::model::residual_stream::{
    lowrank_residual_step_branch_thresholds_relu_native,
    lowrank_residual_step_next_branch_thresholds_relu_native,
};

impl<B: Backend> BDH<B> {
    pub(super) fn forward_with_state_from_embedded_single_pass(
        &self,
        embedded: Tensor<B, 3>,
        state: &mut ModelState<B>,
        start_pos: usize,
        advance_position: bool,
        position_mode: RecurrentPositionMode,
        summary_event_mask: Option<Tensor<B, 2, Int>>,
    ) -> (Tensor<B, 3>, Tensor<B, 3>) {
        let hidden = self.forward_hidden_with_state_from_embedded_single_pass(
            embedded,
            state,
            start_pos,
            advance_position,
            position_mode,
            summary_event_mask,
        );
        let logits = self.project_hidden_to_logits(hidden.clone());
        (hidden, logits)
    }

    pub(super) fn forward_hidden_with_state_from_embedded_single_pass(
        &self,
        embedded: Tensor<B, 3>,
        state: &mut ModelState<B>,
        start_pos: usize,
        advance_position: bool,
        position_mode: RecurrentPositionMode,
        summary_event_mask: Option<Tensor<B, 2, Int>>,
    ) -> Tensor<B, 3> {
        self.forward_hidden_with_state_from_embedded_single_pass_layer_limit(
            embedded,
            state,
            start_pos,
            advance_position,
            position_mode,
            summary_event_mask,
            self.n_layer,
        )
    }

    pub(super) fn initialize_language_pipeline_state(
        &self,
        embedded: Tensor<B, 3>,
    ) -> LanguagePipelineState<B> {
        let [batch, time, embd] = embedded.shape().dims::<3>();
        let current = self.norm.forward(embedded.reshape([batch, 1, time, embd]));
        LanguagePipelineState {
            current: current.clone(),
            residual_history: self.initialize_language_residual_history(&current),
        }
    }

    pub(super) fn forward_language_pipeline_state_layer_range(
        &self,
        mut pipeline_state: LanguagePipelineState<B>,
        state: &mut ModelState<B>,
        start_pos: usize,
        position_mode: RecurrentPositionMode,
        summary_event_mask: Option<Tensor<B, 2, Int>>,
        layer_range: Range<usize>,
    ) -> LanguagePipelineState<B> {
        assert!(
            !self.y_neuron_recurrence.enabled,
            "layer-range pipeline execution is not supported with y-neuron recurrence enabled"
        );

        assert_eq!(
            state.layers.len(),
            self.n_layer,
            "model state layers mismatch"
        );
        let fused = self.kernel.enabled;
        let static_mhc_coefficients = self.mhc_shared.as_ref().and_then(|mhc| {
            (!mhc.coefficient_policy().uses_dynamic_stream_controller()).then(|| mhc.coefficients())
        });
        let layer_end = layer_range.end.min(self.n_layer);

        for layer_idx in layer_range.start.min(layer_end)..layer_end {
            let layer_state = &mut state.layers[layer_idx];
            let connector = self.residual_connector_for_layer(layer_idx);
            let current_before = pipeline_state
                .residual_history
                .capture_previous(&pipeline_state.current);
            let mhc_coefficients = match connector {
                ResidualConnectorRef::Mhc(_) => static_mhc_coefficients.as_ref(),
                ResidualConnectorRef::Vanilla
                | ResidualConnectorRef::AttentionResidual(_)
                | ResidualConnectorRef::BlockAttentionResidual(_) => None,
            };
            let bindings = self.split_language_residuals_for_layer(
                pipeline_state.current,
                &connector,
                pipeline_state.residual_history.as_slice(),
                mhc_coefficients,
            );
            let LanguageMhcSplitBindings {
                branch_input: split_branch_input,
                merge: merge_bindings,
            } = bindings;
            let branch_input = if self.summary_memory_applies_to_layer(layer_idx) {
                self.forward_branch_summary_memory(
                    split_branch_input,
                    layer_state,
                    start_pos,
                    summary_event_mask.clone(),
                )
            } else {
                layer_state.summary_memory_hidden = None;
                split_branch_input
            };

            if self.clocked_slow_memory_applies_to_layer(layer_idx) {
                let branch_out = self.forward_branch_clocked_slow_layer(
                    layer_idx,
                    branch_input,
                    layer_state,
                    start_pos,
                    position_mode,
                );
                let next = self.merge_language_residuals_for_layer(
                    branch_out,
                    merge_bindings,
                    &connector,
                    mhc_coefficients,
                );
                pipeline_state.current =
                    if self.residual_connector_needs_post_merge_norm(&connector) {
                        self.norm.forward(next)
                    } else {
                        next
                    };
                self.update_language_residual_history(
                    &mut pipeline_state.residual_history,
                    current_before,
                    &pipeline_state.current,
                );
                continue;
            }
            layer_state.clocked_slow_hidden = None;

            let [branch_batch, branch_views, branch_time, branch_dim] =
                branch_input.shape().dims::<4>();
            let branch_flat =
                branch_input.reshape([branch_batch * branch_views, 1, branch_time, branch_dim]);
            let (encoder, encoder_v, decoder, latent) = self.layer_lowrank_weights(layer_idx);
            let latent_pattern = &self.kernel.block_sparse.latent;
            let sparse_mask = if fused && latent_pattern.is_sparse() {
                Some(latent_pattern.mask::<B>(latent, &branch_flat.device()))
            } else {
                None
            };
            let fused_recurrent_plan = if matches!(
                (
                    self.sequence_kernel.memory_system,
                    self.sequence_kernel.executor,
                ),
                (
                    SequenceMemorySystem::LinearAttention,
                    SequenceTrainingExecutor::Reference,
                )
            ) && self.kernel.enabled
                && self.kernel.wgpu_recurrent_kernel
                && supports_recurrent_backend::<B>()
            {
                Some(CompiledRecurrentAttentionPlan::new(
                    branch_batch * branch_views,
                    self.n_head,
                    1,
                    branch_time,
                    latent,
                    branch_dim,
                    &branch_flat.device(),
                ))
            } else {
                None
            };
            let rwkv_decay = matches!(
                self.sequence_kernel.memory_system,
                SequenceMemorySystem::Rwkv8StateSpace
            )
            .then(|| self.rwkv_decay(latent));
            let shared_lowrank_cbp_runtime = self.shared_lowrank_continual_backprop_runtime();
            let should_capture_shared_lowrank_cbp = shared_lowrank_cbp_runtime
                .map(|runtime| runtime.should_sample_step())
                .unwrap_or(false);
            let output = (cfg!(any(feature = "viz", feature = "probe"))
                || should_capture_shared_lowrank_cbp)
                .then(|| {
                    lowrank_residual_step_branch_thresholds_relu_native(
                        branch_flat.clone(),
                        encoder.clone(),
                        encoder_v.clone(),
                        decoder.clone(),
                        &self.dropout,
                        fused && self.kernel.projection_executor.use_x(),
                        fused && self.kernel.projection_executor.use_y(),
                        self.x_relu_threshold,
                        self.y_relu_threshold,
                        true,
                        self.low_bit_projection_plan(),
                        self.low_bit_quant.saved_activations.clone(),
                        self.packed_low_bit_projection_artifacts(),
                        latent_pattern,
                        self.kernel.lowrank_grad_input_executor,
                        sparse_mask.clone(),
                        |query, value| {
                            if let Some(decay) = rwkv_decay.as_ref() {
                                self.recurrent_rwkv8_with_decay(
                                    query,
                                    value,
                                    layer_state,
                                    decay.clone(),
                                )
                            } else {
                                self.recurrent_attention_with_plan(
                                    query,
                                    value,
                                    layer_state,
                                    start_pos,
                                    position_mode,
                                    fused_recurrent_plan.as_ref(),
                                )
                            }
                        },
                        |values| activation::relu(values),
                        |values| self.norm.forward(values),
                    )
                });
            if should_capture_shared_lowrank_cbp {
                if let (Some(runtime), Some(output)) =
                    (shared_lowrank_cbp_runtime.as_ref(), output.as_ref())
                {
                    runtime.record_y_neuron_stats(output.y_neuron.clone());
                }
            }
            let branch_out = if let Some(output) = output.as_ref() {
                output
                    .next
                    .clone()
                    .reshape([branch_batch, branch_views, branch_time, branch_dim])
            } else {
                lowrank_residual_step_next_branch_thresholds_relu_native(
                    branch_flat,
                    encoder.clone(),
                    encoder_v.clone(),
                    decoder.clone(),
                    &self.dropout,
                    fused && self.kernel.projection_executor.use_x(),
                    fused && self.kernel.projection_executor.use_y(),
                    self.x_relu_threshold,
                    self.y_relu_threshold,
                    true,
                    self.low_bit_projection_plan(),
                    self.low_bit_quant.saved_activations.clone(),
                    self.packed_low_bit_projection_artifacts(),
                    latent_pattern,
                    self.kernel.lowrank_grad_input_executor,
                    sparse_mask.clone(),
                    |query, value| {
                        if let Some(decay) = rwkv_decay.as_ref() {
                            self.recurrent_rwkv8_with_decay(
                                query,
                                value,
                                layer_state,
                                decay.clone(),
                            )
                        } else {
                            self.recurrent_attention_with_plan(
                                query,
                                value,
                                layer_state,
                                start_pos,
                                position_mode,
                                fused_recurrent_plan.as_ref(),
                            )
                        }
                    },
                    |values| activation::relu(values),
                    |values| self.norm.forward(values),
                )
                .reshape([branch_batch, branch_views, branch_time, branch_dim])
            };

            #[cfg(any(feature = "viz", feature = "probe"))]
            let mixed = output
                .as_ref()
                .expect("viz/probe path should retain full residual output")
                .y_neuron
                .clone()
                .swap_dims(1, 2);
            #[cfg(any(feature = "viz", feature = "probe"))]
            let [flat_batch, time, heads, latent] = mixed.shape().dims();

            #[cfg(any(feature = "viz", feature = "probe"))]
            if time > 0 {
                let last = time - 1;
                let viz_batch = branch_batch.max(1);
                let viz_views = branch_views.max(1);
                let x_neuron_last = output
                    .as_ref()
                    .expect("viz/probe path should retain full residual output")
                    .x_neuron
                    .clone()
                    .slice_dim(2, last..time)
                    .reshape([viz_batch, viz_views, heads, latent])
                    .mean_dim(1)
                    .slice_dim(0, 0..1)
                    .reshape([heads, latent]);
                let y_gate_last = output
                    .as_ref()
                    .expect("viz/probe path should retain full residual output")
                    .y_gate
                    .clone()
                    .slice_dim(2, last..time)
                    .reshape([viz_batch, viz_views, heads, latent])
                    .mean_dim(1)
                    .slice_dim(0, 0..1)
                    .reshape([heads, latent]);
                let y_neuron_last = output
                    .as_ref()
                    .expect("viz/probe path should retain full residual output")
                    .y_neuron
                    .clone()
                    .slice_dim(2, last..time)
                    .reshape([viz_batch, viz_views, heads, latent])
                    .mean_dim(1)
                    .slice_dim(0, 0..1)
                    .reshape([heads, latent]);
                let device = x_neuron_last.device();
                let rho_last = match self.resolve_linear_attention_rho_state(layer_state, &device) {
                    Some(rho) => {
                        let dims = rho.shape().dims::<4>();
                        if dims == [flat_batch, heads, latent, self.n_embd] {
                            let rho_energy =
                                rho.clone().abs().sum_dim(3).div_scalar(self.n_embd as f32);
                            let rho_energy = rho_energy
                                .reshape([viz_batch, viz_views, heads, latent])
                                .mean_dim(1)
                                .sum_dim(0)
                                .div_scalar(viz_batch as f32);
                            rho_energy.reshape([heads, latent])
                        } else {
                            Tensor::<B, 2>::zeros([heads, latent], &device)
                        }
                    }
                    None => Tensor::<B, 2>::zeros([heads, latent], &device),
                };

                layer_state.viz = Some(LayerVizState {
                    x_neuron_last,
                    y_gate_last,
                    y_neuron_last,
                    rho_last,
                });
            }

            #[cfg(any(feature = "viz", feature = "probe"))]
            let branch_out = output
                .as_ref()
                .expect("viz/probe path should retain full residual output")
                .next
                .clone()
                .reshape([branch_batch, branch_views, branch_time, branch_dim]);
            let next = self.merge_language_residuals_for_layer(
                branch_out,
                merge_bindings,
                &connector,
                mhc_coefficients,
            );
            pipeline_state.current = if self.residual_connector_needs_post_merge_norm(&connector) {
                self.norm.forward(next)
            } else {
                next
            };
            self.update_language_residual_history(
                &mut pipeline_state.residual_history,
                current_before,
                &pipeline_state.current,
            );
        }

        pipeline_state
    }

    pub(super) fn forward_hidden_with_state_from_embedded_single_pass_layer_limit(
        &self,
        embedded: Tensor<B, 3>,
        state: &mut ModelState<B>,
        start_pos: usize,
        advance_position: bool,
        position_mode: RecurrentPositionMode,
        summary_event_mask: Option<Tensor<B, 2, Int>>,
        layer_limit: usize,
    ) -> Tensor<B, 3> {
        if self.y_neuron_recurrence.enabled {
            assert_eq!(
                layer_limit, self.n_layer,
                "layer-limited profiling is not supported with y-neuron recurrence enabled"
            );
            return self.forward_hidden_with_state_from_embedded_single_pass_y_neuron_recurrence(
                embedded,
                state,
                start_pos,
                advance_position,
                position_mode,
            );
        }
        let pipeline_state = self.initialize_language_pipeline_state(embedded);
        let pipeline_state = self.forward_language_pipeline_state_layer_range(
            pipeline_state,
            state,
            start_pos,
            position_mode,
            summary_event_mask,
            0..layer_limit.min(self.n_layer),
        );
        let hidden = self.collapse_language_streams(pipeline_state.current);
        let [_batch, time, _dim] = hidden.shape().dims::<3>();
        if advance_position {
            state.position = state.position.saturating_add(time);
        }

        hidden
    }
}