File size: 25,685 Bytes
34e468d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
"""
Taken and modified from alxndrTL's othello_mamba repository:
https://github.com/alxndrTL/othello_mamba
"""

import math
import inspect
from dataclasses import dataclass
from typing import Union

import torch
import torch.nn as nn
import torch.nn.functional as F

# Official fused CUDA selective-scan kernel (mamba_ssm). Optional: imported here
# behind a guard so the module still loads (and the pure-PyTorch pscan path runs)
# when mamba_ssm is not installed. Used only when config.use_cuda=True.
try:
    from mamba_ssm.ops.selective_scan_interface import selective_scan_fn
    _HAS_SELECTIVE_SCAN = True
except Exception:  # pragma: no cover - import guard
    selective_scan_fn = None
    _HAS_SELECTIVE_SCAN = False


class PScan(torch.autograd.Function):
    @staticmethod
    def pscan(A, X):
        # A : (B, D, L, N)
        # X : (B, D, L, N)

        # modifies X in place by doing a parallel scan.
        # more formally, X will be populated by these values :
        # H[t] = A[t] * H[t-1] + X[t] with H[0] = 0
        # which are computed in parallel (2*log2(T) sequential steps (ideally), instead of T sequential steps)

        B, D, L, _ = A.size()
        num_steps = int(math.log2(L))

        # up sweep or reduction step
        Aa = A
        Xa = X
        for k in range(num_steps):
            T = 2 * (Xa.size(2) // 2)

            Aa = Aa[:, :, :T].view(B, D, T // 2, 2, -1)
            Xa = Xa[:, :, :T].view(B, D, T // 2, 2, -1)

            Xa[:, :, :, 1].add_(Aa[:, :, :, 1].mul(Xa[:, :, :, 0]))
            Aa[:, :, :, 1].mul_(Aa[:, :, :, 0])

            Aa = Aa[:, :, :, 1]
            Xa = Xa[:, :, :, 1]

        # down sweep
        for k in range(num_steps - 1, -1, -1):
            Aa = A[:, :, 2**k - 1 : L : 2**k]
            Xa = X[:, :, 2**k - 1 : L : 2**k]

            T = 2 * (Xa.size(2) // 2)

            if T < Xa.size(2):
                Xa[:, :, -1].add_(Aa[:, :, -1].mul(Xa[:, :, -2]))
                Aa[:, :, -1].mul_(Aa[:, :, -2])

            Aa = Aa[:, :, :T].view(B, D, T // 2, 2, -1)
            Xa = Xa[:, :, :T].view(B, D, T // 2, 2, -1)

            Xa[:, :, 1:, 0].add_(Aa[:, :, 1:, 0].mul(Xa[:, :, :-1, 1]))
            Aa[:, :, 1:, 0].mul_(Aa[:, :, :-1, 1])

    @staticmethod
    def forward(ctx, A_in, X_in):
        """
        Applies the parallel scan operation, as defined above. Returns a new tensor.

        Args:
            A_in : (B, L, D, N)
            X_in : (B, L, D, N)

        Returns:
            H : (B, L, D, N)
        """

        # clone tensor (in-place ops)
        A = A_in.clone()  # (B, L, D, N)
        X = X_in.clone()  # (B, L, D, N)

        # prepare tensors
        A = A.transpose(2, 1)  # (B, D, L, N)
        X = X.transpose(2, 1)  # (B, D, L, N)

        # parallel scan
        PScan.pscan(A, X)

        ctx.save_for_backward(A_in, X)

        return X.transpose(2, 1)

    @staticmethod
    def backward(ctx, grad_output_in):
        """
        Flows the gradient from the output to the input. Returns two new tensors.

        Args:
            ctx : A_in : (B, L, D, N), X : (B, D, L, N)
            grad_output_in : (B, L, D, N)

        Returns:
            gradA : (B, L, D, N), gradX : (B, L, D, N)
        """

        A_in, X = ctx.saved_tensors

        # clone tensors
        A = A_in.clone()
        # grad_output_in will be cloned with flip()

        # prepare tensors
        A = A.transpose(2, 1)  # noqa: FURB184
        A = torch.cat((A[:, :, :1], A[:, :, 1:].flip(2)), dim=2)
        grad_output_b = grad_output_in.transpose(2, 1)

        # reverse parallel scan
        grad_output_b = grad_output_b.flip(2)  # noqa: FURB184
        PScan.pscan(A, grad_output_b)
        grad_output_b = grad_output_b.flip(2)

        Q = torch.zeros_like(X)
        Q[:, :, 1:].add_(X[:, :, :-1] * grad_output_b[:, :, 1:])

        return Q.transpose(2, 1), grad_output_b.transpose(2, 1)


@dataclass
class MambaConfig:
    n_embd: int  # D
    n_layer: int
    dt_rank: Union[int, str] = "auto"
    d_state: int = 16  # N in paper/comments
    expand_factor: int = 2  # E in paper/comments
    d_conv: int = 4
    vocab_size: int = 64

    dt_min: float = 0.001
    dt_max: float = 0.1
    dt_init: str = "random"  # "random" or "constant"
    dt_scale: float = 1.0
    dt_init_floor = 1e-4

    rms_norm_eps: float = 1e-5

    bias: bool = False
    conv_bias: bool = True
    inner_layernorms: bool = False  # apply layernorms to internal activations

    pscan: bool = True  # use parallel scan mode or sequential mode when training
    use_cuda: bool = True  # use official CUDA implementation when training

    model_type: str = "mamba"  # mamba or mamba_ssm

    # For transformer
    num_states: int = 64
    num_state_dimensions: int = 1
    predict_type: str = "next_token"  # "next_token" or "state"
    pad_id: int = -1
    freeze_reps: bool = False

    def __post_init__(self):
        self.d_inner = self.expand_factor * self.n_embd  # E*D = ED in comments

        if self.dt_rank == "auto":
            self.dt_rank = math.ceil(self.n_embd / 16)


class Mamba(nn.Module):
    def __init__(self, config: MambaConfig):
        super().__init__()

        self.config = config
        self.embedding = nn.Embedding(config.vocab_size, config.n_embd, padding_idx=0)
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
        self.lm_head.weight = self.embedding.weight

        if config.model_type in ("mamba", "mamba2"):
            self.layers = nn.ModuleList(
                [ResidualBlock(config) for _ in range(config.n_layer)]
            )
            self.out_norm = RMSNorm(config.n_embd, config.rms_norm_eps)
        elif config.model_type == "lstm":
            self.layers = nn.LSTM(
                config.n_embd, config.n_embd, config.n_layer, batch_first=True
            )
        elif config.model_type == "rnn":
            self.layers = nn.RNN(
                config.n_embd, config.n_embd, config.n_layer, batch_first=True
            )
        else:
            raise ValueError("Invalid model_type")

        if config.predict_type == "state":
            self.state_predictor = nn.Linear(
                config.n_embd,
                config.num_states * config.num_state_dimensions,
                bias=True,
            )

        self.apply(self._init_weights)
        for pn, p in self.named_parameters():
            if pn.endswith(("fc_3.weight", "c_proj.weight")):
                torch.nn.init.normal_(
                    p, mean=0.0, std=0.02 / math.sqrt(2 * self.config.n_layer)
                )

        if self.config.freeze_reps:
            for name, param in self.named_parameters():
                if "lm_head" not in name and "state_predictor" not in name:
                    param.requires_grad = False

        print(f"number of parameters: {self.get_num_params() / 1e6:.2f}M")

    def forward(self, idx, targets=None):
        # x : (B, L, D)

        # y : (B, L, D)
        b, t = idx.size()
        x = self.embedding(idx)

        if self.config.model_type in ("mamba", "mamba2"):
            for layer in self.layers:
                x = layer(x)
            x = self.out_norm(x)
        elif self.config.model_type in ("lstm", "rnn"):
            x, _ = self.layers(x)

        if self.config.freeze_reps:
            x = x.detach()

        if self.config.predict_type == "state":
            logits = self.state_predictor(x)
            if self.config.num_state_dimensions > 1:
                logits = logits.view(
                    b, t, self.config.num_state_dimensions, self.config.num_states
                )
            loss = F.cross_entropy(
                logits.view(-1, logits.size(-1)),
                targets.view(-1),
                reduction="none",
            )
            mask = idx != self.config.pad_id
            if self.config.num_state_dimensions > 1:
                loss = loss.view(b, t, self.config.num_state_dimensions).sum(-1)
            else:
                loss = loss.view(b, t)
            loss = (loss * mask).sum() / mask.sum()  # mean only over unmasked elements
        else:
            if targets is not None:
                # if we are given some desired targets also calculate the loss
                logits = self.lm_head(x)
                loss = F.cross_entropy(
                    logits.view(-1, logits.size(-1)),
                    targets.view(-1),
                    ignore_index=self.config.pad_id,
                )
            else:
                # inference-time mini-optimization: only forward the lm_head on the very last position
                logits = self.lm_head(
                    x[:, [-1], :]
                )  # note: using list [-1] to preserve the time dim
                loss = None

        return logits, loss

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=None, return_confidence=False):
        """Autoregressively complete idx (B, T) by re-forwarding the full sequence each step.

        Mamba is recurrent and has no fixed context window, so no cropping is needed.
        Matches the return contract of model.transformer.GPT.generate:
            return_confidence=False -> idx
            return_confidence=True  -> (idx, confidences, top3_tokens, top3_probs)
        For B == 1 the confidence outputs are flat lists indexed by time step; for
        B > 1 they are per-sample lists of shape (B, T[, 3]).
        """
        confidences = [] if return_confidence else None
        top3_tokens = [] if return_confidence else None
        top3_probs = [] if return_confidence else None
        B = idx.size(0)

        for _ in range(max_new_tokens):
            logits, _ = self(idx)  # targets=None -> logits is (B, 1, V) for last position
            if temperature <= 0:
                # Greedy decoding (argmax); probs are the raw softmax for confidence reporting.
                probs = F.softmax(logits[:, -1, :], dim=-1)
                idx_next = probs.argmax(dim=-1, keepdim=True)  # (B, 1)
            else:
                logits = logits[:, -1, :] / temperature
                if top_k is not None:
                    v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
                    logits[logits < v[:, [-1]]] = -float('Inf')
                probs = F.softmax(logits, dim=-1)
                idx_next = torch.multinomial(probs, num_samples=1)  # (B, 1)

            if return_confidence:
                sampled_probs = probs.gather(1, idx_next).squeeze(-1)  # (B,)
                confidences.append(sampled_probs.cpu().tolist())
                top3_prob_vals, top3_token_ids = torch.topk(probs, 3, dim=-1)  # (B, 3)
                top3_tokens.append(top3_token_ids.cpu().tolist())
                top3_probs.append(top3_prob_vals.cpu().tolist())

            idx = torch.cat((idx, idx_next), dim=1)

        if return_confidence:
            if B == 1:
                return (idx,
                        [c[0] for c in confidences],
                        [t[0] for t in top3_tokens],
                        [p[0] for p in top3_probs])
            T = len(confidences)
            conf_bs = [[confidences[t][b] for t in range(T)] for b in range(B)]
            tok_bs = [[top3_tokens[t][b] for t in range(T)] for b in range(B)]
            prob_bs = [[top3_probs[t][b] for t in range(T)] for b in range(B)]
            return idx, conf_bs, tok_bs, prob_bs
        return idx

    def step(self, x, caches):
        # x : (B, L, D)
        # caches : [cache(layer) for all layers], cache : (h, inputs)

        # y : (B, L, D)
        # caches : [cache(layer) for all layers], cache : (h, inputs)

        for i, layer in enumerate(self.layers):
            x, caches[i] = layer.step(x, caches[i])

        return x, caches

    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
            if module.bias is not None:
                torch.nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)

    def configure_optimizers(self, weight_decay, learning_rate, betas, device_type):
        # start with all of the candidate parameters
        param_dict = {pn: p for pn, p in self.named_parameters()}
        # filter out those that do not require grad
        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
        # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
        # i.e. all weight tensors in matmuls + embeddings decay, all biases and layernorms don't.
        decay_params = [p for n, p in param_dict.items() if p.dim() >= 2]
        nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2]
        optim_groups = [
            {"params": decay_params, "weight_decay": weight_decay},
            {"params": nodecay_params, "weight_decay": 0.0},
        ]
        num_decay_params = sum(p.numel() for p in decay_params)
        num_nodecay_params = sum(p.numel() for p in nodecay_params)
        print(
            f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters"
        )
        print(
            f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters"
        )
        # Create AdamW optimizer and use the fused version if it is available
        fused_available = "fused" in inspect.signature(torch.optim.AdamW).parameters
        use_fused = fused_available and device_type == "cuda"
        extra_args = dict(fused=True) if use_fused else {}
        optimizer = torch.optim.AdamW(
            optim_groups, lr=learning_rate, betas=betas, **extra_args
        )
        print(f"using fused AdamW: {use_fused}")
        return optimizer

    def estimate_mfu(self, fwdbwd_per_iter, dt):
        return -1

    def get_num_params(self, non_embedding=True):
        """
        Return the number of parameters in the model.
        For non-embedding count (default), the position embeddings get subtracted.
        The token embeddings would too, except due to the parameter sharing these
        params are actually used as weights in the final layer, so we include them.
        """
        n_params = sum(p.numel() for p in self.parameters())
        if non_embedding:
            n_params -= self.embedding.weight.numel()
        return n_params


class ResidualBlock(nn.Module):
    def __init__(self, config: MambaConfig):
        super().__init__()

        if config.model_type == "mamba":
            self.mixer = MambaBlock(config)
        elif config.model_type == "mamba2":
            from mamba_ssm import Mamba2 as Mamba2SSM

            self.mixer = Mamba2SSM(
                # This module uses roughly 3 * expand * d_model^2 parameters
                d_model=config.n_embd,  # Model dimension d_model
                d_state=config.d_state,  # SSM state expansion factor
                d_conv=config.d_conv,  # Local convolution width
                expand=config.expand_factor,  # Block expansion factor
            )

        self.norm = RMSNorm(config.n_embd, config.rms_norm_eps)

    def forward(self, x):
        # x : (B, L, D)

        # output : (B, L, D)

        output = self.mixer(self.norm(x)) + x
        return output

    def step(self, x, cache):
        # x : (B, D)
        # cache : (h, inputs)
        # h : (B, ED, N)
        # inputs : (B, ED, d_conv-1)

        # output : (B, D)
        # cache : (h, inputs)

        output, cache = self.mixer.step(self.norm(x), cache)
        output = output + x
        return output, cache


class MambaBlock(nn.Module):
    def __init__(self, config: MambaConfig):
        super().__init__()

        self.config = config
        assert isinstance(config.dt_rank, int)
        assert isinstance(self.config.dt_rank, int)

        # projects block input from D to 2*ED (two branches)
        self.in_proj = nn.Linear(config.n_embd, 2 * config.d_inner, bias=config.bias)

        self.conv1d = nn.Conv1d(
            in_channels=config.d_inner,
            out_channels=config.d_inner,
            kernel_size=config.d_conv,
            bias=config.conv_bias,
            groups=config.d_inner,
            padding=config.d_conv - 1,
        )

        # projects x to input-dependent delta, B, C
        self.x_proj = nn.Linear(
            config.d_inner, config.dt_rank + 2 * config.d_state, bias=False
        )

        # projects delta from dt_rank to d_inner
        self.dt_proj = nn.Linear(config.dt_rank, config.d_inner, bias=True)

        # dt initialization
        # dt weights
        dt_init_std = config.dt_rank**-0.5 * config.dt_scale
        if config.dt_init == "constant":
            nn.init.constant_(self.dt_proj.weight, dt_init_std)
        elif config.dt_init == "random":
            nn.init.uniform_(self.dt_proj.weight, -dt_init_std, dt_init_std)
        else:
            raise NotImplementedError

        # delta bias
        dt = torch.exp(
            torch.rand(config.d_inner)
            * (math.log(config.dt_max) - math.log(config.dt_min))
            + math.log(config.dt_min)
        ).clamp(min=config.dt_init_floor)
        inv_dt = dt + torch.log(
            -torch.expm1(-dt)
        )  # inverse of softplus: https://github.com/pytorch/pytorch/issues/72759
        with torch.no_grad():
            self.dt_proj.bias.copy_(inv_dt)
        # self.dt_proj.bias._no_reinit = True # initialization would set all Linear.bias to zero, need to mark this one as _no_reinit
        # todo : explain why removed

        # S4D real initialization
        A = torch.arange(1, config.d_state + 1, dtype=torch.float32).repeat(
            config.d_inner, 1
        )
        self.A_log = nn.Parameter(
            torch.log(A)
        )  # why store A in log ? to keep A < 0 (cf -torch.exp(...)) ? for gradient stability ?
        self.A_log._no_weight_decay = True

        self.D = nn.Parameter(torch.ones(config.d_inner))

        # projects block output from ED back to D
        self.out_proj = nn.Linear(config.d_inner, config.n_embd, bias=config.bias)

        self.dt_layernorm: RMSNorm | None = None
        self.B_layernorm: RMSNorm | None = None
        self.C_layernorm: RMSNorm | None = None

        if self.config.inner_layernorms:
            self.dt_layernorm = RMSNorm(self.config.dt_rank, config.rms_norm_eps)
            self.B_layernorm = RMSNorm(self.config.d_state, config.rms_norm_eps)
            self.C_layernorm = RMSNorm(self.config.d_state, config.rms_norm_eps)

        if self.config.use_cuda:
            if not _HAS_SELECTIVE_SCAN:
                raise ImportError(
                    "config.use_cuda=True but the official mamba_ssm selective-scan "
                    "kernel is not available. Install mamba-ssm, or set use_cuda=False "
                    "to use the pure-PyTorch parallel scan.")
            self.selective_scan_cuda = selective_scan_fn

    def _apply_layernorms(self, dt, B, C):
        if self.dt_layernorm is not None:
            dt = self.dt_layernorm(dt)
        if self.B_layernorm is not None:
            B = self.B_layernorm(B)
        if self.C_layernorm is not None:
            C = self.C_layernorm(C)
        return dt, B, C

    def forward(self, x):
        # x : (B, L, D)

        # y : (B, L, D)

        _, L, _ = x.shape

        xz = self.in_proj(x)  # (B, L, 2*ED)
        x, z = xz.chunk(2, dim=-1)  # (B, L, ED), (B, L, ED)

        # x branch
        x = x.transpose(1, 2)  # (B, ED, L)
        x = self.conv1d(x)[
            :, :, :L
        ]  # depthwise convolution over time, with a short filter
        x = x.transpose(1, 2)  # noqa: FURB184

        x = F.silu(x)
        y = self.ssm(x, z)

        if self.config.use_cuda:
            output = self.out_proj(y)  # (B, L, D)
            return output

        # z branch
        z = F.silu(z)

        output = y * z
        output = self.out_proj(output)  # (B, L, D)

        return output

    def ssm(self, x, z):
        # x : (B, L, ED)

        # y : (B, L, ED)

        A = -torch.exp(self.A_log.float())  # (ED, N)
        D = self.D.float()

        deltaBC = self.x_proj(x)  # (B, L, dt_rank+2*N)
        delta, B, C = torch.split(
            deltaBC,
            [self.config.dt_rank, self.config.d_state, self.config.d_state],
            dim=-1,
        )  # (B, L, dt_rank), (B, L, N), (B, L, N)
        delta, B, C = self._apply_layernorms(delta, B, C)
        delta = self.dt_proj.weight @ delta.transpose(
            1, 2
        )  # (ED, dt_rank) @ (B, L, dt_rank) -> (B, ED, L)

        if self.config.use_cuda:
            x = x.transpose(1, 2)
            B = B.transpose(1, 2).to(x.dtype)  # NOTE: casting added by KV
            C = C.transpose(1, 2).to(x.dtype)
            z = z.transpose(1, 2).to(x.dtype)
            y = self.selective_scan_cuda(
                x,
                delta,
                A,
                B,
                C,
                D,
                z=z,
                delta_softplus=True,
                delta_bias=self.dt_proj.bias.float(),
            )
            y = y.transpose(1, 2)  # (B, L, ED)

        else:
            delta = delta.transpose(1, 2)
            delta = F.softplus(delta + self.dt_proj.bias)

            if self.config.pscan:
                y = self.selective_scan(x, delta, A, B, C, D)
            else:
                y = self.selective_scan_seq(x, delta, A, B, C, D)

        return y

    def selective_scan(self, x, delta, A, B, C, D):
        # x : (B, L, ED)
        # Δ : (B, L, ED)
        # A : (ED, N)
        # B : (B, L, N)
        # C : (B, L, N)
        # D : (ED)

        # y : (B, L, ED)

        deltaA = torch.exp(delta.unsqueeze(-1) * A)  # (B, L, ED, N)
        deltaB = delta.unsqueeze(-1) * B.unsqueeze(2)  # (B, L, ED, N)

        BX = deltaB * (x.unsqueeze(-1))  # (B, L, ED, N)

        hs = PScan.apply(deltaA, BX)

        y = (hs @ C.unsqueeze(-1)).squeeze(
            3
        )  # (B, L, ED, N) @ (B, L, N, 1) -> (B, L, ED, 1)

        y = y + D * x

        return y

    def selective_scan_seq(self, x, delta, A, B, C, D):
        # x : (B, L, ED)
        # Δ : (B, L, ED)
        # A : (ED, N)
        # B : (B, L, N)
        # C : (B, L, N)
        # D : (ED)

        # y : (B, L, ED)

        _, L, _ = x.shape

        deltaA = torch.exp(delta.unsqueeze(-1) * A)  # (B, L, ED, N)
        deltaB = delta.unsqueeze(-1) * B.unsqueeze(2)  # (B, L, ED, N)

        BX = deltaB * (x.unsqueeze(-1))  # (B, L, ED, N)

        h = torch.zeros(
            x.size(0), self.config.d_inner, self.config.d_state, device=deltaA.device
        )  # (B, ED, N)
        hs = []

        for t in range(0, L):
            h = deltaA[:, t] * h + BX[:, t]
            hs.append(h)

        hs = torch.stack(hs, dim=1)  # (B, L, ED, N)

        y = (hs @ C.unsqueeze(-1)).squeeze(
            3
        )  # (B, L, ED, N) @ (B, L, N, 1) -> (B, L, ED, 1)

        y = y + D * x

        return y

    def step(self, x, cache):
        # x : (B, D)
        # cache : (h, inputs)
        # h : (B, ED, N)
        # inputs : (B, ED, d_conv-1)

        # output : (B, D)
        # cache : (h, inputs)

        h, inputs = cache

        xz = self.in_proj(x)  # (B, 2*ED)
        x, z = xz.chunk(2, dim=1)  # (B, ED), (B, ED)

        # x branch
        x_cache = x.unsqueeze(2)
        x = self.conv1d(torch.cat([inputs, x_cache], dim=2))[
            :, :, self.config.d_conv - 1
        ]  # (B, ED)

        x = F.silu(x)
        y, h = self.ssm_step(x, h)

        # z branch
        z = F.silu(z)

        output = y * z
        output = self.out_proj(output)  # (B, D)

        # prepare cache for next call
        inputs = torch.cat([inputs[:, :, 1:], x_cache], dim=2)  # (B, ED, d_conv-1)
        cache = (h, inputs)

        return output, cache

    def ssm_step(self, x, h):
        # x : (B, ED)
        # h : (B, ED, N)

        # y : (B, ED)
        # h : (B, ED, N)

        A = -torch.exp(
            self.A_log.float()
        )  # (ED, N) # todo : ne pas le faire tout le temps, puisque c'est indépendant de la timestep
        D = self.D.float()

        deltaBC = self.x_proj(x)  # (B, dt_rank+2*N)

        delta, B, C = torch.split(
            deltaBC,
            [self.config.dt_rank, self.config.d_state, self.config.d_state],
            dim=-1,
        )  # (B, dt_rank), (B, N), (B, N)
        delta, B, C = self._apply_layernorms(delta, B, C)
        delta = F.softplus(self.dt_proj(delta))  # (B, ED)

        deltaA = torch.exp(delta.unsqueeze(-1) * A)  # (B, ED, N)
        deltaB = delta.unsqueeze(-1) * B.unsqueeze(1)  # (B, ED, N)

        BX = deltaB * (x.unsqueeze(-1))  # (B, ED, N)

        if h is None:
            h = torch.zeros(
                x.size(0),
                self.config.d_inner,
                self.config.d_state,
                device=deltaA.device,
            )  # (B, ED, N)

        h = deltaA * h + BX  # (B, ED, N)

        y = (h @ C.unsqueeze(-1)).squeeze(2)  # (B, ED, N) @ (B, N, 1) -> (B, ED, 1)

        y = y + D * x

        return y, h


# taken straight from https://github.com/johnma2006/mamba-minimal/blob/master/model.py
class RMSNorm(nn.Module):
    def __init__(self, n_embd: int, eps: float = 1e-5):
        super().__init__()

        self.eps = eps
        self.weight = nn.Parameter(torch.ones(n_embd))

    def forward(self, x):
        output = (
            x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps) * self.weight
        )

        return output