File size: 30,168 Bytes
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
14f6f8d
c68c1b8
 
14f6f8d
 
 
c68c1b8
 
 
14f6f8d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c68c1b8
 
14f6f8d
ad95cec
 
14f6f8d
 
 
ad95cec
 
 
 
14f6f8d
 
ad95cec
 
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
 
 
c68c1b8
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
ad95cec
c68c1b8
 
 
 
 
 
 
ad95cec
14f6f8d
c68c1b8
 
 
14f6f8d
 
ad95cec
 
 
 
14f6f8d
c68c1b8
 
 
14f6f8d
 
c68c1b8
 
ad95cec
c68c1b8
 
 
 
14f6f8d
c68c1b8
14f6f8d
ad95cec
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
 
 
 
 
c68c1b8
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
ad95cec
 
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
14f6f8d
c68c1b8
 
14f6f8d
 
c68c1b8
de75931
 
 
 
 
 
 
14f6f8d
de75931
 
 
 
 
 
 
 
 
 
 
14f6f8d
 
 
 
 
de75931
14f6f8d
 
 
 
 
 
 
 
de75931
 
14f6f8d
 
de75931
14f6f8d
de75931
 
14f6f8d
 
de75931
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c68c1b8
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
14f6f8d
 
c68c1b8
 
14f6f8d
 
c68c1b8
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
 
c68c1b8
 
14f6f8d
 
c68c1b8
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
14f6f8d
 
 
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
 
 
 
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
14f6f8d
 
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
14f6f8d
 
c68c1b8
 
14f6f8d
 
 
c68c1b8
 
 
 
 
 
 
 
 
14f6f8d
 
 
 
 
 
c68c1b8
 
14f6f8d
c68c1b8
 
 
 
 
14f6f8d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c68c1b8
14f6f8d
c68c1b8
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
14f6f8d
 
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
14f6f8d
 
 
c68c1b8
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f6f8d
c68c1b8
14f6f8d
c68c1b8
 
 
14f6f8d
 
c68c1b8
 
 
 
 
14f6f8d
c68c1b8
 
 
14f6f8d
c68c1b8
 
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
# coding=utf-8
# Copyright 2023 Bo Peng and HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""PyTorch RWKV5 World model."""

from dataclasses import dataclass
from typing import List, Optional, Tuple, Union

import torch
import torch.nn.functional as F
import torch.utils.checkpoint
from torch import nn

from transformers.modeling_utils import PreTrainedModel
from transformers.utils import (
    ModelOutput,
    add_code_sample_docstrings,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    logging,
)

from .configuration_rwkv5 import Rwkv5Config


logger = logging.get_logger(__name__)

_CHECKPOINT_FOR_DOC = "RWKV/rwkv-5-world-1b5"
_CONFIG_FOR_DOC = "Rwkv5Config"

RWKV5_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "RWKV/rwkv-5-world-1b5",
    # See all RWKV models at https://huggingface.co/models?filter=rwkv
]


def rwkv_linear_attention_v5(
    B,
    H,
    S,
    T,
    n_head,
    hidden,
    time_decay,
    time_first,
    receptance,
    key,
    value,
    gate,
    lxw,
    lxb,
    ow,
    state,
    return_state=False,
    seq_mode=True,
):
    time_decay = torch.exp(-torch.exp(time_decay.float())).reshape(-1, 1, 1).reshape(n_head, -1, 1)
    time_first = time_first.float().reshape(-1, 1, 1).reshape(n_head, -1, 1)
    lxw = lxw.float()
    lxb = lxb.float()
    # if seq_mode:
    out = torch.empty((B, T, H, S), dtype=receptance.dtype, device=receptance.device)
    for t in range(T):
        rt = receptance[:, :, t : t + 1, :]
        kt = key[:, :, :, t : t + 1]
        vt = value[:, :, t : t + 1, :]
        at = kt @ vt
        out[:, t] = (rt @ (time_first * at + state)).squeeze(2)
        state = at + time_decay * state

    out = out.reshape(B * T, H * S)
    out = F.group_norm(out, num_groups=H, weight=lxw, bias=lxb).reshape(B, T, H * S)
    out = out.to(dtype=hidden.dtype) * gate
    out = out @ ow

    return out, state


class RwkvSelfAttention(nn.Module):
    def __init__(self, config, layer_id=0):
        super().__init__()
        self.config = config
        self.layer_id = layer_id
        hidden_size = config.hidden_size
        # https://github.com/BlinkDL/RWKV-LM/blob/main/RWKV-v4neo/src/model.py#L146
        num_attention_heads = hidden_size // config.head_size
        self.num_attention_heads = num_attention_heads
        attention_hidden_size = (
            config.attention_hidden_size if config.attention_hidden_size is not None else hidden_size
        )
        self.attention_hidden_size = attention_hidden_size

        self.time_decay = nn.Parameter(torch.empty(num_attention_heads, config.head_size))
        self.time_faaaa = nn.Parameter(torch.empty(num_attention_heads, config.head_size))
        self.time_mix_gate = nn.Parameter(torch.empty(1, 1, hidden_size))

        self.time_mix_key = nn.Parameter(torch.empty(1, 1, hidden_size))
        self.time_mix_value = nn.Parameter(torch.empty(1, 1, hidden_size))
        self.time_mix_receptance = nn.Parameter(torch.empty(1, 1, hidden_size))

        self.time_shift = nn.ZeroPad2d((0, 0, 1, -1))
        self.key = nn.Linear(hidden_size, attention_hidden_size, bias=False)
        self.value = nn.Linear(hidden_size, attention_hidden_size, bias=False)
        self.receptance = nn.Linear(hidden_size, attention_hidden_size, bias=False)
        self.gate = nn.Linear(hidden_size, attention_hidden_size, bias=False)
        self.output = nn.Linear(attention_hidden_size, hidden_size, bias=False)
        # https://github.com/BlinkDL/RWKV-LM/blob/3db37a72356b736966ddd377268f02b80963af3f/RWKV-v4neo/src/model.py#L190C1-L190C1
        self.ln_x = nn.GroupNorm(hidden_size // config.head_size, hidden_size)

    # TODO: maybe jit, otherwise move inside forward
    def extract_key_value(self, B, H, S, T, hidden, state=None):
        # Mix hidden with the previous timestep to produce key, value, receptance
        if hidden.size(1) == 1 and state is not None:
            shifted = state[0][:, :, self.layer_id]
        else:
            shifted = self.time_shift(hidden)
            if state is not None:
                shifted[:, 0] = state[0][:, :, self.layer_id]
        if len(shifted.size()) == 2:
            shifted = shifted.unsqueeze(1)
        key = hidden * self.time_mix_key + shifted * (1 - self.time_mix_key)
        value = hidden * self.time_mix_value + shifted * (1 - self.time_mix_value)
        receptance = hidden * self.time_mix_receptance + shifted * (1 - self.time_mix_receptance)
        gate = hidden * self.time_mix_gate + shifted * (1 - self.time_mix_gate)

        # https://github.com/BlinkDL/ChatRWKV/blob/main/rwkv_pip_package/src/rwkv/model.py#L693
        key = self.key(key).to(torch.float32).view(B, T, H, S).transpose(1, 2).transpose(-2, -1)
        value = self.value(value).to(torch.float32).view(B, T, H, S).transpose(1, 2)
        receptance = self.receptance(receptance).to(torch.float32).view(B, T, H, S).transpose(1, 2)
        gate = F.silu(self.gate(gate))

        if state is not None:
            state[0][:, :, self.layer_id] = hidden[:, -1]

        return receptance, key, value, gate, state

    def forward(self, hidden, state=None, use_cache=False, seq_mode=True):
        B = hidden.shape[0]
        H = self.time_decay.shape[0]
        S = hidden.shape[-1] // H
        T = hidden.shape[1]

        receptance, key, value, gate, state = self.extract_key_value(B, H, S, T, hidden, state=state)
        layer_state = state[1][:, :, :, :, self.layer_id] if state is not None else None
        rwkv, layer_state = rwkv_linear_attention_v5(
            B,
            H,
            S,
            T,
            self.num_attention_heads,
            hidden,
            self.time_decay,
            self.time_faaaa,
            receptance,
            key,
            value,
            gate,
            self.ln_x.weight,
            self.ln_x.bias,
            self.output.weight.t(),
            state=layer_state,
            return_state=use_cache,
            seq_mode=seq_mode,
        )

        if layer_state is not None:
            state[1][:, :, :, :, self.layer_id] = layer_state

        return rwkv, state


class RwkvFeedForward(nn.Module):
    def __init__(self, config, layer_id=0):
        super().__init__()
        self.config = config
        self.layer_id = layer_id
        hidden_size = config.hidden_size
        # https://github.com/BlinkDL/RWKV-LM/blob/3db37a72356b736966ddd377268f02b80963af3f/RWKV-v4neo/train.py#L168
        intermediate_size = (
            config.intermediate_size
            if config.intermediate_size is not None
            else int((config.hidden_size * 3.5) // 32 * 32)
        )

        self.time_shift = nn.ZeroPad2d((0, 0, 1, -1))
        self.time_mix_key = nn.Parameter(torch.empty(1, 1, hidden_size))
        self.time_mix_receptance = nn.Parameter(torch.empty(1, 1, hidden_size))

        self.key = nn.Linear(hidden_size, intermediate_size, bias=False)
        self.receptance = nn.Linear(hidden_size, hidden_size, bias=False)
        self.value = nn.Linear(intermediate_size, hidden_size, bias=False)

    def forward(self, hidden, state=None):
        if hidden.size(1) == 1 and state is not None:
            shifted = state[2][:, :, self.layer_id]
        else:
            shifted = self.time_shift(hidden)
            if state is not None:
                shifted[:, 0] = state[2][:, :, self.layer_id]
        if len(shifted.size()) == 2:
            shifted = shifted.unsqueeze(1)
        key = hidden * self.time_mix_key + shifted * (1 - self.time_mix_key)
        receptance = hidden * self.time_mix_receptance + shifted * (1 - self.time_mix_receptance)

        key = torch.square(torch.relu(self.key(key)))
        value = self.value(key)
        receptance = torch.sigmoid(self.receptance(receptance))

        if state is not None:
            state[2][:, :, self.layer_id] = hidden[:, -1]

        return receptance * value, state


class RwkvBlock(nn.Module):
    def __init__(self, config, layer_id):
        super().__init__()
        self.config = config
        self.layer_id = layer_id

        if layer_id == 0:
            self.pre_ln = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_epsilon)

        self.ln1 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_epsilon)
        self.ln2 = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_epsilon)

        self.attention = RwkvSelfAttention(config, layer_id)
        self.feed_forward = RwkvFeedForward(config, layer_id)

    def forward(self, hidden, state=None, use_cache=False, output_attentions=False, seq_mode=True):
        attention, state = self.attention(self.ln1(hidden), state=state, use_cache=use_cache, seq_mode=seq_mode)
        hidden = hidden + attention

        feed_forward, state = self.feed_forward(self.ln2(hidden), state=state)
        hidden = hidden + feed_forward

        outputs = (hidden, state)
        if output_attentions:
            outputs += (attention,)
        else:
            outputs += (None,)

        return outputs


class Rwkv5PreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = Rwkv5Config
    base_model_prefix = "rwkv"
    _no_split_modules = ["RwkvBlock"]
    _keep_in_fp32_modules = ["time_decay", "time_first"]
    supports_gradient_checkpointing = True
    training = False

    def _init_weights(self, module):
        """Initialize the weights."""
        if isinstance(module, RwkvSelfAttention):
            layer_id = module.layer_id
            num_hidden_layers = module.config.num_hidden_layers
            hidden_size = module.config.hidden_size
            attention_hidden_size = module.attention_hidden_size
            num_attention_heads = hidden_size // module.config.num_attention_heads

            ratio_0_to_1 = layer_id / (num_hidden_layers - 1)  # 0 to 1
            ratio_1_to_almost0 = 1.0 - (layer_id / num_hidden_layers)  # 1 to ~0

            time_weight = torch.tensor(
                [i / hidden_size for i in range(hidden_size)],
                dtype=module.time_mix_key.dtype,
                device=module.time_mix_key.device,
            )
            time_weight = time_weight[None, None, :]

            # https://github.com/BlinkDL/RWKV-LM/blob/main/RWKV-v4neo/src/model.py#L398
            decay_speed = [
                -6.0 + 5.0 * (h / (attention_hidden_size - 1)) ** (0.7 + 1.3 * ratio_0_to_1)
                for h in range(attention_hidden_size)
            ]
            decay_speed = torch.tensor(decay_speed, dtype=module.time_decay.dtype, device=module.time_decay.device)
            tmp = torch.tensor(
                [
                    (1.0 - (i / (attention_hidden_size - 1.0))) * ratio_0_to_1 + 0.1 * ((i + 1) % 3 - 1)
                    for i in range(attention_hidden_size)
                ],
                dtype=module.time_faaaa.dtype,
                device=module.time_faaaa.device,
            )

            with torch.no_grad():
                module.time_decay.data = decay_speed.reshape(num_attention_heads, module.config.num_attention_heads)
                module.time_faaaa.data = tmp.reshape(num_attention_heads, module.config.num_attention_heads)
                module.time_mix_key.data = torch.pow(time_weight, ratio_1_to_almost0)

                module.time_mix_value.data = torch.pow(time_weight, ratio_1_to_almost0) + 0.3 * ratio_0_to_1
                module.time_mix_receptance.data = torch.pow(time_weight, 0.5 * ratio_1_to_almost0)
                module.time_mix_gate.data = torch.pow(time_weight, 0.5 * ratio_1_to_almost0)

        elif isinstance(module, RwkvFeedForward):
            layer_id = module.layer_id
            num_hidden_layers = module.config.num_hidden_layers
            hidden_size = module.config.hidden_size

            ratio_1_to_almost0 = 1.0 - (layer_id / num_hidden_layers)  # 1 to ~0

            time_weight = torch.tensor(
                [i / hidden_size for i in range(hidden_size)],
                dtype=module.time_mix_key.dtype,
                device=module.time_mix_key.device,
            )
            time_weight = time_weight[None, None, :]

            with torch.no_grad():
                module.time_mix_key.data = torch.pow(time_weight, ratio_1_to_almost0)
                module.time_mix_receptance.data = torch.pow(time_weight, ratio_1_to_almost0)


@dataclass
class Rwkv5Output(ModelOutput):
    """
    Class for the RWKV model outputs.

    Args:
        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        state (list of five `torch.FloatTensor` of shape `(batch_size, hidden_size, num_hidden_layers)`):
            The state of the model at the last time step. Can be used in a forward method with the next `input_ids` to
            avoid providing the old `input_ids`.
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of
            the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in
            the self-attention heads.
    """

    last_hidden_state: torch.FloatTensor = None
    state: Optional[List[torch.FloatTensor]] = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None


@dataclass
class Rwkv5CausalLMOutput(ModelOutput):
    """
    Base class for causal language model (or autoregressive) outputs.

    Args:
        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Language modeling loss (for next-token prediction).
        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        state (list of five `torch.FloatTensor` of shape `(batch_size, hidden_size, num_hidden_layers)`):
            The state of the model at the last time step. Can be used in a forward method with the next `input_ids` to
            avoid providing the old `input_ids`.
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of
            the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in
            the self-attention heads.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: torch.FloatTensor = None
    state: Optional[List[torch.FloatTensor]] = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None


RWKV_START_DOCSTRING = r"""
    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
    etc.) This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)
    subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to
    general usage and behavior.

    Parameters:
        config ([`Rwkv5Config`]): Model configuration class with all the parameters of the model.
            Initializing with a config file does not load the weights associated with the model, only the
            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
"""

RWKV_INPUTS_DOCSTRING = r"""
    Args:
        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
            `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input
            sequence tokens in the vocabulary. If `past_key_values` is used, only `input_ids` that do not have their
            past calculated should be passed as `input_ids`. Indices can be obtained using [`AutoTokenizer`]. See
            [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input
            IDs?](../glossary#input-ids)
        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
            model's internal embedding lookup matrix.
        state (tuple of five `torch.FloatTensor` of shape `(batch_size, hidden_size, num_hidden_layers)`, *optional*):
            If passed along, the model uses the previous state in all the blocks (which will give the output for the
            `input_ids` provided as if the model add `state_input_ids + input_ids` as context).
        use_cache (`bool`, *optional*):
            If set to `True`, the last state is returned and can be used to quickly generate the next logits.
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        return_dict (`bool`, *optional*):
            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
"""


@add_start_docstrings(
    "The bare RWKV Model transformer outputting raw hidden-states without any specific head on top.",
    RWKV_START_DOCSTRING,
)
class Rwkv5Model(Rwkv5PreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size)
        self.blocks = nn.ModuleList([RwkvBlock(config, layer_id=idx) for idx in range(config.num_hidden_layers)])
        self.ln_out = nn.LayerNorm(config.hidden_size)

        self.layers_are_rescaled = False
        self.pre_ln_flag = False

        self.gradient_checkpointing = False

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embeddings

    def set_input_embeddings(self, new_embeddings):
        self.embeddings = new_embeddings

    @add_start_docstrings_to_model_forward(RWKV_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=Rwkv5Output,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.LongTensor] = None,  # noqa
        inputs_embeds: Optional[torch.FloatTensor] = None,
        state: Optional[List[torch.FloatTensor]] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, Rwkv5Output]:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        # rwkv5 only support inference in huggingface.
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if self.training == self.layers_are_rescaled and (
            self.embeddings.weight.dtype == torch.float16 or self.embeddings.weight.dtype == torch.bfloat16
        ):
            self._rescale_layers()

        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is None and inputs_embeds is None:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        if inputs_embeds is None:
            if not self.pre_ln_flag:
                normalized_weight = F.layer_norm(
                    self.embeddings.weight,
                    (self.config.hidden_size,),
                    weight=self.blocks[0].pre_ln.weight,
                    bias=self.blocks[0].pre_ln.bias,
                )
                self.embeddings.weight = nn.Parameter(normalized_weight)
                self.pre_ln_flag = True

            inputs_embeds = self.embeddings(input_ids)

        if use_cache and state is None:
            # https://github.com/BlinkDL/ChatRWKV/blob/main/rwkv_pip_package/src/rwkv/model.py#L904-L906
            state = []
            num_attention_heads = self.config.hidden_size // self.config.num_attention_heads
            state.append(
                torch.zeros(
                    (inputs_embeds.size(0), self.config.hidden_size, self.config.num_hidden_layers),
                    dtype=inputs_embeds.dtype,
                    requires_grad=False,
                    device=inputs_embeds.device,
                ).contiguous()
            )
            state.append(
                torch.zeros(
                    (
                        inputs_embeds.size(0),
                        num_attention_heads,
                        self.config.hidden_size // num_attention_heads,
                        self.config.hidden_size // num_attention_heads,
                        self.config.num_hidden_layers,
                    ),
                    dtype=torch.float32,
                    requires_grad=False,
                    device=inputs_embeds.device,
                ).contiguous()
            )
            state.append(
                torch.zeros(
                    (inputs_embeds.size(0), self.config.hidden_size, self.config.num_hidden_layers),
                    dtype=inputs_embeds.dtype,
                    requires_grad=False,
                    device=inputs_embeds.device,
                ).contiguous()
            )

        seq_mode = inputs_embeds.shape[1] > 1
        hidden_states = inputs_embeds

        all_self_attentions = () if output_attentions else None
        all_hidden_states = () if output_hidden_states else None
        for idx, block in enumerate(self.blocks):
            hidden_states, state, attentions = block(
                hidden_states, state=state, use_cache=use_cache, output_attentions=output_attentions, seq_mode=seq_mode
            )
            if (
                self.layers_are_rescaled
                and self.config.rescale_every > 0
                and (idx + 1) % self.config.rescale_every == 0
            ):
                hidden_states = hidden_states / 2

            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            if output_attentions:
                all_self_attentions = all_self_attentions + (attentions,)

        hidden_states = self.ln_out(hidden_states)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return (hidden_states, state, all_hidden_states, all_self_attentions)

        return Rwkv5Output(
            last_hidden_state=hidden_states,
            state=state,
            hidden_states=all_hidden_states,  # None
            attentions=all_self_attentions,  # None
        )

    def _rescale_layers(self):
        # Layers should be rescaled for inference only.
        if self.layers_are_rescaled == (not self.training):
            return
        if self.config.rescale_every > 0:
            with torch.no_grad():
                for block_id, block in enumerate(self.blocks):
                    if self.training:
                        block.attention.output.weight.mul_(2 ** int(block_id // self.config.rescale_every))
                        block.feed_forward.value.weight.mul_(2 ** int(block_id // self.config.rescale_every))
                    else:
                        block.attention.output.weight.div_(2 ** int(block_id // self.config.rescale_every))
                        block.feed_forward.value.weight.div_(2 ** int(block_id // self.config.rescale_every))

        self.layers_are_rescaled = not self.training


@add_start_docstrings(
    """
    The RWKV Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    """,
    RWKV_START_DOCSTRING,
)
class Rwkv5ForCausalLM(Rwkv5PreTrainedModel):
    _tied_weights_keys = ["head.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.rwkv = Rwkv5Model(config)
        self.head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        # Initialize weights and apply final processing
        self.post_init()

    def get_output_embeddings(self):
        return self.head

    def set_output_embeddings(self, new_embeddings):
        self.head = new_embeddings

    def prepare_inputs_for_generation(self, input_ids, state=None, inputs_embeds=None, **kwargs):
        # only last token for inputs_ids if the state is passed along.
        if state is not None:
            input_ids = input_ids[:, -1].unsqueeze(-1)

        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
        if inputs_embeds is not None and state is None:
            model_inputs = {"inputs_embeds": inputs_embeds}
        else:
            model_inputs = {"input_ids": input_ids}

        model_inputs["state"] = state
        return model_inputs

    @add_start_docstrings_to_model_forward(RWKV_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=Rwkv5CausalLMOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        state: Optional[List[torch.FloatTensor]] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, Rwkv5CausalLMOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        rwkv_outputs = self.rwkv(
            input_ids,
            inputs_embeds=inputs_embeds,
            state=state,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        hidden_states = rwkv_outputs[0]

        logits = self.head(hidden_states)

        loss = None
        if labels is not None:
            # https://github.com/BlinkDL/ChatRWKV/blob/main/rwkv_pip_package/src/rwkv/model.py#L984
            loss = torch.tensor(0.0, device=logits.device, dtype=logits.dtype)

        if not return_dict:
            output = (logits,) + rwkv_outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return Rwkv5CausalLMOutput(
            loss=loss,
            logits=logits,
            state=rwkv_outputs.state,
            hidden_states=rwkv_outputs.hidden_states,
            attentions=rwkv_outputs.attentions,
        )