File size: 38,409 Bytes
56f6578
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
# Copyright 2023 Baichuan Inc. All Rights Reserved.

# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# 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.


from .configuration_baichuan import BaichuanConfig
from .generation_utils import build_chat_input, TextIterStreamer

import math
from typing import List, Optional, Tuple, Union
from threading import Thread

import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
from torch.nn import functional as F
from transformers import PreTrainedModel, PretrainedConfig
from transformers.activations import ACT2FN
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
from transformers.generation.utils import GenerationConfig
from transformers.utils import logging, ContextManagers

import os
from contextlib import contextmanager
logger = logging.get_logger(__name__)

try:
    from xformers import ops as xops
except ImportError:
    xops = None
    logger.warning(
        "Xformers is not installed correctly. If you want to use memory_efficient_attention to accelerate training use the following command to install Xformers\npip install xformers."
    )


# Copied from transformers.models.bart.modeling_bart._make_causal_mask
def _make_causal_mask(
        input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
):
    """
    Make causal mask used for bi-directional self-attention.
    """
    bsz, tgt_len = input_ids_shape
    mask = torch.full((tgt_len, tgt_len), torch.tensor(torch.finfo(dtype).min, device=device), device=device)
    mask_cond = torch.arange(mask.size(-1), device=device)
    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
    mask = mask.to(dtype)

    if past_key_values_length > 0:
        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)

def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    if len(mask.size()) == 3:
        bsz, src_len, _ = mask.size()
        tgt_len = tgt_len if tgt_len is not None else src_len
        expanded_mask = mask[:,None,:,:].expand(bsz, 1, tgt_len, src_len).to(dtype)
    else:
        bsz, src_len = mask.size()
        tgt_len = tgt_len if tgt_len is not None else src_len
        expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)


class RMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        RMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)

        # convert into half-precision if necessary
        if self.weight.dtype in [torch.float16, torch.bfloat16]:
            hidden_states = hidden_states.to(self.weight.dtype)

        return self.weight * hidden_states


class RotaryEmbedding(torch.nn.Module):
    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
        super().__init__()
        self.inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float().to(device) / dim))
        self.max_seq_len_cached = max_position_embeddings
        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        self.cos_cached = emb.cos()[None, None, :, :].to(torch.float32)
        self.sin_cached = emb.sin()[None, None, :, :].to(torch.float32)
    def forward(self, x, seq_len=None):
        # x: [bs, num_attention_heads, seq_len, head_size]
        # This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
        if seq_len > self.max_seq_len_cached:
            self.max_seq_len_cached = seq_len
            t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
            freqs = torch.outer(t, self.inv_freq)
            emb = torch.cat((freqs, freqs), dim=-1)
            self.cos_cached = emb.cos()[None, None, :, :].to(torch.float32).to(x.device)
            self.sin_cached = emb.sin()[None, None, :, :].to(torch.float32).to(x.device)
        elif self.cos_cached.device != x.device:
            self.cos_cached = self.cos_cached.to(x.device)
            self.sin_cached = self.sin_cached.to(x.device)  
        return (
            self.cos_cached[:, :, :seq_len, ...],
            self.sin_cached[:, :, :seq_len, ...],
        )


def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2:]
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(q, k, cos_, sin_, position_ids):
    cos = cos_.squeeze(1).squeeze(0)  # [seq_len, dim]
    sin = sin_.squeeze(1).squeeze(0)  # [seq_len, dim]
    cos = cos[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    sin = sin[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    q_embed = (q.float() * cos) + (rotate_half(q.float()) * sin)
    k_embed = (k.float() * cos) + (rotate_half(k.float()) * sin)
    return q_embed.to(q.dtype), k_embed.to(k.dtype)


class MLP(nn.Module):
    def __init__(
            self,
            hidden_size: int,
            intermediate_size: int,
            hidden_act: str,
    ):
        super().__init__()
        self.gate_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
        self.down_proj = nn.Linear(intermediate_size, hidden_size, bias=False)
        self.up_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
        self.act_fn = ACT2FN[hidden_act]

    def forward(self, x):
        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))


class DocLLMAttention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""
    def __init__(self, config: BaichuanConfig):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.hidden_size // self.num_heads
        self.max_position_embeddings = config.max_position_embeddings
        self._lambda_ts = config.lambda_ts
        self._lambda_st = config.lambda_st
        self._lambda_ss = config.lambda_ss

        if (self.head_dim * self.num_heads) != self.hidden_size:
            raise ValueError(
                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
                f" and `num_heads`: {self.num_heads})."
            )
        self.W_pack = nn.Linear(self.hidden_size, 3 * self.hidden_size, bias=False)
        self.spatial_W_pack = nn.Linear(self.hidden_size, 3 * self.hidden_size, bias=False)
        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
        self.rotary_emb = RotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)

    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()

    def forward(
            self,
            hidden_states: torch.Tensor,
            bounding_box_embeddings: torch.Tensor,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_value: Optional[Tuple[torch.Tensor]] = None,
            spatial_past_key_value: Optional[Tuple[torch.Tensor]] = None,
            output_attentions: bool = False,
            use_cache: bool = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        bsz, q_len, _ = hidden_states.size()



        proj = self.W_pack(hidden_states)
        proj = proj.unflatten(-1, (3, self.hidden_size)).unsqueeze(0).transpose(0, -2).squeeze(-2)
        query_states = proj[0].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = proj[1].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        value_states = proj[2].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)

        spatial_proj = self.spatial_W_pack(bounding_box_embeddings)
        spatial_proj = spatial_proj.unflatten(-1, (3, self.hidden_size)).unsqueeze(0).transpose(0, -2).squeeze(-2)
        spatial_query_states = spatial_proj[0].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        spatial_key_states = spatial_proj[1].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        spatial_value_states = spatial_proj[2].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)

        kv_seq_len = key_states.shape[-2]
        if past_key_value is not None:
            kv_seq_len += past_key_value[0].shape[-2]

        spatial_kv_seq_len = spatial_key_states.shape[-2] 
        if spatial_past_key_value is not None:
            spatial_kv_seq_len += spatial_past_key_value[0].shape[-2]

        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)

        # FIXME do we need applying pos embeddings for spatial modality?
        spatial_query_states, spatial_key_states = apply_rotary_pos_emb(spatial_query_states, spatial_key_states, cos, sin, position_ids)
        # [bsz, nh, t, hd]

        if past_key_value is not None:
            # reuse k, v, self_attention
            key_states = torch.cat([past_key_value[0], key_states], dim=2)
            value_states = torch.cat([past_key_value[1], value_states], dim=2)

        if spatial_past_key_value is not None:
            # reuse k, v, self_attention
            spatial_key_states = torch.cat([spatial_past_key_value[0], spatial_key_states], dim=2)
            spatial_value_states = torch.cat([spatial_past_key_value[1], spatial_value_states], dim=2)

        past_key_value = (key_states, value_states) if use_cache else None
        spatial_past_key_value = (spatial_key_states, spatial_value_states) if use_cache else None

        if xops is not None and self.training:
            attn_weights = None
            query_states = query_states.transpose(1, 2)
            key_states = key_states.transpose(1, 2)
            value_states = value_states.transpose(1, 2)
            attn_output = xops.memory_efficient_attention(
                query_states, key_states, value_states, attn_bias=xops.LowerTriangularMask()
            )
        else:
            with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=True):


                # attn_output = F.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask = attention_mask)
                L, S = query_states.size(-2), key_states.size(-2)
                scale_factor = 1 / math.sqrt(query_states.size(-1)) 
                # attn_bias = torch.zeros(L, S, dtype=query_states.dtype)
                attn_bias = torch.zeros_like(attention_mask)

                if attention_mask is not None:
                    if attention_mask.dtype == torch.bool:
                        attn_bias.masked_fill_(attention_mask.logical_not(), float("-inf"))
                    else:
                        attn_bias += attention_mask

                attn_weight_tt = query_states @ key_states.transpose(-2, -1) * scale_factor
                attn_weight_ts = query_states @ spatial_key_states.transpose(-2, -1) * scale_factor
                attn_weight_st = spatial_query_states @ key_states.transpose(-2, -1) * scale_factor
                attn_weight_ss = spatial_query_states @ spatial_key_states.transpose(-2, -1) * scale_factor

                attn_weight = attn_weight_tt + self._lambda_ts * attn_weight_ts + self._lambda_st * attn_weight_st + self._lambda_ss * attn_weight_ss
                attn_weight += attn_bias
                attn_weight = torch.softmax(attn_weight, dim=-1)
                attn_weight = torch.dropout(attn_weight, 0.0, train=True)
                attn_output = attn_weight @ value_states

            attn_output = attn_output.transpose(1, 2)
        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
        attn_output = self.o_proj(attn_output)

        if not output_attentions:
            attn_weights = None

        return attn_output, attn_weights, past_key_value, spatial_past_key_value


class DecoderLayer(nn.Module):
    def __init__(self, config: BaichuanConfig):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.self_attn = DocLLMAttention(config=config)
        self.mlp = MLP(
            hidden_size=self.hidden_size,
            intermediate_size=config.intermediate_size,
            hidden_act=config.hidden_act,
        )
        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(
            self,
            hidden_states: torch.Tensor,
            bounding_box_embeddings: torch.Tensor,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_value: Optional[Tuple[torch.Tensor]] = None,
            spatial_past_key_value: Optional[Tuple[torch.Tensor]] = None,
            output_attentions: Optional[bool] = False,
            use_cache: Optional[bool] = False,
    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:

        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states, self_attn_weights, present_key_value, spatial_present_key_value = self.self_attn(
            hidden_states=hidden_states,
            bounding_box_embeddings=bounding_box_embeddings,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value,
            spatial_past_key_value=spatial_past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)
            outputs += (spatial_present_key_value,)

        return outputs


class BaichuanPreTrainedModel(PreTrainedModel):
    config_class = BaichuanConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["DecoderLayer"]
    _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]

    def _init_weights(self, module):
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, BaichuanModel):
            module.gradient_checkpointing = value

class DocLLMBaseModelOutputWithPast(BaseModelOutputWithPast):
    def __init__(self, *args, spatial_past_key_values=None, **kwargs):
        super().__init__(*args, **kwargs)
        self.spatial_past_key_values = spatial_past_key_values

class DocLLMCausalLMOutputWithPast(CausalLMOutputWithPast):
    def __init__(self, *args, spatial_past_key_values=None, **kwargs):
        super().__init__(*args, **kwargs)
        self.spatial_past_key_values = spatial_past_key_values


class BaichuanModel(BaichuanPreTrainedModel):
    def __init__(self, config: BaichuanConfig):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
        self.embed_spatial = nn.Linear(4, config.hidden_size, bias=False)
        self.layers = nn.ModuleList([DecoderLayer(config) for _ in range(config.num_hidden_layers)])
        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

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

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        self.embed_tokens = value

    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
        # create causal mask
        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
        combined_attention_mask = None
        if input_shape[-1] > 1:
            combined_attention_mask = _make_causal_mask(
                input_shape,
                inputs_embeds.dtype,
                device=inputs_embeds.device,
                past_key_values_length=past_key_values_length,
            )

        if attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
                inputs_embeds.device
            )
            combined_attention_mask = (
                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
            )

        return combined_attention_mask

    def forward(
            self,
            input_ids: torch.LongTensor = None,
            input_coordinates: torch.FloatTensor = None,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[List[torch.FloatTensor]] = None,
            spatial_past_key_values: Optional[List[torch.FloatTensor]] = None,
            inputs_embeds: Optional[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, BaseModelOutputWithPast]:
        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
        )
        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

        # retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
        elif input_ids is not None:
            batch_size, seq_length = input_ids.shape
        elif inputs_embeds is not None:
            batch_size, seq_length, _ = inputs_embeds.shape
        else:
            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
        assert input_ids.device == input_coordinates.device

        seq_length_with_past = seq_length
        past_key_values_length = 0

        spatial_seq_length_with_past = seq_length
        spatial_past_key_values_length = 0

        if past_key_values is not None:
            past_key_values_length = past_key_values[0][0].shape[2]
            seq_length_with_past = seq_length_with_past + past_key_values_length

        if spatial_past_key_values is not None:
            spatial_past_key_values_length = spatial_past_key_values[0][0].shape[2]
            spatial_seq_length_with_past = spatial_seq_length_with_past + spatial_past_key_values_length

        if position_ids is None:
            device = input_ids.device if input_ids is not None else inputs_embeds.device
            position_ids = torch.arange(
                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
            )
            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
        else:
            position_ids = position_ids.view(-1, seq_length).long()

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)
        
        coordinate_embeds = self.embed_spatial(input_coordinates)
    

        # embed positions
        if attention_mask is None:
            attention_mask = torch.ones(
                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
            )
        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
        )

        hidden_states = inputs_embeds

        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None
        next_decoder_cache = () if use_cache else None
        spatial_next_decoder_cache = () if use_cache else None

        for idx, decoder_layer in enumerate(self.layers):
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            past_key_value = past_key_values[idx] if past_key_values is not None else None
            spatial_past_key_value = spatial_past_key_values[idx] if spatial_past_key_values is not None else None

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        # None for past_key_value
                        return module(*inputs, output_attentions, None)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),
                    hidden_states,
                    coordinate_embeds,
                    attention_mask,
                    position_ids,
                    None,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    coordinate_embeds,
                    attention_mask=attention_mask,
                    position_ids=position_ids,
                    past_key_value=past_key_value,
                    spatial_past_key_value=spatial_past_key_value,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                )

            hidden_states = layer_outputs[0]

            if use_cache:
                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
                spatial_next_decoder_cache += (layer_outputs[3 if output_attentions else 1],)

            if output_attentions:
                all_self_attns += (layer_outputs[1],)

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        next_cache = next_decoder_cache if use_cache else None
        spatial_next_cache = spatial_next_decoder_cache if use_cache else None

        if not return_dict:
            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
        return DocLLMBaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=next_cache,
            spatial_past_key_values=spatial_next_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
        )


class NormHead(nn.Module):
    def __init__(self, hidden_size, vocab_size, bias=False):
        super().__init__()
        self.weight = nn.Parameter(torch.empty((vocab_size, hidden_size)))
        nn.init.kaiming_uniform_(self.weight, a=math.sqrt(5))
        self.first_flag = True

    def forward(self, hidden_states):
        if self.training:
            norm_weight = nn.functional.normalize(self.weight)
        elif self.first_flag:
            self.first_flag = False
            self.weight = nn.Parameter(nn.functional.normalize(self.weight))
            norm_weight = self.weight
        else:
            norm_weight = self.weight
        return nn.functional.linear(hidden_states, norm_weight)

_init_weights = True
@contextmanager
def no_init_weights(_enable=True):
    global _init_weights
    old_init_weights = _init_weights
    if _enable:
        _init_weights = False
    try:
        yield
    finally:
        _init_weights = old_init_weights

class BaichuanForCausalLM(BaichuanPreTrainedModel):
    def __init__(self, config, *model_args, **model_kwargs):
        super().__init__(config, *model_args, **model_kwargs)
        self.model = BaichuanModel(config)

        self.lm_head = NormHead(config.hidden_size, config.vocab_size, bias=False)
        if hasattr(config, "quantization_config") and config.quantization_config['load_in_4bit']:
            try:
                from .quantizer import quantize_offline, init_model_weight_int4
            except ImportError:
                raise ImportError(f"Needs QLinear to run quantize.")
            quantize_offline(self, 4)
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        self.model.embed_tokens = value

    def get_output_embeddings(self):
        return self.lm_head

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

    def set_decoder(self, decoder):
        self.model = decoder

    def get_decoder(self):
        return self.model
    
    @classmethod
    def from_pretrained(
        cls,
        pretrained_model_name_or_path: Optional[Union[str, os.PathLike]],
        *model_args,
        config: Optional[Union[PretrainedConfig, str, os.PathLike]] = None,
        cache_dir: Optional[Union[str, os.PathLike]] = None,
        ignore_mismatched_sizes: bool = False,
        force_download: bool = False,
        local_files_only: bool = False,
        token: Optional[Union[str, bool]] = None,
        revision: str = "main",
        use_safetensors: bool = None,
        **kwargs,
    ):
        # Load config if we don't provide a configuration
        if not isinstance(config, PretrainedConfig):
            config_path = config if config is not None else pretrained_model_name_or_path
            config, model_kwargs = cls.config_class.from_pretrained(
                config_path,
                cache_dir=cache_dir,
                return_unused_kwargs=True,
                force_download=force_download,
                resume_download=False,
                proxies=None,
                local_files_only=local_files_only,
                token=token,
                revision=revision,
                subfolder="",
                _from_auto=False,
                _from_pipeline=None,
                **kwargs,
            )
        else:
            model_kwargs = kwargs
        
        if hasattr(config, "quantization_config") and config.quantization_config['load_in_4bit']:
            try:
                from .quantizer import init_model_weight_int4
                from accelerate import init_empty_weights, dispatch_model, infer_auto_device_map
                from accelerate.utils import CustomDtype
                from accelerate.utils import get_balanced_memory
            except ImportError:
                raise ImportError(f"Needs import model weight init func to run quantize.") 
            # Instantiate model.
            init_contexts = [no_init_weights(_enable=True)]
            init_contexts.append(init_empty_weights())
            with ContextManagers(init_contexts):
                model = cls(config)
            
            model_file = os.path.join(pretrained_model_name_or_path, 'pytorch_model.bin')
            state_dict = torch.load(model_file, map_location="cpu") 
            model.is_quantized = True
                        
            device_map = kwargs.pop("device_map", None)
            torch_dtype = kwargs.pop("torch_dtype", None)
            
            kwargs = {"no_split_module_classes": model._no_split_modules}
            target_dtype = CustomDtype.INT4
            max_memory = get_balanced_memory(
                model,
                dtype=target_dtype,
                low_zero=(device_map == "balanced_low_0"),
                max_memory=None,
                **kwargs,
            )
            kwargs["max_memory"] = max_memory
            
            device_map = infer_auto_device_map(model, dtype=target_dtype, **kwargs)
            model = init_model_weight_int4(config, model, state_dict)
            
            # Set model in evaluation mode to deactivate DropOut modules by default
            model.eval()
            # If it is a model with generation capabilities, attempt to load the generation config
            if model.can_generate():
                try:
                    model.generation_config = GenerationConfig.from_pretrained(
                        pretrained_model_name_or_path,
                        cache_dir=cache_dir,
                        force_download=force_download,
                        resume_download=False,
                        proxies=None,
                        local_files_only=local_files_only,
                        token=token,
                        revision=revision,
                        subfolder="",
                        _from_auto=False,
                        _from_pipeline=None,
                        **kwargs,
                    )
                except (OSError, TypeError):
                    logger.info(
                        "Generation config file not found, using a generation config created from the model config."
                    )
                    pass
            
            if device_map is not None:
                dispatch_model(model, device_map=device_map)
            
            return model
        return super(BaichuanForCausalLM, cls).from_pretrained(pretrained_model_name_or_path, *model_args, 
                config=config, cache_dir=cache_dir, ignore_mismatched_sizes=ignore_mismatched_sizes, 
                force_download=force_download, local_files_only=local_files_only, token=token, revision=revision, 
                use_safetensors=use_safetensors, **kwargs)   

    def forward(
            self,
            input_ids: torch.LongTensor = None,
            input_coordinates: torch.FloatTensor = None,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[List[torch.FloatTensor]] = None,
            spatial_past_key_values: Optional[List[torch.FloatTensor]] = None,
            inputs_embeds: Optional[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, CausalLMOutputWithPast]:

        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
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.model(
            input_ids=input_ids,
            input_coordinates=input_coordinates,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            spatial_past_key_values=spatial_past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        hidden_states = outputs[0]
        logits = self.lm_head(hidden_states)
        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = CrossEntropyLoss()
            shift_logits = shift_logits.view(-1, self.config.vocab_size)
            shift_labels = shift_labels.view(-1)
            # Enable model parallelism
            shift_labels = shift_labels.to(shift_logits.device)
            loss = loss_fct(shift_logits, shift_labels)

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

        return DocLLMCausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            spatial_past_key_values=outputs.spatial_past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

    def prepare_inputs_for_generation(
            self, input_ids, past_key_values=None, spatial_past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
    ):
        if past_key_values:
            input_ids = input_ids[:, -1:]

        position_ids = kwargs.get("position_ids", None)
        if attention_mask is not None and position_ids is None:
            # create position_ids on the fly for batch generation
            position_ids = attention_mask.long().cumsum(-1) - 1
            position_ids.masked_fill_(attention_mask == 0, 1)
            if past_key_values:
                position_ids = position_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 past_key_values is None:
            model_inputs = {"inputs_embeds": inputs_embeds}
        else:
            model_inputs = {"input_ids": input_ids}

        model_inputs.update(
            {
                "position_ids": position_ids,
                "past_key_values": past_key_values,
                "spatial_past_key_values":spatial_past_key_values,
                "use_cache": kwargs.get("use_cache"),
                "attention_mask": attention_mask,
            }
        )
        return model_inputs

    @staticmethod
    def _reorder_cache(past_key_values, beam_idx):
        reordered_past = ()
        for layer_past in past_key_values:
            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
        return reordered_past

    def quantize(self, bits: int):
        try:
            from .quantizer import quantize_online
        except ImportError:
            raise ImportError(f"Needs QLinear to run quantize.")
        return quantize_online(self, bits)

    def chat(self, tokenizer, messages: List[dict], stream=False,
             generation_config: Optional[GenerationConfig]=None):
        generation_config = generation_config or self.generation_config
        input_ids = build_chat_input(self, tokenizer, messages, generation_config.max_new_tokens)
        if stream:
            streamer = TextIterStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
            Thread(target=self.generate, kwargs=dict(
                inputs=input_ids, streamer=streamer,
                generation_config=generation_config,
            )).start()
            return streamer
        else:
            outputs = self.generate(input_ids, generation_config=generation_config)
            response = tokenizer.decode(outputs[0][len(input_ids[0]):], skip_special_tokens=True)
            return response