File size: 31,178 Bytes
31112ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""
Convert a Flux model from Diffusers (folder or single-file) into the original
single-file Flux transformer checkpoint used by Black Forest Labs / ComfyUI.

Input  : /path/to/diffusers   (root or .../transformer)  OR  /path/to/*.safetensors (single file)
Output : /path/to/flux1-your-model.safetensors  (transformer only)

Usage:
  python diffusers_to_flux_transformer.py /path/to/diffusers /out/flux1-dev.safetensors
  python diffusers_to_flux_transformer.py /path/to/diffusion_pytorch_model.safetensors /out/flux1-dev.safetensors
  # optional quantization:
  #   --fp8           (float8_e4m3fn, simple)
  #   --fp8-scaled    (scaled float8 for 2D weights; adds .scale_weight tensors)
"""

import argparse
import json
from pathlib import Path
from collections import OrderedDict

import torch
from safetensors import safe_open
import safetensors.torch
from tqdm import tqdm


def parse_args():
    ap = argparse.ArgumentParser()
    ap.add_argument("diffusers_path", type=str,
                    help="Path to Diffusers checkpoint folder OR a single .safetensors file.")
    ap.add_argument("output_path", type=str,
                    help="Output .safetensors path for the Flux transformer.")
    ap.add_argument("--fp8", action="store_true",
                    help="Experimental: write weights as float8_e4m3fn via stochastic rounding (transformer only).")
    ap.add_argument("--fp8-scaled", action="store_true",
                    help="Experimental: scaled float8_e4m3fn for 2D weight tensors; adds .scale_weight tensors.")
    return ap.parse_args()


# Mapping from original Flux keys -> list of Diffusers keys (per block where applicable).
DIFFUSERS_MAP = {
    # global embeds
    "time_in.in_layer.weight": ["time_text_embed.timestep_embedder.linear_1.weight"],
    "time_in.in_layer.bias":   ["time_text_embed.timestep_embedder.linear_1.bias"],
    "time_in.out_layer.weight": ["time_text_embed.timestep_embedder.linear_2.weight"],
    "time_in.out_layer.bias":   ["time_text_embed.timestep_embedder.linear_2.bias"],

    "vector_in.in_layer.weight": ["time_text_embed.text_embedder.linear_1.weight"],
    "vector_in.in_layer.bias":   ["time_text_embed.text_embedder.linear_1.bias"],
    "vector_in.out_layer.weight": ["time_text_embed.text_embedder.linear_2.weight"],
    "vector_in.out_layer.bias":   ["time_text_embed.text_embedder.linear_2.bias"],

    "guidance_in.in_layer.weight": ["time_text_embed.guidance_embedder.linear_1.weight"],
    "guidance_in.in_layer.bias":   ["time_text_embed.guidance_embedder.linear_1.bias"],
    "guidance_in.out_layer.weight": ["time_text_embed.guidance_embedder.linear_2.weight"],
    "guidance_in.out_layer.bias":   ["time_text_embed.guidance_embedder.linear_2.bias"],

    "txt_in.weight": ["context_embedder.weight"],
    "txt_in.bias":   ["context_embedder.bias"],
    "img_in.weight": ["x_embedder.weight"],
    "img_in.bias":   ["x_embedder.bias"],

    # dual-stream (image/text) blocks
    "double_blocks.().img_mod.lin.weight": ["norm1.linear.weight"],
    "double_blocks.().img_mod.lin.bias":   ["norm1.linear.bias"],
    "double_blocks.().txt_mod.lin.weight": ["norm1_context.linear.weight"],
    "double_blocks.().txt_mod.lin.bias":   ["norm1_context.linear.bias"],

    "double_blocks.().img_attn.qkv.weight": [
        ["attn.to_q.weight", "attn.to_k.weight", "attn.to_v.weight"],
        ["qkv_proj.weight"],
    ],
    "double_blocks.().img_attn.qkv.bias": [
        ["attn.to_q.bias", "attn.to_k.bias", "attn.to_v.bias"],
        ["qkv_proj.bias"],
    ],
    "double_blocks.().txt_attn.qkv.weight": [
        ["attn.add_q_proj.weight", "attn.add_k_proj.weight", "attn.add_v_proj.weight"],
        ["qkv_proj_context.weight"],
    ],
    "double_blocks.().txt_attn.qkv.bias": [
        ["attn.add_q_proj.bias", "attn.add_k_proj.bias", "attn.add_v_proj.bias"],
        ["qkv_proj_context.bias"],
    ],

    "double_blocks.().img_attn.norm.query_norm.scale": [
        ["attn.norm_q.weight"],
        ["norm_q.weight"],
    ],
    "double_blocks.().img_attn.norm.key_norm.scale": [
        ["attn.norm_k.weight"],
        ["norm_k.weight"],
    ],
    "double_blocks.().txt_attn.norm.query_norm.scale": [
        ["attn.norm_added_q.weight"],
        ["norm_added_q.weight"],
    ],
    "double_blocks.().txt_attn.norm.key_norm.scale": [
        ["attn.norm_added_k.weight"],
        ["norm_added_k.weight"],
    ],

    "double_blocks.().img_mlp.0.weight": [
        ["ff.net.0.proj.weight"],
        ["mlp_fc1.weight"],
    ],
    "double_blocks.().img_mlp.0.bias": [
        ["ff.net.0.proj.bias"],
        ["mlp_fc1.bias"],
    ],
    "double_blocks.().img_mlp.2.weight": [
        ["ff.net.2.weight"],
        ["mlp_fc2.weight"],
    ],
    "double_blocks.().img_mlp.2.bias": [
        ["ff.net.2.bias"],
        ["mlp_fc2.bias"],
    ],

    "double_blocks.().txt_mlp.0.weight": [
        ["ff_context.net.0.proj.weight"],
        ["mlp_context_fc1.weight"],
    ],
    "double_blocks.().txt_mlp.0.bias": [
        ["ff_context.net.0.proj.bias"],
        ["mlp_context_fc1.bias"],
    ],
    "double_blocks.().txt_mlp.2.weight": [
        ["ff_context.net.2.weight"],
        ["mlp_context_fc2.weight"],
    ],
    "double_blocks.().txt_mlp.2.bias": [
        ["ff_context.net.2.bias"],
        ["mlp_context_fc2.bias"],
    ],

    "double_blocks.().img_attn.proj.weight": [
        ["attn.to_out.0.weight"],
        ["out_proj.weight"],
    ],
    "double_blocks.().img_attn.proj.bias": [
        ["attn.to_out.0.bias"],
        ["out_proj.bias"],
    ],
    "double_blocks.().txt_attn.proj.weight": [
        ["attn.to_add_out.weight"],
        ["out_proj_context.weight"],
    ],
    "double_blocks.().txt_attn.proj.bias": [
        ["attn.to_add_out.bias"],
        ["out_proj_context.bias"],
    ],

    # single-stream blocks
    "single_blocks.().modulation.lin.weight": ["norm.linear.weight"],
    "single_blocks.().modulation.lin.bias":   ["norm.linear.bias"],
    "single_blocks.().linear1.weight": [
        ["attn.to_q.weight", "attn.to_k.weight", "attn.to_v.weight", "proj_mlp.weight"],
        ["qkv_proj.weight", "mlp_fc1.weight"],
    ],
    "single_blocks.().linear1.bias": [
        ["attn.to_q.bias", "attn.to_k.bias", "attn.to_v.bias", "proj_mlp.bias"],
        ["qkv_proj.bias", "mlp_fc1.bias"],
    ],
    "single_blocks.().norm.query_norm.scale": [
        ["attn.norm_q.weight"],
        ["norm_q.weight"],
    ],
    "single_blocks.().norm.key_norm.scale": [
        ["attn.norm_k.weight"],
        ["norm_k.weight"],
    ],
    "single_blocks.().linear2.weight": [
        ["proj_out.weight"],
        ["out_proj.weight", "mlp_fc2.weight"],
    ],
    "single_blocks.().linear2.bias": [
        ["proj_out.bias"],
        ["out_proj.bias", "mlp_fc2.bias"],
    ],

    # final
    "final_layer.linear.weight":              ["proj_out.weight"],
    "final_layer.linear.bias":                ["proj_out.bias"],
    # these two are built from norm_out.linear.{weight,bias} by swapping [shift,scale] -> [scale,shift]
    "final_layer.adaLN_modulation.1.weight":  ["norm_out.linear.weight"],
    "final_layer.adaLN_modulation.1.bias":    ["norm_out.linear.bias"],
}

_TARGET_SUFFIXES = {}
for _tgt_key in DIFFUSERS_MAP:
    _base, _suffix = _tgt_key.rsplit(".", 1)
    _TARGET_SUFFIXES.setdefault(_base, set()).add("." + _suffix)


def _strip_prefix(key: str) -> str:
    return key[6:] if key.startswith("model.") else key


class StateDictSource:
    """
    Provide DiffusersSource-like access over an in-memory state dict.
    """

    POSSIBLE_PREFIXES = ["", "model."]

    def __init__(self, state_dict: dict):
        self._state_dict = state_dict
        self._all_keys = list(state_dict.keys())

    def _resolve(self, want: str):
        for pref in self.POSSIBLE_PREFIXES:
            key = pref + want
            if key in self._state_dict:
                return key
        return None

    def has(self, want: str) -> bool:
        return self._resolve(want) is not None

    def get(self, want: str) -> torch.Tensor:
        real_key = self._resolve(want)
        if real_key is None:
            raise KeyError(f"Missing key: {want}")
        return self._state_dict[real_key]

    @property
    def base_keys(self):
        return [_strip_prefix(k) for k in self._all_keys]


def detect_diffusers_state_dict(state_dict: dict) -> bool:
    base_keys = [_strip_prefix(k) for k in state_dict.keys()]
    if any(k.startswith(("double_blocks.", "single_blocks.")) for k in base_keys):
        return False
    return any(k.startswith(("transformer_blocks.", "single_transformer_blocks.")) for k in base_keys)


def convert_state_dict(state_dict: dict, *, verbose: bool = False) -> dict:
    if not detect_diffusers_state_dict(state_dict):
        return state_dict
    converted = _convert_from_source(StateDictSource(state_dict), verbose=verbose)
    return converted if converted else state_dict


def _count_blocks(base_keys):
    num_dual = 0
    num_single = 0
    for key in base_keys:
        if key.startswith("transformer_blocks."):
            try:
                idx = int(key.split(".")[1])
                num_dual = max(num_dual, idx + 1)
            except Exception:
                pass
        elif key.startswith("single_transformer_blocks."):
            try:
                idx = int(key.split(".")[1])
                num_single = max(num_single, idx + 1)
            except Exception:
                pass
    return num_dual, num_single


def _swap_scale_shift(vec: torch.Tensor) -> torch.Tensor:
    if vec is None or vec.ndim != 1 or vec.numel() % 2 != 0:
        return vec
    shift, scale = vec.chunk(2, dim=0)
    return torch.cat([scale, shift], dim=0)


def _swap_scale_shift_matrix(mat: torch.Tensor) -> torch.Tensor:
    if mat is None or mat.ndim != 2 or mat.size(0) % 2 != 0:
        return mat
    shift, scale = mat.chunk(2, dim=0)
    return torch.cat([scale, shift], dim=0)


def _collect_suffixes(base_keys, base):
    prefix = base + "."
    suffixes = set()
    for key in base_keys:
        if key.startswith(prefix):
            suffixes.add("." + key[len(prefix):])
    return suffixes


def _normalize_suffix(suffix: str) -> str:
    if suffix == ".smooth":
        return ".smooth_factor"
    if suffix == ".smooth_orig":
        return ".smooth_factor_orig"
    if suffix == ".lora_down":
        return ".proj_down"
    if suffix == ".lora_up":
        return ".proj_up"
    return suffix


def _match_dtype_device(ref: torch.Tensor, other: torch.Tensor) -> torch.Tensor:
    if ref.dtype != other.dtype or ref.device != other.device:
        return other.to(device=ref.device, dtype=ref.dtype)
    return other


def _concat(values, dim=0):
    if any(v is None for v in values):
        return None
    ref = values[0]
    if ref.ndim == 0:
        return ref
    merged = [ref] + [_match_dtype_device(ref, v) for v in values[1:]]
    return torch.cat(merged, dim=dim)


def _maybe_load_nunchaku():
    try:
        from shared.qtypes import nunchaku_int4 as _nunchaku_int4  # pylint: disable=import-outside-toplevel
    except Exception:
        return None
    return _nunchaku_int4


def _get_qweight_dims(qweight: torch.Tensor | None) -> tuple[int | None, int | None]:
    if not torch.is_tensor(qweight):
        return None, None
    if qweight.dtype == torch.int8:
        out_features = qweight.size(0)
    else:
        out_features = qweight.size(0) * 4
    in_features = qweight.size(1) * 2
    return out_features, in_features


def _merge_packed_scales_out(values, qweights, group_size: int = 64):
    if any(v is None for v in values):
        return None
    nunchaku = _maybe_load_nunchaku()
    if nunchaku is None or any(q is None for q in qweights):
        return _concat(values, dim=1)
    unpacked = []
    out_total = 0
    in_features = None
    for value, qweight in zip(values, qweights):
        out_i, in_i = _get_qweight_dims(qweight)
        if out_i is None or in_i is None:
            return _concat(values, dim=1)
        if in_features is None:
            in_features = in_i
        if in_i != in_features:
            return _concat(values, dim=1)
        unpacked_i = nunchaku._unpack_nunchaku_wscales(value, out_i, in_i, group_size)
        if not torch.is_tensor(unpacked_i):
            return _concat(values, dim=1)
        unpacked.append(unpacked_i)
        out_total += out_i
    merged = [unpacked[0]] + [_match_dtype_device(unpacked[0], u) for u in unpacked[1:]]
    merged = torch.cat(merged, dim=1)
    return nunchaku._pack_nunchaku_wscales(merged, out_total, in_features, group_size)


def _merge_packed_scales_in(values, qweights, group_size: int = 64):
    if any(v is None for v in values):
        return None
    nunchaku = _maybe_load_nunchaku()
    if nunchaku is None or any(q is None for q in qweights):
        return _concat(values, dim=0)
    out_a, in_a = _get_qweight_dims(qweights[0])
    out_b, in_b = _get_qweight_dims(qweights[1])
    if out_a is None or in_a is None or out_b is None or in_b is None:
        return _concat(values, dim=0)
    if out_a != out_b:
        return _concat(values, dim=0)
    unpack_a = nunchaku._unpack_nunchaku_wscales(values[0], out_a, in_a, group_size)
    unpack_b = nunchaku._unpack_nunchaku_wscales(values[1], out_b, in_b, group_size)
    if not torch.is_tensor(unpack_a) or not torch.is_tensor(unpack_b):
        return _concat(values, dim=0)
    if unpack_a.dtype != unpack_b.dtype or unpack_a.device != unpack_b.device:
        unpack_b = unpack_b.to(device=unpack_a.device, dtype=unpack_a.dtype)
    merged = torch.cat([unpack_a, unpack_b], dim=0)
    return nunchaku._pack_nunchaku_wscales(merged, out_a, in_a + in_b, group_size)


def _pad_to_multiple(tensor: torch.Tensor | None, divisor: int | tuple[int, int]):
    if tensor is None:
        return None
    if isinstance(divisor, int):
        div0 = div1 = divisor
    else:
        div0, div1 = divisor
    height, width = tensor.shape
    new_h = ((height + div0 - 1) // div0) * div0
    new_w = ((width + div1 - 1) // div1) * div1
    if new_h == height and new_w == width:
        return tensor
    padded = torch.zeros((new_h, new_w), dtype=tensor.dtype, device=tensor.device)
    padded[:height, :width] = tensor
    return padded


def _pack_lowrank_weight(weight: torch.Tensor | None, down: bool):
    if weight is None or weight.ndim != 2:
        return weight
    lane_n, lane_k = 1, 2
    n_pack_size, k_pack_size = 2, 2
    num_n_lanes, num_k_lanes = 8, 4
    frag_n = n_pack_size * num_n_lanes * lane_n
    frag_k = k_pack_size * num_k_lanes * lane_k
    weight = _pad_to_multiple(weight, (frag_n, frag_k))
    if weight is None:
        return None
    if down:
        rows, cols = weight.shape
        r_frags, c_frags = rows // frag_n, cols // frag_k
        weight = weight.view(r_frags, frag_n, c_frags, frag_k).permute(2, 0, 1, 3)
    else:
        cols, rows = weight.shape
        c_frags, r_frags = cols // frag_n, rows // frag_k
        weight = weight.view(c_frags, frag_n, r_frags, frag_k).permute(0, 2, 1, 3)
    weight = weight.reshape(c_frags, r_frags, n_pack_size, num_n_lanes, k_pack_size, num_k_lanes, lane_k)
    weight = weight.permute(0, 1, 3, 5, 2, 4, 6).contiguous()
    return weight.view(cols, rows)


def _pack_nunchaku_w4a4_weight(qvals, out_features, in_features):
    if qvals is None or qvals.ndim != 2:
        return qvals
    if qvals.dtype not in (torch.int8, torch.int16, torch.int32):
        qvals = qvals.to(torch.int32)
    if qvals.dtype != torch.int32:
        qvals = qvals.to(torch.int32)
    if qvals.shape != (out_features, in_features):
        return None
    mem_n = 128
    mem_k = 64
    num_k_unrolls = 2
    if out_features % mem_n != 0 or in_features % (mem_k * num_k_unrolls) != 0:
        return None
    n_pack_size = 2
    k_pack_size = 2
    num_n_lanes = 8
    num_k_lanes = 4
    reg_n = 1
    reg_k = 8
    num_n_packs = mem_n // (n_pack_size * num_n_lanes * reg_n)
    num_k_packs = mem_k // (k_pack_size * num_k_lanes * reg_k)
    n_tiles = out_features // mem_n
    k_tiles = in_features // mem_k
    weight = qvals.reshape(
        n_tiles,
        num_n_packs,
        n_pack_size,
        num_n_lanes,
        reg_n,
        k_tiles,
        num_k_packs,
        k_pack_size,
        num_k_lanes,
        reg_k,
    )
    weight = weight.permute(0, 5, 6, 1, 3, 8, 2, 7, 4, 9).contiguous()
    weight = weight.bitwise_and_(0xF)
    shifts = torch.arange(0, 32, 4, dtype=torch.int32, device=weight.device)
    weight = weight.bitwise_left_shift_(shifts)
    weight = weight.sum(dim=-1, dtype=torch.int32)
    return weight.view(dtype=torch.int8).view(out_features, -1)


def _unpack_lowrank_weight(weight: torch.Tensor | None, down: bool):
    if weight is None:
        return None
    nunchaku = _maybe_load_nunchaku()
    if nunchaku is None:
        return weight
    return nunchaku._unpack_lowrank_weight(weight, down)


def _block_diag_out(mats):
    if not mats:
        return None
    ref = mats[0]
    total_rows = sum(m.size(0) for m in mats)
    total_cols = sum(m.size(1) for m in mats)
    out = torch.zeros((total_rows, total_cols), dtype=ref.dtype, device=ref.device)
    row = 0
    col = 0
    for mat in mats:
        mat = _match_dtype_device(ref, mat)
        out[row : row + mat.size(0), col : col + mat.size(1)] = mat
        row += mat.size(0)
        col += mat.size(1)
    return out


def _merge_lowrank_down(values):
    if any(v is None for v in values):
        return None
    unpacked = [_unpack_lowrank_weight(v, down=True) for v in values]
    ref = unpacked[0]
    merged = [ref] + [_match_dtype_device(ref, v) for v in unpacked[1:]]
    merged = torch.cat(merged, dim=0)
    return _pack_lowrank_weight(merged, down=True)


def _merge_lowrank_up(values):
    if any(v is None for v in values):
        return None
    unpacked = [_unpack_lowrank_weight(v, down=False) for v in values]
    merged = _block_diag_out(unpacked)
    return _pack_lowrank_weight(merged, down=False)


def _merge_lowrank_down_block_diag(values):
    if any(v is None for v in values):
        return None
    unpacked = [_unpack_lowrank_weight(v, down=True) for v in values]
    merged = _block_diag_out(unpacked)
    return _pack_lowrank_weight(merged, down=True)


def _merge_lowrank_up_concat(values):
    if any(v is None for v in values):
        return None
    unpacked = [_unpack_lowrank_weight(v, down=False) for v in values]
    ref = unpacked[0]
    merged = [ref] + [_match_dtype_device(ref, v) for v in unpacked[1:]]
    merged = torch.cat(merged, dim=1)
    return _pack_lowrank_weight(merged, down=False)


def _merge_qweight_in(values):
    if any(v is None for v in values):
        return None
    a, b = values
    out_a, in_a = _get_qweight_dims(a)
    out_b, in_b = _get_qweight_dims(b)
    if out_a is None or in_a is None or out_b is None or in_b is None:
        return _concat(values, dim=1)
    if out_a != out_b:
        return _concat(values, dim=1)
    nunchaku = _maybe_load_nunchaku()
    if nunchaku is None or a.dtype != torch.int8 or b.dtype != torch.int8:
        return _concat(values, dim=1)
    unpack_a = nunchaku._unpack_nunchaku_w4a4_weight(a, out_a, in_a)
    unpack_b = nunchaku._unpack_nunchaku_w4a4_weight(b, out_b, in_b)
    if not torch.is_tensor(unpack_a) or not torch.is_tensor(unpack_b):
        return _concat(values, dim=1)
    if unpack_a.dtype != torch.int32:
        unpack_a = unpack_a.to(torch.int32)
    if unpack_b.dtype != torch.int32:
        unpack_b = unpack_b.to(torch.int32)
    if unpack_a.dtype != unpack_b.dtype or unpack_a.device != unpack_b.device:
        unpack_b = unpack_b.to(device=unpack_a.device, dtype=unpack_a.dtype)
    merged = torch.cat([unpack_a, unpack_b], dim=1)
    packed = _pack_nunchaku_w4a4_weight(merged, out_a, in_a + in_b)
    if packed is None:
        return _concat(values, dim=1)
    return packed


def _merge_multi(values, suffix, qweights):
    if any(v is None for v in values):
        return None
    if suffix in (".wscales", ".wzeros"):
        return _merge_packed_scales_out(values, qweights)
    if suffix == ".proj_down":
        return _merge_lowrank_down(values)
    if suffix == ".proj_up":
        return _merge_lowrank_up(values)
    if suffix in (".smooth_factor", ".smooth_factor_orig", ".input_scale", ".output_scale", ".scale_weight"):
        return values[0]
    return _concat(values, dim=0)


def _merge_multi_in(values, suffix, qweights):
    if any(v is None for v in values):
        return None
    if suffix == ".qweight":
        return _merge_qweight_in(values)
    if suffix in (".wscales", ".wzeros"):
        return _merge_packed_scales_in(values, qweights)
    if suffix == ".proj_down":
        return _merge_lowrank_down_block_diag(values)
    if suffix == ".proj_up":
        return _merge_lowrank_up_concat(values)
    if suffix in (".smooth_factor", ".smooth_factor_orig"):
        return _concat(values, dim=0)
    if suffix == ".bias":
        ref = values[0]
        total = ref
        for val in values[1:]:
            total = total + _match_dtype_device(ref, val)
        return total
    if suffix in (".input_scale", ".output_scale", ".scale_weight"):
        return values[0]
    return _concat(values, dim=1)


def _convert_from_source(src, *, verbose: bool = False) -> dict:
    base_keys = src.base_keys
    num_dual, num_single = _count_blocks(base_keys)
    if verbose:
        print(f"Found {num_dual} dual-stream blocks, {num_single} single-stream blocks")

    out = {}
    quant_suffixes = {
        ".qweight",
        ".wscales",
        ".wzeros",
        ".smooth",
        ".smooth_orig",
        ".lora_down",
        ".lora_up",
        ".bias",
        ".input_scale",
        ".output_scale",
        ".scale_weight",
    }

    def _map_entry(src_prefix, tgt_template, dvals):
        tgt_base_template, tgt_suffix = tgt_template.rsplit(".", 1)
        tgt_suffix = "." + tgt_suffix
        candidates = dvals if isinstance(dvals[0], (list, tuple)) else [dvals]

        for candidate in candidates:
            src_suffix = "." + candidate[0].rsplit(".", 1)[1]
            if any(d.rsplit(".", 1)[1] != src_suffix.lstrip(".") for d in candidate):
                continue
            src_bases = [d.rsplit(".", 1)[0] for d in candidate]
            suffix_sets = [_collect_suffixes(base_keys, src_prefix + base) for base in src_bases]
            if any(not suffix_set for suffix_set in suffix_sets):
                continue
            common_suffixes = set.intersection(*suffix_sets)
            if not common_suffixes:
                continue

            explicit_suffixes = _TARGET_SUFFIXES.get(tgt_base_template, set())
            allow_extra = src_suffix == tgt_suffix
            if src_suffix in common_suffixes:
                suffixes = {src_suffix}
                if allow_extra and tgt_suffix == ".weight":
                    suffixes |= (common_suffixes - explicit_suffixes)
            else:
                suffixes = common_suffixes & quant_suffixes
                if not suffixes:
                    continue

            tgt_base = (
                tgt_base_template.replace("()", str(block_idx))
                if "()" in tgt_base_template
                else tgt_base_template
            )
            merge_in_features = tgt_base_template.endswith(".linear2") and len(src_bases) > 1

            for suffix in suffixes:
                values = [src.get(src_prefix + base + suffix) for base in src_bases]
                if len(values) == 1:
                    merged = values[0]
                else:
                    qweights = [
                        src.get(src_prefix + base + ".qweight") if src.has(src_prefix + base + ".qweight") else None
                        for base in src_bases
                    ]
                    norm_suffix = _normalize_suffix(suffix)
                    if merge_in_features:
                        merged = _merge_multi_in(values, norm_suffix, qweights)
                    else:
                        merged = _merge_multi(values, norm_suffix, qweights)
                if merged is None:
                    continue
                out_suffix = tgt_suffix if suffix == src_suffix else _normalize_suffix(suffix)
                out[tgt_base + out_suffix] = merged
            break

    for block_idx in range(num_dual):
        prefix = f"transformer_blocks.{block_idx}."
        for tgt_key, dvals in DIFFUSERS_MAP.items():
            if not tgt_key.startswith("double_blocks."):
                continue
            _map_entry(prefix, tgt_key, dvals)

    for block_idx in range(num_single):
        prefix = f"single_transformer_blocks.{block_idx}."
        for tgt_key, dvals in DIFFUSERS_MAP.items():
            if not tgt_key.startswith("single_blocks."):
                continue
            _map_entry(prefix, tgt_key, dvals)

    block_idx = None
    for tgt_key, dvals in DIFFUSERS_MAP.items():
        if tgt_key.startswith(("double_blocks.", "single_blocks.")):
            continue
        _map_entry("", tgt_key, dvals)

    if "final_layer.adaLN_modulation.1.weight" in out:
        out["final_layer.adaLN_modulation.1.weight"] = _swap_scale_shift_matrix(
            out["final_layer.adaLN_modulation.1.weight"]
        )
    if "final_layer.adaLN_modulation.1.bias" in out:
        out["final_layer.adaLN_modulation.1.bias"] = _swap_scale_shift(
            out["final_layer.adaLN_modulation.1.bias"]
        )

    return out


class DiffusersSource:
    """
    Uniform interface over:
      1) Folder with index JSON + shards
      2) Folder with exactly one .safetensors (no index)
      3) Single .safetensors file
    Provides .has(key), .get(key)->Tensor, .base_keys (keys with 'model.' stripped for scanning)
    """

    POSSIBLE_PREFIXES = ["", "model."]  # try in this order

    def __init__(self, path: Path):
        p = Path(path)
        if p.is_dir():
            # use 'transformer' subfolder if present
            if (p / "transformer").is_dir():
                p = p / "transformer"
            self._init_from_dir(p)
        elif p.is_file() and p.suffix == ".safetensors":
            self._init_from_single_file(p)
        else:
            raise FileNotFoundError(f"Invalid path: {p}")

    # ---------- common helpers ----------

    @staticmethod
    def _strip_prefix(k: str) -> str:
        return k[6:] if k.startswith("model.") else k

    def _resolve(self, want: str):
        """
        Return the actual stored key matching `want` by trying known prefixes.
        """
        for pref in self.POSSIBLE_PREFIXES:
            k = pref + want
            if k in self._all_keys:
                return k
        return None

    def has(self, want: str) -> bool:
        return self._resolve(want) is not None

    def get(self, want: str) -> torch.Tensor:
        real_key = self._resolve(want)
        if real_key is None:
            raise KeyError(f"Missing key: {want}")
        return self._get_by_real_key(real_key).to("cpu")

    @property
    def base_keys(self):
        # keys without 'model.' prefix for scanning
        return [self._strip_prefix(k) for k in self._all_keys]

    # ---------- modes ----------

    def _init_from_single_file(self, file_path: Path):
        self._mode = "single"
        self._file = file_path
        self._handle = safe_open(file_path, framework="pt", device="cpu")
        self._all_keys = list(self._handle.keys())

        def _get_by_real_key(real_key: str):
            return self._handle.get_tensor(real_key)

        self._get_by_real_key = _get_by_real_key

    def _init_from_dir(self, dpath: Path):
        index_json = dpath / "diffusion_pytorch_model.safetensors.index.json"
        if index_json.exists():
            with open(index_json, "r", encoding="utf-8") as f:
                index = json.load(f)
            weight_map = index["weight_map"]  # full mapping
            self._mode = "sharded"
            self._dpath = dpath
            self._weight_map = {k: dpath / v for k, v in weight_map.items()}
            self._all_keys = list(self._weight_map.keys())
            self._open_handles = {}

            def _get_by_real_key(real_key: str):
                fpath = self._weight_map[real_key]
                h = self._open_handles.get(fpath)
                if h is None:
                    h = safe_open(fpath, framework="pt", device="cpu")
                    self._open_handles[fpath] = h
                return h.get_tensor(real_key)

            self._get_by_real_key = _get_by_real_key
            return

        # no index: try exactly one safetensors in folder
        files = sorted(dpath.glob("*.safetensors"))
        if len(files) != 1:
            raise FileNotFoundError(
                f"No index found and {dpath} does not contain exactly one .safetensors file."
            )
        self._init_from_single_file(files[0])


def main():
    args = parse_args()
    src = DiffusersSource(Path(args.diffusers_path))
    orig = _convert_from_source(src, verbose=True)

    # Optional FP8 variants (experimental; not required for ComfyUI/BFL)
    if args.fp8 or args.fp8_scaled:
        dtype = torch.float8_e4m3fn  # noqa
        minv, maxv = torch.finfo(dtype).min, torch.finfo(dtype).max

        def stochastic_round_to(t):
            t = t.float().clamp(minv, maxv)
            lower = torch.floor(t * 256) / 256
            upper = torch.ceil(t * 256) / 256
            prob = torch.where(upper != lower, (t - lower) / (upper - lower), torch.zeros_like(t))
            rnd = torch.rand_like(t)
            out = torch.where(rnd < prob, upper, lower)
            return out.to(dtype)

        def scale_to_8bit(weight, target_max=416.0):
            absmax = weight.abs().max()
            scale = absmax / target_max if absmax > 0 else torch.tensor(1.0)
            scaled = (weight / scale).clamp(minv, maxv).to(dtype)
            return scaled, scale

        scales = {}
        for k in tqdm(list(orig.keys()), desc="Quantizing to fp8"):
            t = orig[k]
            if args.fp8:
                orig[k] = stochastic_round_to(t)
            else:
                if k.endswith(".weight") and t.dim() == 2:
                    qt, s = scale_to_8bit(t)
                    orig[k] = qt
                    scales[k[:-len(".weight")] + ".scale_weight"] = s
                else:
                    orig[k] = t.clamp(minv, maxv).to(dtype)
        if args.fp8_scaled:
            orig.update(scales)
            orig["scaled_fp8"] = torch.tensor([], dtype=dtype)
    else:
        # Default: save in bfloat16
        for k in list(orig.keys()):
            orig[k] = orig[k].to(torch.bfloat16).cpu()

    out_path = Path(args.output_path)
    out_path.parent.mkdir(parents=True, exist_ok=True)
    meta = OrderedDict()
    meta["format"] = "pt"
    meta["modelspec.date"] = __import__("datetime").date.today().strftime("%Y-%m-%d")
    print(f"Saving transformer to: {out_path}")
    safetensors.torch.save_file(orig, str(out_path), metadata=meta)
    print("Done.")


if __name__ == "__main__":
    main()