Skip to content

Commit 28fac13

Browse files
committed
reduce number of models, add create_tensor_gate_up helper
1 parent c7fd777 commit 28fac13

7 files changed

Lines changed: 85 additions & 291 deletions

File tree

gguf-py/gguf/constants.py

Lines changed: 2 additions & 35 deletions
Original file line numberDiff line numberDiff line change
@@ -1368,7 +1368,6 @@ class MODEL_TENSOR(IntEnum):
13681368
MODEL_TENSOR.FFN_GATE_EXP,
13691369
MODEL_TENSOR.FFN_DOWN_EXP,
13701370
MODEL_TENSOR.FFN_UP_EXP,
1371-
MODEL_TENSOR.FFN_GATE_UP_EXP,
13721371
],
13731372
MODEL_ARCH.LLAMA4: [
13741373
MODEL_TENSOR.TOKEN_EMBD,
@@ -1389,7 +1388,6 @@ class MODEL_TENSOR(IntEnum):
13891388
MODEL_TENSOR.FFN_GATE_EXP,
13901389
MODEL_TENSOR.FFN_DOWN_EXP,
13911390
MODEL_TENSOR.FFN_UP_EXP,
1392-
MODEL_TENSOR.FFN_GATE_UP_EXP,
13931391
MODEL_TENSOR.FFN_GATE_SHEXP,
13941392
MODEL_TENSOR.FFN_DOWN_SHEXP,
13951393
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -1413,7 +1411,6 @@ class MODEL_TENSOR(IntEnum):
14131411
MODEL_TENSOR.FFN_GATE_EXP,
14141412
MODEL_TENSOR.FFN_DOWN_EXP,
14151413
MODEL_TENSOR.FFN_UP_EXP,
1416-
MODEL_TENSOR.FFN_GATE_UP_EXP,
14171414
],
14181415
MODEL_ARCH.GROK: [
14191416
MODEL_TENSOR.TOKEN_EMBD,
@@ -1435,7 +1432,6 @@ class MODEL_TENSOR(IntEnum):
14351432
MODEL_TENSOR.FFN_GATE_EXP,
14361433
MODEL_TENSOR.FFN_DOWN_EXP,
14371434
MODEL_TENSOR.FFN_UP_EXP,
1438-
MODEL_TENSOR.FFN_GATE_UP_EXP,
14391435
MODEL_TENSOR.FFN_POST_NORM,
14401436
MODEL_TENSOR.LAYER_OUT_NORM,
14411437
],
@@ -1750,7 +1746,6 @@ class MODEL_TENSOR(IntEnum):
17501746
MODEL_TENSOR.FFN_GATE_EXP,
17511747
MODEL_TENSOR.FFN_DOWN_EXP,
17521748
MODEL_TENSOR.FFN_UP_EXP,
1753-
MODEL_TENSOR.FFN_GATE_UP_EXP,
17541749
MODEL_TENSOR.FFN_GATE_INP_SHEXP,
17551750
MODEL_TENSOR.FFN_GATE_SHEXP,
17561751
MODEL_TENSOR.FFN_DOWN_SHEXP,
@@ -1789,7 +1784,6 @@ class MODEL_TENSOR(IntEnum):
17891784
MODEL_TENSOR.FFN_GATE_EXP,
17901785
MODEL_TENSOR.FFN_DOWN_EXP,
17911786
MODEL_TENSOR.FFN_UP_EXP,
1792-
MODEL_TENSOR.FFN_GATE_UP_EXP,
17931787
],
17941788
MODEL_ARCH.QWEN3NEXT: [
17951789
MODEL_TENSOR.TOKEN_EMBD,
@@ -1855,7 +1849,6 @@ class MODEL_TENSOR(IntEnum):
18551849
MODEL_TENSOR.FFN_GATE_EXP,
18561850
MODEL_TENSOR.FFN_DOWN_EXP,
18571851
MODEL_TENSOR.FFN_UP_EXP,
1858-
MODEL_TENSOR.FFN_GATE_UP_EXP,
18591852
],
18601853
MODEL_ARCH.QWEN35: [
18611854
MODEL_TENSOR.TOKEN_EMBD,
@@ -1904,6 +1897,7 @@ class MODEL_TENSOR(IntEnum):
19041897
MODEL_TENSOR.FFN_DOWN_EXP,
19051898
MODEL_TENSOR.FFN_UP_EXP,
19061899
MODEL_TENSOR.FFN_GATE_EXP,
1900+
MODEL_TENSOR.FFN_GATE_UP_EXP,
19071901
MODEL_TENSOR.SSM_A,
19081902
MODEL_TENSOR.SSM_CONV1D,
19091903
MODEL_TENSOR.SSM_DT,
@@ -2031,7 +2025,6 @@ class MODEL_TENSOR(IntEnum):
20312025
MODEL_TENSOR.FFN_GATE_EXP,
20322026
MODEL_TENSOR.FFN_DOWN_EXP,
20332027
MODEL_TENSOR.FFN_UP_EXP,
2034-
MODEL_TENSOR.FFN_GATE_UP_EXP,
20352028
],
20362029
MODEL_ARCH.CODESHELL: [
20372030
MODEL_TENSOR.TOKEN_EMBD,
@@ -2098,7 +2091,6 @@ class MODEL_TENSOR(IntEnum):
20982091
MODEL_TENSOR.FFN_GATE_EXP,
20992092
MODEL_TENSOR.FFN_DOWN_EXP,
21002093
MODEL_TENSOR.FFN_UP_EXP,
2101-
MODEL_TENSOR.FFN_GATE_UP_EXP,
21022094
],
21032095
MODEL_ARCH.MINICPM3: [
21042096
MODEL_TENSOR.TOKEN_EMBD,
@@ -2411,7 +2403,6 @@ class MODEL_TENSOR(IntEnum):
24112403
MODEL_TENSOR.FFN_GATE_EXP,
24122404
MODEL_TENSOR.FFN_DOWN_EXP,
24132405
MODEL_TENSOR.FFN_UP_EXP,
2414-
MODEL_TENSOR.FFN_GATE_UP_EXP,
24152406
],
24162407
MODEL_ARCH.XVERSE: [
24172408
MODEL_TENSOR.TOKEN_EMBD,
@@ -2467,7 +2458,6 @@ class MODEL_TENSOR(IntEnum):
24672458
MODEL_TENSOR.FFN_GATE_EXP,
24682459
MODEL_TENSOR.FFN_DOWN_EXP,
24692460
MODEL_TENSOR.FFN_UP_EXP,
2470-
MODEL_TENSOR.FFN_GATE_UP_EXP,
24712461
],
24722462
MODEL_ARCH.OLMO: [
24732463
MODEL_TENSOR.TOKEN_EMBD,
@@ -2526,7 +2516,6 @@ class MODEL_TENSOR(IntEnum):
25262516
MODEL_TENSOR.FFN_GATE_EXP,
25272517
MODEL_TENSOR.FFN_UP_EXP,
25282518
MODEL_TENSOR.FFN_DOWN_EXP,
2529-
MODEL_TENSOR.FFN_GATE_UP_EXP,
25302519
],
25312520
MODEL_ARCH.OPENELM: [
25322521
MODEL_TENSOR.TOKEN_EMBD,
@@ -2561,7 +2550,6 @@ class MODEL_TENSOR(IntEnum):
25612550
MODEL_TENSOR.FFN_GATE_EXP,
25622551
MODEL_TENSOR.FFN_DOWN_EXP,
25632552
MODEL_TENSOR.FFN_UP_EXP,
2564-
MODEL_TENSOR.FFN_GATE_UP_EXP,
25652553
],
25662554
MODEL_ARCH.DEEPSEEK: [
25672555
MODEL_TENSOR.TOKEN_EMBD,
@@ -2582,7 +2570,6 @@ class MODEL_TENSOR(IntEnum):
25822570
MODEL_TENSOR.FFN_GATE_EXP,
25832571
MODEL_TENSOR.FFN_DOWN_EXP,
25842572
MODEL_TENSOR.FFN_UP_EXP,
2585-
MODEL_TENSOR.FFN_GATE_UP_EXP,
25862573
MODEL_TENSOR.FFN_GATE_SHEXP,
25872574
MODEL_TENSOR.FFN_DOWN_SHEXP,
25882575
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -2609,10 +2596,10 @@ class MODEL_TENSOR(IntEnum):
26092596
MODEL_TENSOR.FFN_GATE,
26102597
MODEL_TENSOR.FFN_DOWN,
26112598
MODEL_TENSOR.FFN_UP,
2612-
MODEL_TENSOR.FFN_GATE_UP_EXP,
26132599
MODEL_TENSOR.FFN_GATE_EXP,
26142600
MODEL_TENSOR.FFN_DOWN_EXP,
26152601
MODEL_TENSOR.FFN_UP_EXP,
2602+
MODEL_TENSOR.FFN_GATE_UP_EXP,
26162603
MODEL_TENSOR.FFN_GATE_SHEXP,
26172604
MODEL_TENSOR.FFN_DOWN_SHEXP,
26182605
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -2635,7 +2622,6 @@ class MODEL_TENSOR(IntEnum):
26352622
MODEL_TENSOR.FFN_GATE_EXP,
26362623
MODEL_TENSOR.FFN_DOWN_EXP,
26372624
MODEL_TENSOR.FFN_UP_EXP,
2638-
MODEL_TENSOR.FFN_GATE_UP_EXP,
26392625
MODEL_TENSOR.FFN_GATE_SHEXP,
26402626
MODEL_TENSOR.FFN_DOWN_SHEXP,
26412627
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -2713,7 +2699,6 @@ class MODEL_TENSOR(IntEnum):
27132699
MODEL_TENSOR.FFN_GATE_EXP,
27142700
MODEL_TENSOR.FFN_DOWN_EXP,
27152701
MODEL_TENSOR.FFN_UP_EXP,
2716-
MODEL_TENSOR.FFN_GATE_UP_EXP,
27172702
MODEL_TENSOR.FFN_GATE_SHEXP,
27182703
MODEL_TENSOR.FFN_DOWN_SHEXP,
27192704
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -2968,7 +2953,6 @@ class MODEL_TENSOR(IntEnum):
29682953
MODEL_TENSOR.FFN_GATE_EXP,
29692954
MODEL_TENSOR.FFN_DOWN_EXP,
29702955
MODEL_TENSOR.FFN_UP_EXP,
2971-
MODEL_TENSOR.FFN_GATE_UP_EXP,
29722956
MODEL_TENSOR.FFN_GATE_SHEXP,
29732957
MODEL_TENSOR.FFN_DOWN_SHEXP,
29742958
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -3009,7 +2993,6 @@ class MODEL_TENSOR(IntEnum):
30092993
MODEL_TENSOR.FFN_GATE_EXP,
30102994
MODEL_TENSOR.FFN_DOWN_EXP,
30112995
MODEL_TENSOR.FFN_UP_EXP,
3012-
MODEL_TENSOR.FFN_GATE_UP_EXP,
30132996
MODEL_TENSOR.FFN_GATE_SHEXP,
30142997
MODEL_TENSOR.FFN_UP_SHEXP,
30152998
MODEL_TENSOR.FFN_DOWN_SHEXP,
@@ -3036,7 +3019,6 @@ class MODEL_TENSOR(IntEnum):
30363019
MODEL_TENSOR.FFN_GATE_EXP,
30373020
MODEL_TENSOR.FFN_DOWN_EXP,
30383021
MODEL_TENSOR.FFN_UP_EXP,
3039-
MODEL_TENSOR.FFN_GATE_UP_EXP,
30403022
MODEL_TENSOR.FFN_GATE_SHEXP,
30413023
MODEL_TENSOR.FFN_UP_SHEXP,
30423024
MODEL_TENSOR.FFN_DOWN_SHEXP,
@@ -3098,7 +3080,6 @@ class MODEL_TENSOR(IntEnum):
30983080
MODEL_TENSOR.FFN_GATE_EXP,
30993081
MODEL_TENSOR.FFN_DOWN_EXP,
31003082
MODEL_TENSOR.FFN_UP_EXP,
3101-
MODEL_TENSOR.FFN_GATE_UP_EXP,
31023083
MODEL_TENSOR.FFN_GATE_SHEXP,
31033084
MODEL_TENSOR.FFN_DOWN_SHEXP,
31043085
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -3121,7 +3102,6 @@ class MODEL_TENSOR(IntEnum):
31213102
MODEL_TENSOR.FFN_GATE_EXP,
31223103
MODEL_TENSOR.FFN_DOWN_EXP,
31233104
MODEL_TENSOR.FFN_UP_EXP,
3124-
MODEL_TENSOR.FFN_GATE_UP_EXP,
31253105
MODEL_TENSOR.FFN_GATE_SHEXP,
31263106
MODEL_TENSOR.FFN_DOWN_SHEXP,
31273107
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -3156,7 +3136,6 @@ class MODEL_TENSOR(IntEnum):
31563136
MODEL_TENSOR.FFN_UP,
31573137
MODEL_TENSOR.FFN_UP_EXP,
31583138
MODEL_TENSOR.FFN_UP_SHEXP,
3159-
MODEL_TENSOR.FFN_GATE_UP_EXP,
31603139
],
31613140
MODEL_ARCH.ARCEE: [
31623141
MODEL_TENSOR.TOKEN_EMBD,
@@ -3193,7 +3172,6 @@ class MODEL_TENSOR(IntEnum):
31933172
MODEL_TENSOR.FFN_GATE_EXP,
31943173
MODEL_TENSOR.FFN_DOWN_EXP,
31953174
MODEL_TENSOR.FFN_UP_EXP,
3196-
MODEL_TENSOR.FFN_GATE_UP_EXP,
31973175
MODEL_TENSOR.FFN_GATE_SHEXP,
31983176
MODEL_TENSOR.FFN_UP_SHEXP,
31993177
MODEL_TENSOR.FFN_DOWN_SHEXP,
@@ -3280,7 +3258,6 @@ class MODEL_TENSOR(IntEnum):
32803258
MODEL_TENSOR.FFN_GATE_EXP,
32813259
MODEL_TENSOR.FFN_DOWN_EXP,
32823260
MODEL_TENSOR.FFN_UP_EXP,
3283-
MODEL_TENSOR.FFN_GATE_UP_EXP,
32843261
MODEL_TENSOR.FFN_GATE_SHEXP,
32853262
MODEL_TENSOR.FFN_DOWN_SHEXP,
32863263
MODEL_TENSOR.FFN_UP_SHEXP,
@@ -3329,7 +3306,6 @@ class MODEL_TENSOR(IntEnum):
33293306
MODEL_TENSOR.ATTN_OUT,
33303307
MODEL_TENSOR.ATTN_SINKS,
33313308
MODEL_TENSOR.FFN_GATE_INP,
3332-
MODEL_TENSOR.FFN_GATE_UP_EXP,
33333309
MODEL_TENSOR.FFN_GATE_EXP,
33343310
MODEL_TENSOR.FFN_DOWN_EXP,
33353311
MODEL_TENSOR.FFN_UP_EXP,
@@ -3375,7 +3351,6 @@ class MODEL_TENSOR(IntEnum):
33753351
MODEL_TENSOR.FFN_GATE_EXP,
33763352
MODEL_TENSOR.FFN_DOWN_EXP,
33773353
MODEL_TENSOR.FFN_UP_EXP,
3378-
MODEL_TENSOR.FFN_GATE_UP_EXP,
33793354
MODEL_TENSOR.FFN_EXP_PROBS_B,
33803355
],
33813356
MODEL_ARCH.SMALLTHINKER: [
@@ -3395,7 +3370,6 @@ class MODEL_TENSOR(IntEnum):
33953370
MODEL_TENSOR.FFN_GATE_EXP,
33963371
MODEL_TENSOR.FFN_DOWN_EXP,
33973372
MODEL_TENSOR.FFN_UP_EXP,
3398-
MODEL_TENSOR.FFN_GATE_UP_EXP,
33993373
],
34003374
MODEL_ARCH.APERTUS: [
34013375
MODEL_TENSOR.TOKEN_EMBD,
@@ -3431,7 +3405,6 @@ class MODEL_TENSOR(IntEnum):
34313405
MODEL_TENSOR.FFN_GATE_EXP,
34323406
MODEL_TENSOR.FFN_UP_EXP,
34333407
MODEL_TENSOR.FFN_DOWN_EXP,
3434-
MODEL_TENSOR.FFN_GATE_UP_EXP,
34353408
],
34363409
MODEL_ARCH.GROVEMOE: [
34373410
MODEL_TENSOR.TOKEN_EMBD,
@@ -3449,7 +3422,6 @@ class MODEL_TENSOR(IntEnum):
34493422
MODEL_TENSOR.FFN_GATE_EXP,
34503423
MODEL_TENSOR.FFN_DOWN_EXP,
34513424
MODEL_TENSOR.FFN_UP_EXP,
3452-
MODEL_TENSOR.FFN_GATE_UP_EXP,
34533425
MODEL_TENSOR.FFN_GATE_CHEXP,
34543426
MODEL_TENSOR.FFN_DOWN_CHEXP,
34553427
MODEL_TENSOR.FFN_UP_CHEXP,
@@ -3470,7 +3442,6 @@ class MODEL_TENSOR(IntEnum):
34703442
MODEL_TENSOR.FFN_GATE_EXP,
34713443
MODEL_TENSOR.FFN_DOWN_EXP,
34723444
MODEL_TENSOR.FFN_UP_EXP,
3473-
MODEL_TENSOR.FFN_GATE_UP_EXP,
34743445
MODEL_TENSOR.FFN_EXP_PROBS_B,
34753446
],
34763447
MODEL_ARCH.COGVLM: [
@@ -3506,7 +3477,6 @@ class MODEL_TENSOR(IntEnum):
35063477
MODEL_TENSOR.FFN_GATE_EXP,
35073478
MODEL_TENSOR.FFN_DOWN_EXP,
35083479
MODEL_TENSOR.FFN_UP_EXP,
3509-
MODEL_TENSOR.FFN_GATE_UP_EXP,
35103480
],
35113481
MODEL_ARCH.PANGU_EMBED: [
35123482
MODEL_TENSOR.TOKEN_EMBD,
@@ -3541,7 +3511,6 @@ class MODEL_TENSOR(IntEnum):
35413511
MODEL_TENSOR.FFN_GATE_EXP,
35423512
MODEL_TENSOR.FFN_DOWN_EXP,
35433513
MODEL_TENSOR.FFN_UP_EXP,
3544-
MODEL_TENSOR.FFN_GATE_UP_EXP,
35453514
],
35463515
MODEL_ARCH.MIMO2: [
35473516
MODEL_TENSOR.TOKEN_EMBD,
@@ -3561,7 +3530,6 @@ class MODEL_TENSOR(IntEnum):
35613530
MODEL_TENSOR.FFN_GATE_EXP,
35623531
MODEL_TENSOR.FFN_DOWN_EXP,
35633532
MODEL_TENSOR.FFN_UP_EXP,
3564-
MODEL_TENSOR.FFN_GATE_UP_EXP,
35653533
MODEL_TENSOR.FFN_EXP_PROBS_B,
35663534
],
35673535
MODEL_ARCH.STEP35: [
@@ -3609,7 +3577,6 @@ class MODEL_TENSOR(IntEnum):
36093577
MODEL_TENSOR.FFN_GATE_EXP,
36103578
MODEL_TENSOR.FFN_DOWN_EXP,
36113579
MODEL_TENSOR.FFN_UP_EXP,
3612-
MODEL_TENSOR.FFN_GATE_UP_EXP,
36133580
],
36143581
MODEL_ARCH.MAINCODER: [
36153582
MODEL_TENSOR.TOKEN_EMBD,

gguf-py/gguf/tensor_mapping.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -568,7 +568,7 @@ class TensorNameMap:
568568
),
569569

570570
MODEL_TENSOR.FFN_GATE_UP_EXP: (
571-
"model.layers.{bid}.mlp.experts.gate_up_proj", # gpt-oss
571+
"model.layers.{bid}.mlp.experts.gate_up_proj",
572572
),
573573

574574
# Feed-forward down

0 commit comments

Comments
 (0)