llama : Support llama 4 text-only (#12791)

* llama4 conversion

* initial support, no chat template

* clean up a bit

* fix tokenizer conversion

* correct hparams

* try this

* fix shexp

* ffn_inp_normed

* chat template

* clean up model conversion

* add_bos

* add scale_before_ffn

* fix order

* weight_before_ffn

* llm_graph_input_attn_temp

* add chunk attn mask

* build_inp_attn_scale()

* add comment about ggml_repeat

* clarify comments

* fix build
This commit is contained in:
Xuan-Son Nguyen 2025-04-07 23:06:44 +02:00 committed by GitHub
parent 82974011f3
commit 1466621e73
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
17 changed files with 532 additions and 22 deletions

View file

@ -0,0 +1,112 @@
ied 4 ½ months
__ggml_vocab_test__
Führer
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
__ggml_vocab_test__
Hello world
__ggml_vocab_test__
Hello world
__ggml_vocab_test__
Hello World
__ggml_vocab_test__
Hello World
__ggml_vocab_test__
Hello World!
__ggml_vocab_test__
Hello, world!
__ggml_vocab_test__
Hello, world!
__ggml_vocab_test__
this is 🦙.cpp
__ggml_vocab_test__
w048 7tuijk dsdfhu
__ggml_vocab_test__
нещо на Български
__ggml_vocab_test__
កាន់តែពិសេសអាចខលចេញ
__ggml_vocab_test__
🚀 (normal) 😶‍🌫️ (multiple emojis concatenated) ✅ (only emoji that has its own token)
__ggml_vocab_test__
Hello
__ggml_vocab_test__
Hello
__ggml_vocab_test__
Hello
__ggml_vocab_test__
Hello
__ggml_vocab_test__
Hello
__ggml_vocab_test__
Hello
Hello
__ggml_vocab_test__
(
__ggml_vocab_test__
=
__ggml_vocab_test__
' era
__ggml_vocab_test__
Hello, y'all! How are you 😁 ?我想在apple工作1314151天
__ggml_vocab_test__
!!!!!!
__ggml_vocab_test__
3
__ggml_vocab_test__
33
__ggml_vocab_test__
333
__ggml_vocab_test__
3333
__ggml_vocab_test__
33333
__ggml_vocab_test__
333333
__ggml_vocab_test__
3333333
__ggml_vocab_test__
33333333
__ggml_vocab_test__
333333333
__ggml_vocab_test__
Cửa Việt
__ggml_vocab_test__
discards
__ggml_vocab_test__
🚀 (normal) 😶‍🌫️ (multiple emojis concatenated) ✅ 🦙🦙 3 33 333 3333 33333 333333 3333333 33333333 3.3 3..3 3...3 កាន់តែពិសេសអាច😁 ?我想在apple工作1314151天 ------======= нещо на Български ''''''```````""""......!!!!!!?????? I've been 'told he's there, 'RE you sure? 'M not sure I'll make it, 'D you like some tea? We'Ve a'lL
__ggml_vocab_test__

View file

@ -0,0 +1,46 @@
1190 220 32 220 18215 7112
50 16800 258
220
256
277
197
198
368
2946
3271
19873 3817
39715 3817
19873 7353
39715 7353
39715 7353 13
19873 24 3817 13
39715 24 3817 13
544 373 9522 112 247 26 36315
99 39923 220 35 9607 21498 21470 3679 9433
1595 7653 633 79829 34051 1636
8755 102595 115960 21125 148305 96819 102816 39048 14105 22528 160234
114590 222 330 14879 21 51358 127 12817 93293 117 24204 330 68239 881 120327 170428 21 89101 330 7384 88230 511 947 1492 3742 7233 21
19873
39715
220 39715
256 39715
277 39715
277 39715 198 277 39715
330
198 319
19 7359
19873 24 386 87799 13 2403 583 650 51358 223 1663 155736 1522 42056 7544 13336 28785 29 4412 20645
17931 4959
31
1922
12325
12325 31
12325 1922
12325 12325
12325 12325 31
12325 12325 1922
12325 12325 12325
47 19811 12077
3260 3579
198 7283 51499 191231 20192 3271 3322 9287 2143 17860 114590 222 330 14879 21 51358 127 12817 93293 117 24204 330 68239 881 120327 170428 21 89101 9522 112 247 172394 247 220 31 220 1922 220 12325 220 12325 31 220 12325 1922 220 12325 12325 220 12325 12325 31 220 12325 12325 1922 220 31 26 31 220 31 396 31 220 31 1043 31 117131 102595 115960 21125 148305 96819 102816 80883 223 1663 155736 1522 42056 7544 13336 28785 29 4412 20645 79745 150278 117079 633 79829 34051 1636 25611 41990 109428 1488 91054 24072 17931 4959 29795 9296 16517 1806 481 96 1386 36633 1609 24 481 1109 650 5074 43 481 57 702 5074 27088 2170 536 24 481 48 650 1933 1696 30262 43 1665 19 32818 262 27236 56