36 Commits

Author SHA1 Message Date
hiyouga
0e4ee9d9a3 update liger kernel
Former-commit-id: a7dd7d325e68c92c7470c1e9ef83a7c8abcbc616
2024-08-29 20:46:08 +08:00
hiyouga
c765292093 support liger kernel
Former-commit-id: 72bc8f01111ad69b92a647b54b4af988515d9c34
2024-08-27 11:20:14 +08:00
hiyouga
20013e130b fix #5048
Former-commit-id: b7ca6c8dc14f689d0df16684a6121cc0ec24f8ba
2024-08-05 23:48:19 +08:00
hiyouga
14bc7b0551 fix up
Former-commit-id: 29ebcd75d55f70f2891632eba187b643cc3a9e51
2024-07-15 01:04:56 +08:00
hiyouga
12e0e5d0d7 tiny fix
Former-commit-id: d3c01552e0f978f150902175f096f6e3bfb64363
2024-07-14 10:56:45 +08:00
hiyouga
0b26011181 fix gemma2 attention
Former-commit-id: 2f6af73da28c4f8321b625fd09ddec8bd4977b08
2024-07-13 23:33:45 +08:00
hiyouga
d7657d772d tiny fix
Former-commit-id: 0c699de39de06eac96af67e8dd4fc4c53335b17e
2024-07-04 03:47:05 +08:00
hiyouga
7b3c1f29ff fix packing for eager/sdpa attn
Former-commit-id: 6fd6aa4530f81a2ed306eeb2a5167607288b62c6
2024-07-04 01:52:43 +08:00
hoshi-hiyouga
a38ff842d0 Merge pull request #4224 from chuan298/main
Implement efficient packing without cross-contamination attention

Former-commit-id: 87d9b2d00513c163335d3f2e2bb3cb3299cecdaa
2024-07-04 01:18:54 +08:00
hiyouga
bfdaadcc40 update packing
Former-commit-id: cce7083024bed4c7429ddc8288d1c9190fde29f5
2024-07-04 01:10:55 +08:00
hoshi-hiyouga
51c75985b8 Update packing.py
Former-commit-id: a36e8f2dd50e0f1c589457a7e785fdbc905d561d
2024-07-03 23:36:01 +08:00
hiyouga
13cec0cc2f update func name
Former-commit-id: c346f79f99db5296000e4d22a65e53c26e85b344
2024-07-03 23:29:33 +08:00
hiyouga
e671ed520b update arg name
Former-commit-id: 8a6a7b9c8a876da9c16e5ada7df461eb8cabee21
2024-07-03 23:23:24 +08:00
hiyouga
104151d558 tiny fix
Former-commit-id: 8b1172b91085125a83a4150943873141c8bbd8bc
2024-07-03 02:31:50 +08:00
hoshi-hiyouga
2452f57cd7 Merge branch 'main' into main
Former-commit-id: e8e6af26514272e29a50649b38182beb4db4ebfa
2024-07-01 21:01:09 +08:00
hiyouga
de4de5b5ab tiny fix
Former-commit-id: 8c41a0aa6db8bf31200c83b14819d474927268a1
2024-07-01 03:55:20 +08:00
hiyouga
2b006beab1 loose gemma2 attention
Former-commit-id: 2f4b89ace15b7a4d2adf16eeba9feb7de9e25d43
2024-06-29 01:42:14 +08:00
hiyouga
87e60f8bac bf16 by default, gemma2 attns
Gemma2 finetuning cannot work until merging https://github.com/huggingface/transformers/pull/31674


Former-commit-id: 4d35e218b1d60ff24b368ff5bc608be9c85411de
2024-06-28 06:00:26 +08:00
hiyouga
58607ec1b0 add quant checks
Former-commit-id: 96a5044394bff75ca8ef17bd7d07d4da66f797f0
2024-06-27 01:12:25 +08:00
hiyouga
d2d9fa4abb support HQQ/EETQ #4113
Former-commit-id: ad144c2265cdee0d23014dbb3d017ea257cb26ed
2024-06-27 00:29:42 +08:00
hiyouga
6b2733ce12 improve autogptq integration
Former-commit-id: addca926de42f91366185a47eb8e777ed44a8e77
2024-06-26 22:11:44 +08:00
stceum
9aa640f27b Bug Fix: off is parsed as False in yaml file, changed to disabled to avoid this.
Former-commit-id: 3ed063f281d1c2563df1b9eb3800543208c9dc16
2024-06-24 20:39:31 +08:00
ancv
5319447aa5 move configure_packing to llamafactory.model.patcher and fix constants
Former-commit-id: 770f75dc8363bfa284a72159ff8ad25ec9abe4e0
2024-06-21 00:45:06 +07:00
hiyouga
030b4811c7 update patcher
Former-commit-id: 3b040e8e0f78dbb6bc1409a1b2b788e1affc7458
2024-06-19 21:27:00 +08:00
hiyouga
5156114981 fix #4357
Former-commit-id: 4bd77d8563aa85230af65caf901214247e214bed
2024-06-18 22:42:45 +08:00
hiyouga
7ef169ed39 fix #4326
Former-commit-id: e2665e71c7428014d46d91542b01a58c1064d05a
2024-06-17 18:17:48 +08:00
ancv
988231026a update packing with sdpa and eager attention mode
Former-commit-id: 238f5c3d99809c6ae2571b59bdce8d8ea3c700b9
2024-06-16 02:25:47 +07:00
hiyouga
c0c6b8075a tiny fix
Former-commit-id: 38b6b0f52edeb8ba45aa03b415b3c0c1b0e0c1e4
2024-06-16 01:06:41 +08:00
ancv
9d9f8c6531 remove some unused params
Former-commit-id: 04315c3d92ecc25537e45d5807cb38bc290dcb16
2024-06-15 23:00:55 +07:00
hiyouga
2946153cea add license
Former-commit-id: d87108daa68bd40174b262be1ca65fe6e1b7ab56
2024-06-15 17:54:33 +08:00
hiyouga
833aa324c2 clean code
Former-commit-id: 2ed8270112755971e3f2dfd2f29c5939b077330a
2024-06-13 01:58:16 +08:00
ancv
045eb155a2 implement efficient packing without cross-contamination attention
Former-commit-id: b2c367bc61c2778dc359613dca496d9e134c2743
2024-06-12 11:56:01 +07:00
hiyouga
8c574eb3cb fix deepspeed version
Former-commit-id: cca6f351081903ca3b5f79f10accc1bbbae0ee61
2024-06-11 16:52:36 +08:00
hiyouga
e3baa5aa08 tiny fix
Former-commit-id: 3f24337a8a995b145b1e8075bc23878eaa363844
2024-06-11 01:04:16 +08:00
hiyouga
2f164c2c41 fix #4160
The split heads should be concatenated in dim=2


Former-commit-id: a793e8456b664ea0b48f0ba162999f18d06b4c2f
2024-06-11 00:37:17 +08:00
hiyouga
8da149ba40 rename files
Former-commit-id: 74f96efef9bcd63f65d0190c901ff9be54ccd350
2024-06-07 00:09:06 +08:00