c991e2a091
Delete triton_test.py
2023-04-06 02:39:40 +03:00
0d271d5d90
Add files via upload
...
Fix triton kernels
2023-04-06 02:38:06 +03:00
4a2d23aa29
Delete autograd_4bit.py
...
File moved to autograd_4bit module
2023-04-06 02:31:06 +03:00
f20570343f
GPTQv2 support
...
GPTQv2 support.
1. Adds dependency on `triton`
2. Refactors autograd_4bit to include both GPTQv1 and GPTQv2
3. Introduces new environment variable GPTQ_VERSION to select autograd_4bit version
4. Fixes triton kernels
5. Matrix multiplications are in fp16
2023-04-06 02:29:36 +03:00
86387a0a35
update multi gpu support in finetune.py
2023-04-03 23:55:58 +08:00
5655f218ed
add g_idx buffer.\nadd triton matmul utils for future support.
2023-04-02 21:29:06 +08:00
f3a25342e1
fix device_map bug when using lora_apply_dir
2023-03-31 19:44:36 +08:00
00bf0a1e1b
Update README.md
2023-03-31 14:17:35 +08:00
dd0efc721f
Merge pull request #47 from winglian/better-gpt4all
...
fix gpt4all training to more closely match the released logic, other small fixes and optimizations
2023-03-31 11:20:03 +08:00
8791eaee9a
fix gpt4all training to more closely match the released logic, other small fixes and optimizations
2023-03-30 22:40:40 -04:00
878eada8dd
add amp_wrapper for autocast support.
2023-03-30 19:57:19 +08:00
b3c91a5af5
Merge pull request #45 from winglian/fix-missing-bracket
...
fix missing paren
2023-03-30 13:53:55 +08:00
e744aec8bf
fix missing paren
2023-03-29 23:40:30 -04:00
8db4633d84
Update README.md
2023-03-30 11:24:25 +08:00
8a62560e6c
add offload support
2023-03-30 11:21:21 +08:00
32976f91c4
Merge pull request #42 from winglian/multigpu-fix
...
better multi-gpu support, support gpt4all training data
2023-03-30 00:03:27 +08:00
b7361da58a
better multi-gpu support, support gpt4all training data
2023-03-29 11:21:47 -04:00
0fdae9224c
optimized groupsize backward for performance
2023-03-29 17:44:51 +08:00
5986649b37
Update README.md
2023-03-29 14:46:28 +08:00
1c02d4262d
add resume checkpoint to continue a training
2023-03-29 14:35:39 +08:00
2a1cb42966
add padding support as an option
2023-03-29 11:20:16 +08:00
cff57ebfa4
Merge pull request #39 from winglian/fix-prompt-eos-token
...
properly include the eos token so inference doesn't blabber on
2023-03-29 10:35:46 +08:00
daad59f8ef
properly include the eos token so inference doesn't blabber on
2023-03-28 20:53:16 -04:00
1719bd0ce3
fix bug
2023-03-29 08:09:40 +08:00
1043ded7d9
Merge branch 'main' of github.com:johnsmith0031/alpaca_lora_4bit
2023-03-29 01:26:20 +08:00
d28ee06202
fix bug
2023-03-29 01:25:37 +08:00
b5e3dae573
Merge pull request #34 from winglian/v2-fixes
...
fixes for most recent update
2023-03-28 23:49:56 +08:00
b47da33084
fixes for most recent update
2023-03-28 10:56:35 -04:00
234004ceb5
fix bug
2023-03-28 22:05:18 +08:00
f26615fc0c
fix bug
2023-03-28 21:47:22 +08:00
0768d0fdff
update finetune data format
2023-03-28 21:45:33 +08:00
8a6c8661df
Merge branch 'main' of github.com:johnsmith0031/alpaca_lora_4bit
...
merged
2023-03-28 21:14:35 +08:00
211af574b6
fix bug
2023-03-28 21:12:51 +08:00
ac07457473
Update README.md
2023-03-28 20:44:02 +08:00
bff039de95
add v2 model support
2023-03-28 20:33:55 +08:00
667e43cb5b
Merge pull request #30 from winglian/features/python-fixes
...
backwards support for pre-py3.10, add datasets requirement used in train
2023-03-28 09:34:50 +08:00
101d314bd9
add missing dependency to train with LlamaTokenizer
2023-03-27 16:13:46 -04:00
62e54ac1c7
backwards support for pre-py3.10, add datasets requirement used in train
2023-03-27 16:08:20 -04:00
6c8c07e7ad
Update README.md
2023-03-27 18:03:28 +08:00
cf94d7af68
Update README.md
2023-03-27 17:52:35 +08:00
1ca9b8abf8
Update README.md
2023-03-27 17:51:04 +08:00
0b5b376de1
Merge pull request #23 from sterlind/star/repos
...
Get dependencies straight from pip!
2023-03-27 17:47:39 +08:00
399c3d124e
Tested and should be ready!
2023-03-25 20:52:38 -07:00
a2a4c1d117
Remove gitmodules.
2023-03-25 20:23:46 -07:00
96440c8717
Removing submodules actually.
2023-03-25 20:20:38 -07:00
17f3da744c
Fix repos.
2023-03-25 20:16:48 -07:00
8e705eddcb
Add GPTQ and peft as submodules.
2023-03-25 20:02:19 -07:00
82dd6dd13e
Merge pull request #20 from kooshi/multi-gpu
...
Enable model parallelism and distributed data parallelism for multi-gpu setups
2023-03-25 15:06:01 +08:00
8e471516b8
distributed data parallelism with torchrun
2023-03-24 23:56:06 -05:00
2bc64597aa
model parallelism
2023-03-24 23:03:43 -05:00
Andrey Glushenkov
John Smith
Wing Lian
Star Dorminey
kooshi