Commit Graph

571 Commits

Author SHA1 Message Date
Wing Lian
607a4d33f2 make sure to use train split if loading from hf 2023-05-21 22:04:39 -04:00
Wing Lian
99383f14a3 make one cycle lr div factor configurable 2023-05-21 20:25:06 -04:00
Wing Lian
0f74464652 fix new dataset prompt tokenizers 2023-05-21 18:57:09 -04:00
Wing Lian
e0602a9e54 add missing __init__ 2023-05-21 16:36:41 -04:00
Wing Lian
2809f3f21b pygmalion dataset prompts format, cached tokenized datasets should be hashed on the tokenizer too 2023-05-21 16:16:09 -04:00
Wing Lian
4ea9a66dbd tokenization fixes 2023-05-21 08:33:06 -04:00
Wing Lian
1d5ab84486 optionally be able to specify alpaca or chat style prompts 2023-05-20 18:16:22 -04:00
NanoCode012
641f8012f9 Set half using cfg.fp16 for 4bit 2023-05-20 02:29:31 +09:00
Wing Lian
13650732f8 concise multiple choice and tldr summarize 2023-05-17 11:29:17 -04:00
Wing Lian
8c2f3cb0f8 support for replit lm 2023-05-17 08:49:03 -04:00
Wing Lian
b46bc02f0a add alpaca multiple choice instruct dataset support 2023-05-16 21:45:34 -04:00
NanoCode012
2c73c81348 Add lora_modules_to_save 2023-05-16 19:22:00 +09:00
Wing Lian
5e37144754 fix prompters, especially the sharegpt prompter 2023-05-15 22:15:36 -04:00
Wing Lian
bdbca8fa6c more fixes 2023-05-15 14:07:17 -04:00
Wing Lian
42410c783c more fixes 2023-05-14 09:16:41 -04:00
Wing Lian
aef00b6c13 fix torch_dtype for model load 2023-05-14 08:44:22 -04:00
Wing Lian
0d28df0fd2 move filter to before saving so it doesn't happen everytime, update runpod manual script 2023-05-13 21:51:41 -04:00
Wing Lian
84c7bc4b68 whoops, gt vs lt 2023-05-12 14:03:25 -04:00
Wing Lian
aa3c3f97ae optimize dataloading to use cache, fix model token embedding sizes 2023-05-12 13:53:27 -04:00
NanoCode012
89b7f26b9d Merge branch 'main' into patch-2 2023-05-11 21:18:38 +09:00
Wing Lian
2bc1a5bde1 black formatting 2023-05-10 16:01:08 -04:00
Wing Lian
7a490a4646 various fixes 2023-05-10 16:00:09 -04:00
NanoCode012
813aab378f Fix Trainer() got multiple values for keyword argument 'callbacks' 2023-05-10 18:28:28 +09:00
Wing Lian
e2e68c3965 testing mpt triton 2023-05-09 20:57:40 -04:00
Wing Lian
a27d594788 fix conditional so alpaca doesn't choke 2023-05-09 20:57:07 -04:00
NanoCode012
174b74ddc9 Rename variable to use same convention 2023-05-09 02:49:44 +09:00
NanoCode012
cf681537ec Add CompletionPrompt type 2023-05-09 02:49:44 +09:00
Wing Lian
bd3c5a5cb3 Merge pull request #21 from NanoCode012/patch-1
Fix: Scheduler and optimizer condition
2023-05-08 13:34:44 -04:00
Wing Lian
bcbc99e655 Merge pull request #19 from NanoCode012/feat/callback-save-lora
Feat: Add callback save peft_model on_save
2023-05-08 13:34:07 -04:00
NanoCode012
36aaea02b9 Update trainer.py 2023-05-09 02:01:08 +09:00
NanoCode012
5b6690ac25 Fix condition scheduler 2023-05-09 01:44:12 +09:00
Wing Lian
a125693122 add support for trust_remote_code for mpt models 2023-05-08 12:07:27 -04:00
NanoCode012
cc77bab526 Add callbacks to Trainer 2023-05-09 00:41:19 +09:00
NanoCode012
0d6708bfe4 Add callback save peft_model on_save 2023-05-09 00:38:27 +09:00
Wing Lian
a12fb0a8da Jeopardy bot! (#17)
* support for jeopardy dataset

* commit the final config for jeopardy bot
2023-05-08 03:21:40 -04:00
Wing Lian
a4329b1068 fix #16 load best model setting when using 8bit 2023-05-07 18:30:48 -04:00
Wing Lian
550502b321 use micro batch size for eval size if not specified 2023-05-07 18:26:05 -04:00
Wing Lian
247825bd57 refactor inference, warn if model is frozen 2023-05-07 01:54:15 -04:00
Wing Lian
cb9a887047 Merge pull request #13 from winglian/dev
merge dev branch for various fixes
2023-05-07 01:48:02 -04:00
NanoCode012
0e74b6402e Add eval_batch_size for evaluation 2023-05-06 22:21:24 +09:00
Wing Lian
a10a8265ef fix log sweep lr 2023-05-03 15:06:03 -04:00
Wing Lian
9105935b00 support for multi line inference input, log sweep over learning rates 2023-05-03 13:48:54 -04:00
Wing Lian
7748f3d6da fix adam bnb optimizer grouped parameters, fix peft model 8bit conversion logic, black formatting 2023-05-01 16:31:46 -04:00
Wing Lian
2255bb7f4f support llama-adapter zero init attention 2023-05-01 10:42:21 -04:00
Wing Lian
ad2b48c0fa fdsp config dict fix, todo list, add torchdistx support 2023-04-30 13:32:07 -04:00
Wing Lian
9190ada23a 8bit and deepspeed changes 2023-04-30 06:50:35 -04:00
Wing Lian
6dfdd2dec0 don't load models in 8bit unless they are using an adapter, also fix tokenizer load in exceptional case 2023-04-30 03:19:56 -04:00
Wing Lian
29936bba7f fix fsdp training args 2023-04-30 00:56:28 -04:00
Wing Lian
78821815de fix for zero value warmup steps 2023-04-30 00:34:12 -04:00
Wing Lian
5159d00a86 fix sharegpt tokenization, refactor tokenization debugging 2023-04-30 00:23:53 -04:00