Commit Graph

27 Commits

Author SHA1 Message Date
LiangSong
95973b5de1 update header 2023-05-17 22:21:46 +07:00
LiangSong
59b79af9d7 add comment 2023-05-09 16:53:05 +08:00
LiangSong
92caa94490 support peft 2023-05-08 22:26:39 +08:00
LiangSong
ec2b4d6ee7 fix split by shard bug 2023-05-08 14:03:05 +08:00
LiangSong
4a1e7bb44b Optimized the structure of configs, added support for deepspeed stage3, reduced memory usage by using Auto class to load models, and added support for training 65B models. 2023-05-06 23:37:17 +08:00
LiangSong
f893a0f5b8 update dataset 2023-05-05 19:23:16 +08:00
LiangSong
85caa97a6a add xP3 dataset and belle_2M 2023-05-05 17:05:41 +08:00
LiangSong
51686b5fb8 add split dataset by shard option to accelerate data loading 2023-05-04 09:20:23 +08:00
LiangSong
f0d41f937b update instruct_config and set all random seed to 42 2023-05-04 08:45:21 +08:00
LiangSong
154456c976 set dataset shuffle seed to 42 2023-05-04 00:31:12 +08:00
LiangSong
0fdca8b949 update readme 2023-04-28 15:01:01 +08:00
LiangSong
49118aad42 update header config and add padding to concat_multiple_sequence 2023-04-27 23:42:11 +08:00
LiangSong
db6cdb51d0 unified pre-training and instrcution-tuning both use train_lm and dataset 2023-04-27 19:42:06 +08:00
LiangSong
0377b43628 update tokenizer to LlamaTokenizer 2023-04-26 18:53:30 +08:00
LiangSong
f41f5558ec update header 2023-04-24 23:19:07 +08:00
LiangSong
f8f4cde228 using huggingface datasets to accelerate training, using open-llama to pretrain 2023-04-24 19:13:53 +08:00
LiangSong
ae0691c509 update utils 2023-04-12 17:15:40 +08:00
LiangSong
c67d365db3 update format 2023-04-07 23:20:20 +08:00
LiangSong
1a731953da update server 2023-04-07 10:04:05 +08:00
LiangSong
bc16df4751 add more instruction data 2023-04-06 03:45:24 +08:00
LiangSong
562067230f update dataset, add concat sequence from multiple docs 2023-04-05 22:42:34 +08:00
LiangSong
b9bc7eaf35 fix long seq bug 2023-03-31 10:12:28 +08:00
LiangSong
a62ac2658f add instruction-tuning 2023-03-30 23:43:12 +08:00
LiangSong
87776f4370 add BucketBySequenceLengthDataset to accelerate training speed 2023-03-28 10:05:27 +08:00
LiangSong
918a8cdc3d reformat code with black 2023-03-27 14:34:59 +08:00
LiangSong
1976aea5f9 update readme 2023-03-27 02:12:59 +08:00
LiangSong
73a81a4205 add high-performance Llama pre-train code 2023-03-26 23:59:53 +08:00