forked from Beomi/easy-lm-trainer
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.sh
executable file
Β·39 lines (37 loc) Β· 1.31 KB
/
train.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
# Kakao KoGPT 6B μ¬μ© μμ
# python main.py \
# --model_name='kakaobrain/kogpt' \
# --model_revision='KoGPT6B-ryan1.5b-float16' \
# --bos_token='[BOS]' \
# --eos_token='[EOS]' \
# --unk_token='[UNK]' \
# --pad_token='[PAD]' \
# --mask_token='[MASK]' \
# --train_file_path='./data/train.csv' \
# --test_file_path='./data/test.csv' \
# --num_train_epochs=10 \
# --data_text_column='text'
# # Ployglot-Ko 3.8B μ¬μ© μμ
# torchrun --nproc_per_node=4 --master_port=34321 main.py \
# --model_name='EleutherAI/polyglot-ko-5.8b' \
# --train_file_path='data/text_ko_alpaca_data.jsonl' \
# --num_train_epochs=1 \
# --data_text_column='text' \
# --block_size=256 \
# --batch_size=1 \
# --fp16=True \
# --fsdp='auto_wrap' \
# --fsdp_config=fsdp_config.json \
# --deepspeed=ds_config.json
# # --fsdp_transformer_layer_cls_to_wrap='GPTNeoXLayer' \
# Ployglot-Ko 3.8B μ¬μ© μμ
torchrun --nproc_per_node=4 --master_port=34321 main.py \
--model_name='EleutherAI/polyglot-ko-12.8b' \
--train_file_path='data/text_ko_alpaca_data.jsonl' \
--num_train_epochs=1 \
--data_text_column='text' \
--block_size=1024 \
--batch_size=1 \
--fp16=True \
--deepspeed=ds_zero3.json
# --fsdp_transformer_layer_cls_to_wrap='GPTNeoXLayer' \