File size: 4,054 Bytes
f4dcc30 46caca0 f4dcc30 46caca0 f4dcc30 46caca0 f4dcc30 46caca0 f4dcc30 46caca0 | 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 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 | # BASE_MODEL="facebook/opt-125m" #"meta-llama/Llama-2-7b-hf facebook/opt-125m
BASE_MODEL="meta-llama/Llama-2-7b-hf"
# DATA_PATH="./data/MetaMathQA.json"
DATA_PATH="./data/MetaMathQA-40K.json"
OUTPUT="output/cp3e5"
export WANDB_PROJECT="HRA_MetaMath395"
# python finetune_32.py \
# --model_name_or_path $BASE_MODEL \
# --output_dir $OUTPUT \
# --hrft_r 32 \
# --init_a 1e-4 \
# --eps 1e-4 \
# --add_orth "none" \
# --lamda 1e-4 \
# --data_path $DATA_PATH \
# --dataset_split "train[:100000]"\
# --dataset_field query response \
# --num_train_epochs 2 \
# --per_device_train_batch_size 8 \
# --gradient_accumulation_steps 4 \
# --save_strategy "steps" \
# --save_steps 0 \
# --save_total_limit 1 \
# --learning_rate 3e-5 \
# --weight_decay 0. \
# --warmup_ratio 0.005 \
# --lr_scheduler_type "cosine" \
# --logging_steps 1000 \
# --bf16 True \
# --tf32 True \
# --report_to "none" \
# wandb sync wandb/latest-run
# OUTPUT="output/cp1e5N"
# python finetune_32.py \
# --model_name_or_path $BASE_MODEL \
# --output_dir $OUTPUT \
# --hrft_r 32 \
# --init_a 1e-4 \
# --eps 1e-4 \
# --add_orth "none" \
# --lamda 1e-4 \
# --data_path $DATA_PATH \
# --dataset_split "train[:100000]"\
# --dataset_field query response \
# --num_train_epochs 2 \
# --per_device_train_batch_size 8 \
# --gradient_accumulation_steps 4 \
# --save_strategy "steps" \
# --save_steps 0 \
# --save_total_limit 1 \
# --learning_rate 1e-5 \
# --weight_decay 0. \
# --warmup_ratio 0.005 \
# --lr_scheduler_type "cosine" \
# --logging_steps 1000 \
# --bf16 True \
# --tf32 True \
# --report_to "wandb"
# wandb sync wandb/latest-run
# OUTPUT="output/cpr1"
# python finetune_32.py \
# --model_name_or_path $BASE_MODEL \
# --output_dir $OUTPUT \
# --hrft_r 1 \
# --init_a 1e-4 \
# --eps 1e-4 \
# --add_orth "none" \
# --lamda 1e-4 \
# --data_path $DATA_PATH \
# --dataset_split "train"\
# --dataset_field query response \
# --num_train_epochs 2 \
# --per_device_train_batch_size 32 \
# --gradient_accumulation_steps 1 \
# --save_strategy "steps" \
# --save_steps 0 \
# --save_total_limit 1 \
# --learning_rate 3e-5 \
# --weight_decay 0. \
# --warmup_ratio 0.005 \
# --lr_scheduler_type "cosine" \
# --logging_steps 1000 \
# --bf16 True \
# --tf32 True \
# --report_to "wandb"
# wandb sync wandb/latest-run
# OUTPUT="output/cpr2"
# python finetune_32.py \
# --model_name_or_path $BASE_MODEL \
# --output_dir $OUTPUT \
# --hrft_r 1 \
# --init_a 1e-4 \
# --eps 1e-4 \
# --add_orth "none" \
# --lamda 1e-4 \
# --data_path $DATA_PATH \
# --dataset_split "train"\
# --dataset_field query response \
# --num_train_epochs 3 \
# --per_device_train_batch_size 32 \
# --gradient_accumulation_steps 1 \
# --save_strategy "steps" \
# --save_steps 0 \
# --save_total_limit 1 \
# --learning_rate 3e-5 \
# --weight_decay 0. \
# --warmup_ratio 0.005 \
# --lr_scheduler_type "cosine" \
# --logging_steps 200 \
# --bf16 True \
# --tf32 True \
# --report_to "wandb"
# wandb sync wandb/latest-run
OUTPUT="output/cms3"
python finetune_32.py \
--model_name_or_path $BASE_MODEL \
--output_dir $OUTPUT \
--hrft_r 32 \
--init_a 1e-4 \
--eps 1e-4 \
--add_orth "none" \
--lamda 1e-4 \
--data_path $DATA_PATH \
--dataset_split "train"\
--dataset_field query response \
--num_train_epochs 2 \
--per_device_train_batch_size 8 \
--gradient_accumulation_steps 4 \
--save_strategy "steps" \
--save_steps 0 \
--save_total_limit 1 \
--learning_rate 1e-5 \
--weight_decay 0. \
--warmup_ratio 0.005 \
--lr_scheduler_type "cosine" \
--logging_steps 200 \
--bf16 True \
--tf32 True \
--report_to "wandb"
date +"%F %T"
# wandb sync wandb/latest-run |