File size: 8,727 Bytes
0dbbebb | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 | #!/bin/bash
# --- initialize conda ---
source /root/miniconda3/etc/profile.d/conda.sh
# --- activate env ---
conda activate bpe_v2
export CUDA_VISIBLE_DEVICES=6
data_path=$1
lr=$2
output_root=$3
project_name=$4
# Model / tokenizer pairs to sweep
MODELS=(
# "/home/n5huang/dna_token/pretrain/models/model_cpu_test_1/checkpoint-100000"
# "/home/n5huang/dna_token/pretrain/models/model_len_reg/checkpoint-100000"
# "/home/n5huang/dna_token/pretrain/models/model_len_2/checkpoint-100000"
# "/home/n5huang/dna_token/pretrain/models/model_tfidf/checkpoint-100000"
# "/root/NaN/dna-tokenizer/pretrain/models/base_2048/checkpoint-100000"
# "/root/NaN/dna-tokenizer/pretrain/models/len2_2048/checkpoint-100000"
"/root/NaN/dna-tokenizer/pretrain/models/len2_3072/checkpoint-100000"
# "/root/NaN/dna-tokenizer/pretrain/models/base_3072/checkpoint-100000"
# "/root/NaN/dna-tokenizer/pretrain/models/base_4096/checkpoint-100000"
# "/root/NaN/dna-tokenizer/pretrain/models/model_len2_4096/checkpoint-100000"
)
TOKENIZERS=(
# "/home/n5huang/dna_token/tokenizer_evaluation/baseline_bpe/tokenizer.json"
# "/home/n5huang/dna_token/tokenizer_evaluation/merge_bpe/merge_tokenizer_unigram_len.json"
# "/home/n5huang/dna_token/tokenizer_evaluation/merge_bpe/merge_tokenizer_unigram_len2.json"
# "/home/n5huang/dna_token/tokenizer_evaluation/merge_bpe/merge_tokenizer_unigram_tf_idf.json"
# "/root/NaN/dna-tokenizer/baseline_bpe/vocab_2048/2048_tokenizer.json"
# "/root/NaN/dna-tokenizer/merge_bpe/vocab_2048/merge_tokenizer_unigram_len2.json"
"/root/NaN/dna-tokenizer/merge_bpe/vocab_3072/merge_tokenizer_unigram_len2.json"
# "/root/NaN/dna-tokenizer/baseline_bpe/vocab_3072/3072_tokenizer.json"
# "/root/NaN/dna-tokenizer/baseline_bpe/vocab_4096/4096_tokenizer.json"
# "/root/NaN/dna-tokenizer/merge_bpe/vocab_4096/merge_tokenizer_unigram_len2.json"
)
# MODEL_NAMES=("len2_2048")
MODEL_NAMES=("len2_3072")
# MODEL_NAMES=("base_2048" "len2_2048" "len2_3072" "base_3072" "base_4096" "len2_4096")
if [ ${#MODELS[@]} -ne ${#TOKENIZERS[@]} ] || [ ${#MODELS[@]} -ne ${#MODEL_NAMES[@]} ]; then
echo "MODELS, TOKENIZERS, MODEL_NAMES must have the same length" >&2
exit 1
fi
echo "The provided data_path is $data_path"
echo "Output root: $output_root"
for seed in 42; do
for idx in "${!MODELS[@]}"; do
model=${MODELS[$idx]}
tokenizer=${TOKENIZERS[$idx]}
model_name=${MODEL_NAMES[$idx]}
# for data in demo_coding_vs_intergenomic_seqs human_nontata_promoters human_enhancers_cohn human_ocr_ensembl; do # length ~200
# run_output_dir=${output_root}/${data}/${model_name}
# mkdir -p "${run_output_dir}"
# echo "Running ${model_name} on ${data}, seed ${seed}, lr ${lr}, output ${run_output_dir}"
# torchrun --nproc_per_node=1 \
# --master_port=${MASTER_PORT:-29500} \
# train.py \
# --model_name_or_path ${model} \
# --tokenizer_path ${tokenizer} \
# --trust_remote_code True \
# --data_path $data_path/$data/split \
# --kmer -1 \
# --run_name ${model_name}_hg38_BPE_${lr}_${data}_seed${seed} \
# --model_max_length 100 \
# --per_device_train_batch_size 128 \
# --per_device_eval_batch_size 128 \
# --gradient_accumulation_steps 1 \
# --learning_rate ${lr} \
# --num_train_epochs 3 \
# --fp16 \
# --save_steps 200 \
# --output_dir ${run_output_dir} \
# --evaluation_strategy steps \
# --eval_steps 200 \
# --warmup_steps 30 \
# --logging_steps 100000 \
# --overwrite_output_dir True \
# --log_level info \
# --seed ${seed} \
# --find_unused_parameters False \
# --project_name ${project_name}
# done
for data in drosophila_enhancers_stark dummy_mouse_enhancers_ensembl human_enhancers_ensembl; do
run_output_dir=${output_root}/${data}/${model_name}
mkdir -p "${run_output_dir}"
echo "Running ${model_name} on ${data}, seed ${seed}, lr ${lr}, output ${run_output_dir}"
torchrun --nproc_per_node=1 \
--master_port=${MASTER_PORT:-29500} \
train.py \
--model_name_or_path ${model} \
--tokenizer_path ${tokenizer} \
--trust_remote_code True \
--data_path $data_path/$data/split \
--kmer -1 \
--run_name ${model_name}_hg38_BPE_${lr}_${data}_seed${seed} \
--model_max_length 512 \
--per_device_train_batch_size 128 \
--per_device_eval_batch_size 128 \
--gradient_accumulation_steps 1 \
--learning_rate ${lr} \
--num_train_epochs 5 \
--fp16 \
--save_steps 200 \
--output_dir ${run_output_dir} \
--evaluation_strategy steps \
--eval_steps 200 \
--warmup_steps 30 \
--logging_steps 100000 \
--overwrite_output_dir True \
--log_level info \
--seed ${seed} \
--find_unused_parameters False \
--project_name ${project_name}
done
# for data in demo_human_or_worm drosophila_enhancers_stark dummy_mouse_enhancers_ensembl human_enhancers_ensembl; do # length mostly 2000+
# run_output_dir=${output_root}/${data}/${model_name}
# mkdir -p "${run_output_dir}"
# echo "Running ${model_name} on ${data}, seed ${seed}, lr ${lr}, output ${run_output_dir}"
# torchrun --nproc_per_node=1 \
# --master_port=${MASTER_PORT:-29500} \
# train.py \
# --model_name_or_path ${model} \
# --tokenizer_path ${tokenizer} \
# --trust_remote_code True \
# --data_path $data_path/$data/split \
# --kmer -1 \
# --run_name ${model_name}_hg38_BPE_${lr}_${data}_seed${seed} \
# --model_max_length 512 \
# --per_device_train_batch_size 128 \
# --per_device_eval_batch_size 128 \
# --gradient_accumulation_steps 1 \
# --learning_rate ${lr} \
# --num_train_epochs 5 \
# --fp16 \
# --save_steps 200 \
# --output_dir ${run_output_dir} \
# --evaluation_strategy steps \
# --eval_steps 200 \
# --warmup_steps 30 \
# --logging_steps 100000 \
# --overwrite_output_dir True \
# --log_level info \
# --seed ${seed} \
# --find_unused_parameters False \
# --project_name ${project_name}
# done
# for data in human_ensembl_regulatory; do # length ~200-700
# run_output_dir=${output_root}/${data}/${model_name}
# mkdir -p "${run_output_dir}"
# echo "Running ${model_name} on ${data}, seed ${seed}, lr ${lr}, output ${run_output_dir}"
# torchrun --nproc_per_node=1 \
# --master_port=${MASTER_PORT:-29500} \
# train.py \
# --model_name_or_path ${model} \
# --tokenizer_path ${tokenizer} \
# --trust_remote_code True \
# --data_path $data_path/$data/split \
# --kmer -1 \
# --run_name ${model_name}_hg38_BPE_${lr}_${data}_seed${seed} \
# --model_max_length 250 \
# --per_device_train_batch_size 128 \
# --per_device_eval_batch_size 128 \
# --gradient_accumulation_steps 1 \
# --learning_rate ${lr} \
# --num_train_epochs 8 \
# --fp16 \
# --save_steps 200 \
# --output_dir ${run_output_dir} \
# --evaluation_strategy steps \
# --eval_steps 200 \
# --warmup_steps 30 \
# --logging_steps 100000 \
# --overwrite_output_dir True \
# --log_level info \
# --seed ${seed} \
# --find_unused_parameters False \
# --project_name ${project_name}
# done
done
done
|