HierarchicalContextASA/train_memn2n.sh at main · a2d8a4v/HierarchicalContextASA · GitHub

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
#!/usr/bin/env bash

stage=0
stop_stage=10000

data_dir=nictjle
cache_dir=cache/NICTJLE
glove_path=glove/glove.6B.300d.txt
save_root=output/NICTJLE
log_path=log/NICTJLE
sentaspara=para
model_type=HSG
problem_type=regression
head=linear
wandb=
pred_method=
CUDA=0
BCUDA=0

. parse_options.sh
set -euo pipefail

if [ $stage -le 0 ] && [ $stop_stage -ge 0 ] ; then
    lrs="1e-5 8e-6 6e-6 4e-6 2e-6"
    pred_method=all_s
    # count=0
    # for lr in $lrs; do
    lr="1e-4"
    PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python python3 train.py --cuda --gpu $CUDA --bert_gpu $BCUDA \
        --data_dir $data_dir \
        --cache_dir $cache_dir \
        --embedding_path $glove_path \
        --model PAL \
        --save_root $save_root \
        --log_root $log_path \
        --grad_clip \
        --batch_size 16 \
        --gradient_accumulation_steps 8 \
        --n_epochs 25 \
        --num_workers 0 \
        --sentaspara sent \
        --doc_max_timesteps 88 \
        --problem_type regression \
        --reweight \
        --rw_alpha 2 \
        --head $head \
        --word_embedding \
        --pmi_window_width 5 \
        --interviewer \
        --sent_max_len 1600 \
        --lr $lr \
        --memn2n \
        --verbose_show_in_gpumonitor "lr:$lr stage:$stage"
        # --verbose_show_in_gpumonitor "lr:$lr stage:$stage count:$count"
        # count=$((count + 1))
                # --lr_descent \
                        # --wandb \
    # done
fi