File size: 1,417 Bytes
7ed86c3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/bin/bash
#SBATCH -A YOUR_ACCOUNT   # EDIT: your SLURM account
#SBATCH -p gpu   # EDIT: your GPU partition name
#SBATCH --nodes 1
#SBATCH --gpus-per-node 1
#SBATCH --ntasks-per-node 1
#SBATCH -c 64
#SBATCH -t 12:00:00
#SBATCH --job-name=syn-train
#SBATCH -o logs/train-%j.out
#SBATCH -e logs/train-%j.out
# SLURM copies the submitted script into a per-job spool dir before running it on this
# cluster, so locating the repo via ${BASH_SOURCE[0]} resolves to that spool path, not
# the real one ("/var/lib/slurm/..." errors downstream) -- a real failure mode hit
# repeatedly in practice. $SLURM_SUBMIT_DIR is set by sbatch to the directory it was
# invoked from, immune to that copy, and matches this repo's own submit-from-root
# convention (see scripts/slurm/README.md point 6).
STOICHEIA_ROOT="${SLURM_SUBMIT_DIR:-$PWD}"
export STOICHEIA_ROOT
# Train one arm of the biaffine parser (frozen Stoicheia/LemmaDiff + scalar mix + biaffine head).
#   sbatch scripts/slurm/syntax_train.sbatch configs/syntax/parser_char.json
set -euo pipefail
SYN_ROOT=$STOICHEIA_ROOT
source "$SYN_ROOT/env.sh"
CONFIG="${1:?usage: sbatch scripts/train.sbatch <config.json>}"
mkdir -p "$SYN_ROOT/logs"
srun apptainer exec --nv $APPTAINER_BINDS $SIF bash -lc "
  export PYTHONPATH=$SYN_ROOT:$TAGGER_ROOT:$STOICHEIA_ROOT
  export STOICHEIA_CKPT=$STOICHEIA_CKPT
  cd $SYN_ROOT
  python -u -m parser.train --config $CONFIG
"
echo TRAIN_DONE