Spaces:
Sleeping
Sleeping
File size: 2,353 Bytes
be3b34d |
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 |
#!/bin/bash
set -e
set -x
# Directory where output files go
OUTDIR=ag4mout
# Directory containing AG4Masses source files
AG4MDIR=ag4masses
# Directory containing external libraries including ag_ckpt_vocab and meliad
AGLIB=aglib
AGDIR=$AG4MDIR/alphageometry
export PYTHONPATH=$PYTHONPATH:$AGDIR:$AGLIB
# stdout, solution is written here
OUTFILE=$OUTDIR/solution.out
# stderr, a lot of information, error message, log etc.
ERRFILE=$OUTDIR/ag.err
# stdout and stderr are written to both ERRFILF and console
exec > >(tee $ERRFILE) 2>&1
# BATCH_SIZE: number of outputs for each LM query
# BEAM_SIZE: size of the breadth-first search queue
# DEPTH: search depth (number of auxilary points to add)
# NWORKERS: number of parallel run worker processes. Rule of thumb: on a 128G machine with 16 logical CPUs,
# use NWORKERS=8, BATCH_SIZE=24.
#
# Memory usage is affected by BATCH_SIZE, NWORKER and complexity of the problem.
# Larger NWORKER and BATCH_SIZE tends to cause out of memory issue
BATCH_SIZE=32
BEAM_SIZE=512
DEPTH=16
NWORKERS=1
#The results in Google's paper can be obtained by setting BATCH_SIZE=32, BEAM_SIZE=512, DEPTH=16
PROB_FILE=$OUTDIR/output.txt
PROB=Output
# alphageometry | ddar
MODEL=ddar
DATA=$AGLIB/ag_ckpt_vocab
MELIAD_PATH=$AGLIB/meliad
export PYTHONPATH=$PYTHONPATH:$MELIAD_PATH
DDAR_ARGS=(
--defs_file=$AGDIR/defs.txt \
--rules_file=$AGDIR/rules.txt \
);
SEARCH_ARGS=(
--beam_size=$BEAM_SIZE
--search_depth=$DEPTH
)
LM_ARGS=(
--ckpt_path=$DATA \
--vocab_path=$DATA/geometry.757.model \
--gin_search_paths=$MELIAD_PATH/transformer/configs,$AGDIR \
--gin_file=base_htrans.gin \
--gin_file=size/medium_150M.gin \
--gin_file=options/positions_t5.gin \
--gin_file=options/lr_cosine_decay.gin \
--gin_file=options/seq_1024_nocache.gin \
--gin_file=geometry_150M_generate.gin \
--gin_param=DecoderOnlyLanguageModelGenerate.output_token_losses=True \
--gin_param=TransformerTaskConfig.batch_size=$BATCH_SIZE \
--gin_param=TransformerTaskConfig.sequence_length=128 \
--gin_param=Trainer.restore_state_variables=False
);
true "=========================================="
python -m alphageometry \
--alsologtostderr \
--problems_file=$PROB_FILE \
--problem_name=$PROB \
--mode=$MODEL \
"${DDAR_ARGS[@]}" \
"${SEARCH_ARGS[@]}" \
"${LM_ARGS[@]}" \
--out_file=$OUTFILE \
--n_workers=$NWORKERS 2>&1
|