run.sh

#!/usr/bin/env bash

EXPORT_PREFIX_PATH=""
PRETRAINED_MODEL_PREFIX_PATH=""
QA_MODEL_PATH=""

# Iterative - true or ""
ITERATIVE=""
# Planning - None, operator, equation
PLANNING="operator"
# Reward - fluency, number, qa, combined
REWARD="combined"
# Partial reward - true or ""
QA_PARTIAL_REWARD="True"
# Critic / planning - true or ""
CRITIC=""
EPOCHS=50
# e.g. 10% of last epochs
RL_EPOCHS=0.1

RUN_NAME="QG-rewards-all-${NAME_PREFIX}=${REWARD}-plan=${PLANNING}-iter=${ITERATIVE}"
echo "Running ${RUN_NAME}"

PLANNING=$PLANNING ITERATIVE=$ITERATIVE EXPORT_PREFIX_PATH=$EXPORT_PREFIX_PATH PRETRAINED_MODEL_PREFIX_PATH=$PRETRAINED_MODEL_PREFIX_PATH \
RUN_NAME=$RUN_NAME QA_PARTIAL_REWARD=$QA_PARTIAL_REWARD QA_MODEL_PATH=$QA_MODEL_PATH EPOCHS=$EPOCHS \
REWARD=$REWARD MODEL_IMPORT_PATH=$MODEL_IMPORT_PATH MODEL_CKPT_PATH="" WANDB_MODE="disabled" \
RL_EPOCHS=$RL_EPOCHS VALID_EVERY_EPOCH=$VALID_EVERY_EPOCH BATCH_SIZE=$BATCH_SIZE RETURN_SEQ=$RETURN_SEQ
python question_generation_pipeline.py