#!/usr/bin/env bash
set -euo pipefail

RUN_ID="${RUN_ID:-taotern-200m-until-selection-$(date +%Y%m%d-%H%M%S)}"
JOB_ROOT="${JOB_ROOT:-/home/student/YouZheng/jobs/taotern}"
REMOTE_REPO="${REMOTE_REPO:-$(pwd)}"
PYTHON_BIN="${PYTHON_BIN:-/home/student/.venv/bin/python}"
SSM_REPO_PATH="${SSM_REPO_PATH:-/home/student/YouZheng/gamma_ssm_repo}"
DATA_PATH="${DATA_PATH:-/home/student/Data/TaoData/pretrain.jsonl}"
TOKENIZER_PATH="${TOKENIZER_PATH:-/home/student/YouZheng/tokenizers/taodata_pilot_8k/tokenizer.model}"
SEQ_LEN="${SEQ_LEN:-512}"
BATCH_SIZE="${BATCH_SIZE:-8}"
PILOT_TOKENS="${PILOT_TOKENS:-300000000}"
SERIOUS_TOKENS="${SERIOUS_TOKENS:-1000000000}"
PILOT_EVAL_BATCHES="${PILOT_EVAL_BATCHES:-64}"
SERIOUS_EVAL_BATCHES="${SERIOUS_EVAL_BATCHES:-128}"
MAX_TOKENS="${MAX_TOKENS:-50000000}"
MAX_RECORDS="${MAX_RECORDS:-100000}"
LEARNING_RATE="${LEARNING_RATE:-0.0003}"
WEIGHT_DECAY="${WEIGHT_DECAY:-0.01}"

JOB_COMMAND="REMOTE_REPO=$REMOTE_REPO PYTHON_BIN=$PYTHON_BIN SSM_REPO_PATH=$SSM_REPO_PATH DATA_PATH=$DATA_PATH TOKENIZER_PATH=$TOKENIZER_PATH SEQ_LEN=$SEQ_LEN BATCH_SIZE=$BATCH_SIZE PILOT_TOKENS=$PILOT_TOKENS SERIOUS_TOKENS=$SERIOUS_TOKENS PILOT_EVAL_BATCHES=$PILOT_EVAL_BATCHES SERIOUS_EVAL_BATCHES=$SERIOUS_EVAL_BATCHES MAX_TOKENS=$MAX_TOKENS MAX_RECORDS=$MAX_RECORDS LEARNING_RATE=$LEARNING_RATE WEIGHT_DECAY=$WEIGHT_DECAY bash scripts/remote/run_200m_until_selection.sh"

export RUN_ID JOB_ROOT JOB_COMMAND
export OUTPUT_DIR="${OUTPUT_DIR:-$JOB_ROOT/$RUN_ID/outputs}"
export CHECKPOINT_DIR="${CHECKPOINT_DIR:-$JOB_ROOT/$RUN_ID/checkpoints}"

bash scripts/remote/submit_detached_job.sh