#!/bin/bash
# CODI Training Script for nebulactl run mdl
# Usage: sh sim-cot/CODI/run_nebula.sh

rm -rf ms-swift/ms_swift.egg-info

CLUSTER_CONFIG='./cluster.json'
cat <<EOT > ${CLUSTER_CONFIG}
{
    "worker": {
        "gpu": 100,
        "cpu": 1200,
        "memory": 50000
    }
}
EOT

WORKER_COUNT=1

# Select the configuration file to use
# args="--config_file=./args/gsm_codi_pmps_test.yaml"
# args="--config_file=./args/gsm_codi_pmps_llama_test.yaml"
args="--config_file=./args/gsm_codi_pmps_qwen05_test.yaml"
# args="--config_file=./args/gsm_codi_pmps_larger_model_gsm_test.yaml"

nebulactl run mdl --engine=xdl \
                  --refresh \
                  --ignore=data/*,tools/pycocoevalcap/*,tools/pycocotools/*,.git/*,Coconut/*,CODI/*,CODI_CT/* \
                  --queue=pailitao_algo_high_2 \
                  --worker_count=$WORKER_COUNT \
                  --odps_project=pailitao_search_offline \
                  --entry="./test.py" \
                  --oss_access_id=LTAI5tMZb5HHH6HHgAExAyiL \
                  --oss_access_key=6MVP8JnTE01t4TLUHhPoFV0swZNTrb \
                  --oss_bucket=pailitao-ai \
                  --oss_endpoint=oss-cn-zhangjiakou-internal.aliyuncs.com \
                  --algo_name=pytorch260 \
                  --user_params="$args" \
                  --job_name=qwen05-k24_t0.1_mt0.1_fac0.5_l6_lr8e4_btz256_prj768_ep30_svamp-eval \
                  --file.cluster_file=$CLUSTER_CONFIG \
                  --force \
                  --max_failover_times=2 \
                  --requirements_file_name="./requirements.txt" 

rm ${CLUSTER_CONFIG}