| bash scripts/env.sh | |
| CKPT=$1 | |
| OUTPUT=$2 | |
| # torchrun --nnodes=1 \ | |
| # --node_rank=0 \ | |
| # --master_addr=127.0.0.1 \ | |
| # --nproc_per_node=1 \ | |
| # --master_port=63668 \ | |
| torchrun --nproc_per_node $MLP_WORKER_GPU \ | |
| --master_addr $MLP_WORKER_0_HOST \ | |
| --node_rank $MLP_ROLE_INDEX \ | |
| --master_port $MLP_WORKER_0_PORT --nnodes $MLP_WORKER_NUM \ | |
| evaluation/inference_agentthink.py --checkpoint $CKPT --output_name $OUTPUT --out-dir results/agentthink |