#!/bin/bash cd eval export NCCL_DEBUG=warn # 定义评估脚本路径 set -x export WANDB_OFFICIAL=1 export WANDB_API_KEY=TO_BE_FILLED TOTAL_NODES=${ARNOLD_WORKER_NUM:-1} # Default to 1 if not set CURRENT_NODE=${ARNOLD_ID:-0} # Default to 0 if not set add_step_0=false temperature=0.0 max_tokens=16000 top_p=0.95 benchmarks="aime24,aime25,amc23,math500,olympiadbench,minerva_math,livemathbench" output_dir="eval_results" overwrite=false n_sampling=1 specific_steps="" while [[ $# -gt 0 ]]; do case $1 in --run_name) RUN_NAME="$2" shift 2 ;; --init_model) INIT_MODEL_PATH="$2" shift 2 ;; --template) template="$2" shift 2 ;; --tp_size) tp_size="$2" shift 2 ;; --temperature) temperature="$2" shift 2 ;; --top_p) top_p="$2" shift 2 ;; --max_tokens) max_tokens="$2" shift 2 ;; --add_step_0) add_step_0="$2" shift 2 ;; --benchmarks) benchmarks="$2" shift 2 ;; --just_wandb) just_wandb="$2" shift 2 ;; --output_dir) output_dir="$2" shift 2 ;; --overwrite) overwrite="$2" shift 2 ;; --n_sampling) n_sampling="$2" shift 2 ;; --specific_steps) specific_steps="$2" shift 2 ;; --seed) seed="$2" shift 2 ;; *) echo "Unknown parameter: $1" exit 1 ;; esac done # Check required parameters if [ -z "$RUN_NAME" ] || [ -z "$INIT_MODEL_PATH" ] || [ -z "$template" ] || [ -z "$tp_size" ]; then echo "Missing required parameters. Usage:" echo "--run_name --init_model --template