#!/bin/bash # 快速查看训练状态 echo "==========================================" echo "🚀 BEVFusion 训练状态快速查看" echo "==========================================" # 检查进程 PROCESS_COUNT=$(ps aux | grep "python.*train.py" | grep -v grep | wc -l) echo "" echo "📊 进程状态:" echo " 训练进程数: $PROCESS_COUNT" if [ $PROCESS_COUNT -gt 0 ]; then echo " 状态: 🟢 运行中" else echo " 状态: 🔴 未运行" fi # 最新进度 echo "" echo "📈 最新训练进度:" tail -5 enhanced_training_6gpus.log | grep "Epoch \[" | tail -1 | \ awk -F'Epoch' '{print " ", $2}' | \ sed 's/\t/ /g' # GPU状态 echo "" echo "💻 GPU使用状态:" nvidia-smi --query-gpu=index,utilization.gpu,memory.used,memory.total \ --format=csv,noheader | head -6 | \ awk -F, '{printf " GPU %s: 利用率 %s, 显存 %s/%s\n", $1, $2, $3, $4}' # Checkpoints echo "" echo "💾 已生成Checkpoints:" ls -lh runs/enhanced_from_epoch19/epoch_*.pth 2>/dev/null | \ awk '{print " ✅", $9, "-", $5}' | grep -v latest echo "" echo "=========================================="