#!/bin/bash # FlowCache VBench sampling script # Usage: bash flowcache_vbench.sh [yaml_config_path] # Default config: yaml_config/sample/flowcache_vbench.yaml export PAD_HQ=1 export PAD_DURATION=1 export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True export OFFLOAD_T5_CACHE=true export OFFLOAD_VAE_CACHE=true export TORCH_CUDA_ARCH_LIST="8.9;9.0" MAGI_ROOT=$(git rev-parse --show-toplevel) export PYTHONPATH="$MAGI_ROOT:$PYTHONPATH" export MAGI_ROOT="$MAGI_ROOT" # YAML config file path (can be overridden via command line argument) YAML_CONFIG="${1:-yaml_config/sample/flowcache_vbench.yaml}" if [ ! -f "$YAML_CONFIG" ]; then echo "❌ YAML config file not found: $YAML_CONFIG" exit 1 fi echo "📋 Using YAML config: $YAML_CONFIG" # Create log directory LOG_DIR="./logs" mkdir -p "$LOG_DIR" LOG_FILE="$LOG_DIR/flowcache_vbench_$(date +%Y%m%d_%H%M%S).log" exec > >(tee -a "$LOG_FILE") 2>&1 echo "🚀 Starting multi-GPU benchmark sampling" # Define list of dimensions to process DIMENSIONS=("overall_consistency" "subject_consistency" "scene") echo "🔢 Total dimensions to process: ${#DIMENSIONS[@]}" echo "📋 Dimensions: ${DIMENSIONS[*]}" # Loop through each dimension for DIMENSION in "${DIMENSIONS[@]}"; do echo "🔍 Processing dimension: $DIMENSION" # Use Python to temporarily modify the dimension in YAML, then run sampling python3 -c " import yaml import sys # Read YAML config with open('$YAML_CONFIG', 'r') as f: config = yaml.safe_load(f) # Modify dimension config['dimension'] = '$DIMENSION' # Save to temporary file temp_config = '$YAML_CONFIG.tmp' with open(temp_config, 'w') as f: yaml.dump(config, f, default_flow_style=False) print(temp_config) " > /tmp/temp_config_path.txt TEMP_CONFIG=$(cat /tmp/temp_config_path.txt) python sample_video.py "$TEMP_CONFIG" rm "$TEMP_CONFIG" if [ $? -eq 0 ]; then echo "✅ Completed: $DIMENSION" else echo "❌ Failed: $DIMENSION" echo "🛑 Script paused due to error. Fix the issue and rerun." exit 1 fi echo "---" done echo "🎉 All sampling tasks completed."