Upload data3/estimate_budget.sh with huggingface_hub

data3/estimate_budget.sh

@@ -0,0 +1,95 @@
+#!/bin/bash
+# 快速估算不同预算可以生成多少数据
+
+echo "💰 OpenAI API 预算估算工具"
+echo "=========================================="
+echo ""
+echo "基于您的测试结果:"
+echo "  - 2个样本成本: \$0.001819"
+echo "  - 平均每样本: \$0.0009095"
+echo ""
+
+# 从测试结果计算的平均值
+AVG_INPUT_TOKENS=1917
+AVG_OUTPUT_TOKENS=2552
+
+# 标准 API 定价 (gpt-4o-mini)
+STANDARD_INPUT_PRICE=0.15  # per 1M tokens
+STANDARD_OUTPUT_PRICE=0.60 # per 1M tokens
+
+# Batch API 定价 (50% off)
+BATCH_INPUT_PRICE=0.075    # per 1M tokens
+BATCH_OUTPUT_PRICE=0.30    # per 1M tokens
+
+echo "📊 不同预算对比:"
+echo "=========================================="
+printf "%-15s %-15s %-15s %-15s\n" "预算" "标准API" "Batch API" "节省"
+echo "----------------------------------------"
+
+for BUDGET in 1 5 10 20 50 100; do
+    # 计算标准 API 能生成多少
+    STANDARD_COUNT=$(python3 -c "
+import math
+avg_cost_per_sample = ($AVG_INPUT_TOKENS * $STANDARD_INPUT_PRICE / 1_000_000) + ($AVG_OUTPUT_TOKENS * $STANDARD_OUTPUT_PRICE / 1_000_000)
+print(int($BUDGET / avg_cost_per_sample))
+")
+    
+    # 计算 Batch API 能生成多少
+    BATCH_COUNT=$(python3 -c "
+import math
+avg_cost_per_sample = ($AVG_INPUT_TOKENS * $BATCH_INPUT_PRICE / 1_000_000) + ($AVG_OUTPUT_TOKENS * $BATCH_OUTPUT_PRICE / 1_000_000)
+print(int($BUDGET / avg_cost_per_sample))
+")
+    
+    SAVINGS=$((BATCH_COUNT - STANDARD_COUNT))
+    
+    printf "%-15s %-15s %-15s %-15s\n" "\$$BUDGET" "$STANDARD_COUNT" "$BATCH_COUNT" "+$SAVINGS"
+done
+
+echo ""
+echo "🎯 推荐配置 (基于 \$10 预算):"
+echo "=========================================="
+
+# 估算 $10 预算下的详细信息
+python3 -c "
+budget = 10.0
+avg_input = $AVG_INPUT_TOKENS
+avg_output = $AVG_OUTPUT_TOKENS
+
+# Batch API
+batch_input_price = $BATCH_INPUT_PRICE / 1_000_000
+batch_output_price = $BATCH_OUTPUT_PRICE / 1_000_000
+batch_cost_per_sample = (avg_input * batch_input_price) + (avg_output * batch_output_price)
+batch_samples = int(budget / batch_cost_per_sample)
+
+# Standard API  
+std_input_price = $STANDARD_INPUT_PRICE / 1_000_000
+std_output_price = $STANDARD_OUTPUT_PRICE / 1_000_000
+std_cost_per_sample = (avg_input * std_input_price) + (avg_output * std_output_price)
+std_samples = int(budget / std_cost_per_sample)
+
+print(f'使用 Batch API:')
+print(f'  - 可生成样本数: {batch_samples:,}')
+print(f'  - 每样本成本: \${batch_cost_per_sample:.6f}')
+print(f'  - 总输入tokens: {batch_samples * avg_input:,}')
+print(f'  - 总输出tokens: {batch_samples * avg_output:,}')
+print(f'')
+print(f'使用标准 API:')
+print(f'  - 可生成样本数: {std_samples:,}')
+print(f'  - 每样本成本: \${std_cost_per_sample:.6f}')
+print(f'')
+print(f'💰 节省:')
+print(f'  - 多生成样本: {batch_samples - std_samples:,} ({((batch_samples - std_samples) / std_samples * 100):.1f}%)')
+print(f'  - 节省金额: \${budget * 0.5:.2f} (50%)')
+"
+
+echo ""
+echo "📝 使用建议:"
+echo "=========================================="
+echo "1. 先小规模测试 (100-1000 样本)"
+echo "2. 确认质量后再大规模生成"
+echo "3. 使用 --min-score 90+ 保证高质量"
+echo "4. Batch API 处理时间: 通常几小时内完成"
+echo ""
+echo "💡 运行估算命令:"
+echo "   python3 generate_problems_batch.py estimate --num-requests 20000"