fix 2

1 files changed, 45 insertions, 0 deletions

diff --git a/test_first_batch_fix.py b/test_first_batch_fix.py
new file mode 100644
index 0000000..be4e32f
--- /dev/null
+++ b/test_first_batch_fix.py
@@ -0,0 +1,45 @@
+#!/usr/bin/env python3
+"""
+测试第一个批次修复是否有效
+"""
+import sys
+sys.path.append('.')
+
+from dataset.gee_processor import GEEProcessor
+from smart_balanced_dataloader import create_smart_balanced_dataloader
+
+class MockTokenizer:
+    def apply_chat_template(self, messages, tokenize=False, add_generation_prompt=True):
+        return messages[0]["content"]
+
+print("🧪 测试第一个批次修复")
+print("="*50)
+
+# 创建测试数据
+processor = GEEProcessor(MockTokenizer())
+test_data = processor.create_test_data(num_samples=20)
+
+# 测试多次运行，确保第一个批次总是平衡
+print("🔄 测试5次运行，确保第一个批次总是平衡:")
+
+for test_run in range(5):
+    print(f"\n--- 测试运行 {test_run + 1} ---")
+    
+    # 创建数据加载器
+    dataloader = create_smart_balanced_dataloader(test_data, batch_size=2, num_batches=3)
+    
+    # 只关注第一个批次
+    first_batch = next(iter(dataloader))
+    
+    male_count = sum(1 for g in first_batch['gender'] if g == 'male')
+    female_count = sum(1 for g in first_batch['gender'] if g == 'female')
+    
+    print(f"第一批次: {first_batch['gender']}")
+    print(f"统计: male={male_count}, female={female_count}")
+    
+    if male_count > 0 and female_count > 0:
+        print("✅ 第一批次平衡")
+    else:
+        print("❌ 第一批次仍然不平衡！")
+
+print(f"\n🎯 如果以上所有测试都显示'✅ 第一批次平衡'，则修复成功！") 
\ No newline at end of file