diff options
| -rw-r--r-- | smart_balanced_dataloader.py | 218 | ||||
| -rw-r--r-- | train_gee_smart.py | 279 |
2 files changed, 497 insertions, 0 deletions
diff --git a/smart_balanced_dataloader.py b/smart_balanced_dataloader.py new file mode 100644 index 0000000..54eb630 --- /dev/null +++ b/smart_balanced_dataloader.py @@ -0,0 +1,218 @@ +#!/usr/bin/env python3 +""" +智能平衡数据加载器 - 自动检测并重新生成不平衡批次 +""" +import torch +from torch.utils.data import Dataset, DataLoader +import random +from typing import List, Dict +import warnings + +class SmartBalancedGEEDataset(Dataset): + def __init__(self, data: List[Dict]): + self.data = data + # 按性别分组 + self.male_data = [item for item in data if item['gender'] == 'male'] + self.female_data = [item for item in data if item['gender'] == 'female'] + + print(f"📊 数据分布: male={len(self.male_data)}, female={len(self.female_data)}") + + # 确保有足够的数据 + if len(self.male_data) == 0 or len(self.female_data) == 0: + raise ValueError("数据中必须包含男性和女性样本") + + # 确保有足够的样本进行重新生成 + min_samples = min(len(self.male_data), len(self.female_data)) + if min_samples < 10: + warnings.warn(f"样本数量较少 (min={min_samples})，可能影响批次生成质量") + + def __len__(self): + return len(self.data) + + def __getitem__(self, idx): + return self.data[idx] + + def generate_balanced_batch(self, batch_size: int, max_retries: int = 10): + """生成一个平衡的批次，如果失败会重试""" + if batch_size < 2: + raise ValueError("batch_size必须>=2才能保证性别平衡") + + # 计算每个性别需要的样本数 + male_per_batch = batch_size // 2 + female_per_batch = batch_size - male_per_batch + + for attempt in range(max_retries): + try: + batch = [] + + # 随机选择男性样本 + if len(self.male_data) >= male_per_batch: + male_samples = random.sample(self.male_data, male_per_batch) + batch.extend(male_samples) + else: + # 如果男性样本不够，用替换的方式采样 + male_samples = random.choices(self.male_data, k=male_per_batch) + batch.extend(male_samples) + + # 随机选择女性样本 + if len(self.female_data) >= female_per_batch: + female_samples = random.sample(self.female_data, female_per_batch) + batch.extend(female_samples) + else: + # 如果女性样本不够，用替换的方式采样 + female_samples = random.choices(self.female_data, k=female_per_batch) + batch.extend(female_samples) + + # 打乱批次内的顺序 + random.shuffle(batch) + + # 验证批次平衡性 + male_count = sum(1 for item in batch if item['gender'] == 'male') + female_count = sum(1 for item in batch if item['gender'] == 'female') + + if male_count > 0 and female_count > 0: + return batch + else: + print(f"⚠️ 尝试 {attempt+1}: 批次不平衡 (male={male_count}, female={female_count})，重新生成...") + + except Exception as e: + print(f"⚠️ 尝试 {attempt+1} 失败: {e}") + continue + + # 如果所有尝试都失败，返回一个强制平衡的批次 + print(f"❌ {max_retries} 次尝试后仍然失败，强制生成平衡批次") + return self._force_balanced_batch(batch_size) + + def _force_balanced_batch(self, batch_size: int): + """强制生成一个平衡批次""" + batch = [] + male_per_batch = batch_size // 2 + female_per_batch = batch_size - male_per_batch + + # 强制添加男性样本（允许重复） + for _ in range(male_per_batch): + batch.append(random.choice(self.male_data)) + + # 强制添加女性样本（允许重复） + for _ in range(female_per_batch): + batch.append(random.choice(self.female_data)) + + random.shuffle(batch) + return batch + +class SmartBalancedDataLoader: + """智能平衡数据加载器""" + def __init__(self, dataset: SmartBalancedGEEDataset, batch_size: int, num_batches: int): + self.dataset = dataset + self.batch_size = batch_size + self.num_batches = num_batches + self.current_idx = 0 + + print(f"🧠 智能数据加载器初始化: batch_size={batch_size}, num_batches={num_batches}") + + def __iter__(self): + self.current_idx = 0 + return self + + def __next__(self): + if self.current_idx >= self.num_batches: + raise StopIteration + + # 动态生成平衡批次 + batch = self.dataset.generate_balanced_batch(self.batch_size) + self.current_idx += 1 + + # 应用collate函数并验证 + return self._smart_collate(batch) + + def _smart_collate(self, batch, max_regenerate: int = 3): + """智能collate函数，如果检测到不平衡会重新生成""" + inputs = [item["input"] for item in batch] + genders = [item["gender"] for item in batch] + + # 检查批次平衡性 + male_count = sum(1 for g in genders if g == 'male') + female_count = sum(1 for g in genders if g == 'female') + + # 如果不平衡，尝试重新生成 + regenerate_count = 0 + while (male_count == 0 or female_count == 0) and regenerate_count < max_regenerate: + print(f"🔄 检测到不平衡批次 (male={male_count}, female={female_count})，重新生成...") + + # 重新生成批次 + batch = self.dataset.generate_balanced_batch(self.batch_size) + inputs = [item["input"] for item in batch] + genders = [item["gender"] for item in batch] + + # 重新检查 + male_count = sum(1 for g in genders if g == 'male') + female_count = sum(1 for g in genders if g == 'female') + regenerate_count += 1 + + # 最终检查 + if male_count == 0: + print("❌ 最终警告: 批次中没有男性样本！") + if female_count == 0: + print("❌ 最终警告: 批次中没有女性样本！") + + if male_count > 0 and female_count > 0: + print(f"✅ 平衡批次: male={male_count}, female={female_count}") + + return { + "input": inputs, + "gender": genders + } + +def create_smart_balanced_dataloader(data: List[Dict], batch_size: int, num_batches: int = 10): + """创建智能平衡数据加载器""" + + if batch_size < 2: + print("⚠️ 警告: batch_size < 2，无法保证性别平衡") + # 回退到普通DataLoader + dataset = SmartBalancedGEEDataset(data) + return DataLoader(dataset, batch_size=batch_size, shuffle=True) + + dataset = SmartBalancedGEEDataset(data) + + print(f"🧠 创建智能平衡数据加载器") + print(f" 批次大小: {batch_size}") + print(f" 批次数量: {num_batches}") + print(f" 每批次配置: male={batch_size//2}, female={batch_size-batch_size//2}") + + return SmartBalancedDataLoader(dataset, batch_size, num_batches) + +# 测试函数 +if __name__ == "__main__": + # 测试智能平衡数据加载器 + import sys + sys.path.append('.') + from dataset.gee_processor import GEEProcessor + + class MockTokenizer: + def apply_chat_template(self, messages, tokenize=False, add_generation_prompt=True): + return messages[0]["content"] + + processor = GEEProcessor(MockTokenizer()) + test_data = processor.create_test_data(num_samples=20) + + print("🧪 测试智能平衡数据加载器") + dataloader = create_smart_balanced_dataloader(test_data, batch_size=4, num_batches=5) + + for i, batch in enumerate(dataloader): + print(f"\n=== 批次 {i+1} ===") + print(f"输入数量: {len(batch['input'])}") + print(f"性别分布: {batch['gender']}") + + # 验证平衡性 + male_count = sum(1 for g in batch['gender'] if g == 'male') + female_count = sum(1 for g in batch['gender'] if g == 'female') + + if male_count > 0 and female_count > 0: + print(f"✅ 批次完美平衡: male={male_count}, female={female_count}") + else: + print(f"❌ 批次仍然不平衡: male={male_count}, female={female_count}") + + if i >= 4: # 测试5个批次 + break + + print("\n�� 智能平衡数据加载器测试完成!")
\ No newline at end of file diff --git a/train_gee_smart.py b/train_gee_smart.py new file mode 100644 index 0000000..d368762 --- /dev/null +++ b/train_gee_smart.py @@ -0,0 +1,279 @@ +#!/usr/bin/env python3 +""" +GEE训练脚本 - 使用智能平衡数据加载器 +绝对保证每个批次都包含男女样本 +""" +import argparse +import os +import torch +import torch.nn.functional as F +from torch.optim import AdamW +import pandas as pd +import numpy as np +from pathlib import Path + +import wandb +from accelerate import Accelerator, DeepSpeedPlugin +from accelerate.utils import set_seed +from transformers import AutoConfig, AutoTokenizer, AutoModelForCausalLM + +# 导入自定义模块 +import sys +sys.path.append('.') +from dataset.gee_processor import GEEProcessor +from losses.gee_loss import GEELoss, gender_to_label +from smart_balanced_dataloader import create_smart_balanced_dataloader + +os.environ.setdefault("NCCL_TIMEOUT", "2700") +os.environ.setdefault("TORCH_NCCL_HEARTBEAT_TIMEOUT_SEC", "2700") + +def parse_args(): + parser = argparse.ArgumentParser() + # GEE相关参数 + parser.add_argument('--lambda_weight', type=float, default=0.5, help='GEE lambda weight') + parser.add_argument('--use_l1', action='store_true', help='Use L1 loss instead of L2') + parser.add_argument('--auto_anneal', action='store_true', help='Use automatic annealing') + + # 模型参数 + parser.add_argument('--model_name', type=str, default='Qwen2.5-Math-1.5B-Instruct', help='Model name') + parser.add_argument('--model_path', type=str, required=True, help='Model path') + parser.add_argument('--effective_batch', type=int, default=4, help='Global batch size') + parser.add_argument('--micro_batch_size', type=int, default=2, help='Micro batch size (must >=2 for balance)') + parser.add_argument('--learning_rate', type=float, default=2e-5, help='Learning rate') + parser.add_argument('--max_steps', type=int, default=10, help='Maximum training steps') + parser.add_argument('--sample_temp', type=float, default=0.7, help='Generation temperature') + + # 运行参数 + parser.add_argument('--run_name', type=str, default='gee_smart_balanced', help='Run name') + parser.add_argument('--wandb_project', type=str, default='one-shot-gee', help='W&B project name') + parser.add_argument('--use_test_data', action='store_true', help='Use synthetic test data') + parser.add_argument('--seed', type=int, default=42, help='Random seed') + parser.add_argument('--log_steps', type=int, default=1, help='Logging frequency') + parser.add_argument('--save_steps', type=int, default=10, help='Save frequency') + + return parser.parse_args() + +def main(): + args = parse_args() + set_seed(args.seed) + + # 强制检查batch_size + if args.micro_batch_size < 2: + print("❌ 错误: micro_batch_size必须>=2才能保证性别平衡！") + print("请使用: --micro_batch_size 2 或更大") + return + + # DeepSpeed配置 + ds_config = { + "train_micro_batch_size_per_gpu": args.micro_batch_size, + "train_batch_size": args.effective_batch, + "gradient_accumulation_steps": max(1, args.effective_batch // args.micro_batch_size), + "bf16": {"enabled": True}, + "zero_optimization": { + "stage": 2, + "offload_optimizer": {"device": "cpu"}, + "offload_param": {"device": "none"} + }, + "gradient_clipping": 1.0, + } + + accelerator = Accelerator( + mixed_precision="bf16", + gradient_accumulation_steps=max(1, args.effective_batch // args.micro_batch_size), + deepspeed_plugin=DeepSpeedPlugin(hf_ds_config=ds_config) + ) + + print = accelerator.print + print(f"🧠 开始智能GEE训练 - {args.run_name}") + print(f"📊 配置信息:") + print(f" 批次大小: micro={args.micro_batch_size}, effective={args.effective_batch}") + print(f" Lambda权重: {args.lambda_weight}") + print(f" 最大步数: {args.max_steps}") + print(f" 智能平衡: ✅ 启用") + + # 加载模型 + model_path = args.model_path + config = AutoConfig.from_pretrained(model_path, trust_remote_code=True) + config.use_cache = False + model = AutoModelForCausalLM.from_pretrained(model_path, config=config, trust_remote_code=True) + model.gradient_checkpointing_enable() + + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + if tokenizer.pad_token is None: + tokenizer.pad_token = tokenizer.eos_token + + # 初始化GEE处理器和损失函数 + gee_processor = GEEProcessor(tokenizer) + gee_loss_fn = GEELoss(lambda_weight=args.lambda_weight, use_l1=args.use_l1) + + if accelerator.is_main_process: + wandb.init(project=args.wandb_project, name=args.run_name, config=vars(args)) + + # 准备数据 - 使用智能平衡数据加载器 + if args.use_test_data: + print("📊 使用合成测试数据...") + train_data = gee_processor.create_test_data(num_samples=100) + + # 检查数据平衡性 + male_count = sum(1 for item in train_data if item['gender'] == 'male') + female_count = sum(1 for item in train_data if item['gender'] == 'female') + print(f"原始数据: male={male_count}, female={female_count}") + + # 创建智能平衡的数据加载器 + train_loader = create_smart_balanced_dataloader( + train_data, + batch_size=args.micro_batch_size, + num_batches=args.max_steps + 5 # 额外的批次确保有足够数据 + ) + else: + print("❌ 请使用 --use_test_data 进行测试") + return + + optimizer = AdamW(model.parameters(), lr=args.learning_rate) + + # 注意：智能数据加载器不需要accelerator.prepare处理 + model, optimizer = accelerator.prepare(model, optimizer) + + print(f"🎯 开始训练...") + print(f" 期望看到: 每个批次都显示 '✅ 平衡批次'") + print(f" 不应该看到: '❌ 警告: 批次中没有男性/女性样本'") + + # 开始训练 + model.train() + initial_entropy_gap = None + successful_steps = 0 + failed_steps = 0 + + for step, batch in enumerate(train_loader, start=1): + if step > args.max_steps: + print(f"🛑 达到最大步数 {args.max_steps}，训练结束") + break + + with accelerator.accumulate(model): + try: + # 验证批次平衡性 + male_count = sum(1 for g in batch["gender"] if g == 'male') + female_count = sum(1 for g in batch["gender"] if g == 'female') + + if male_count == 0 or female_count == 0: + print(f"💥 Step {step}: 智能加载器失败！male={male_count}, female={female_count}") + failed_steps += 1 + continue + else: + successful_steps += 1 + + # 准备输入 + inputs = tokenizer( + batch["input"], + return_tensors="pt", + padding="longest", + truncation=True, + max_length=1024 + ).to(accelerator.device) + + # 生成回答 + with torch.no_grad(): + gen_ids = accelerator.unwrap_model(model).generate( + **inputs, + max_new_tokens=128, + do_sample=True, + top_p=0.95, + temperature=args.sample_temp, + synced_gpus=True, + repetition_penalty=1.15, + pad_token_id=tokenizer.pad_token_id, + use_cache=False + ) + + # 准备完整序列 + seq = torch.cat([inputs.input_ids, gen_ids[:, inputs.input_ids.shape[1]:]], dim=1) + pad_mask = seq.ne(tokenizer.pad_token_id) + prompt_lengths = pad_mask[:, :inputs.input_ids.shape[1]].sum(-1) + + # 计算logits和熵 + logits = model(seq, attention_mask=pad_mask).logits + H_tok = gee_loss_fn.compute_token_entropy(logits, pad_mask) + H_i = gee_loss_fn.compute_sample_entropy(H_tok, prompt_lengths) + + # 准备性别标签 + gender_labels = torch.tensor([ + gender_to_label(g) for g in batch["gender"] + ], device=accelerator.device) + + # 计算GEE损失 + loss, metrics = gee_loss_fn.compute_gee_loss(H_i, gender_labels) + + # 自动退火（可选） + if args.auto_anneal and initial_entropy_gap is None: + initial_entropy_gap = metrics['entropy_gap'] + + if args.auto_anneal and initial_entropy_gap > 0: + current_gap = metrics['entropy_gap'] + anneal_factor = current_gap / initial_entropy_gap + new_lambda = args.lambda_weight * anneal_factor + gee_loss_fn.update_lambda(new_lambda) + metrics['lambda_weight'] = new_lambda + + # 反向传播 + accelerator.backward(loss) + accelerator.clip_grad_norm_(model.parameters(), 1.0) + optimizer.step() + optimizer.zero_grad() + + # 日志记录 + if accelerator.is_main_process: + if step % args.log_steps == 0: + print(f"🎯 Step {step} | loss={loss.item():.6f} | " + f"gap={metrics['entropy_gap']:.6f} | " + f"H_male={metrics['H_male']:.6f} | " + f"H_female={metrics['H_female']:.6f} | " + f"批次[{male_count}M,{female_count}F]") + + # 添加成功率统计 + success_rate = successful_steps / (successful_steps + failed_steps) * 100 + metrics['success_rate'] = success_rate + + wandb.log({"step": step, **metrics}) + + # 保存检查点 + if accelerator.is_main_process and step % args.save_steps == 0: + ckpt = Path(f"checkpoints/{args.model_name}/{args.run_name}") / f"step_{step}" + ckpt.mkdir(parents=True, exist_ok=True) + accelerator.unwrap_model(model).save_pretrained(ckpt, safe_serialization=True) + tokenizer.save_pretrained(ckpt) + print(f"💾 检查点已保存: {ckpt}") + + except Exception as e: + print(f"❌ 训练步骤 {step} 出错: {e}") + failed_steps += 1 + continue + + if accelerator.is_main_process: + # 最终统计 + total_steps = successful_steps + failed_steps + success_rate = successful_steps / total_steps * 100 if total_steps > 0 else 0 + + print(f"\n🎉 智能GEE训练完成!") + print(f"📊 最终统计:") + print(f" 成功步数: {successful_steps}") + print(f" 失败步数: {failed_steps}") + print(f" 成功率: {success_rate:.1f}%") + + if success_rate >= 95: + print("✅ 智能平衡数据加载器工作完美！") + elif success_rate >= 80: + print("⚠️ 智能平衡数据加载器基本正常，偶有问题") + else: + print("❌ 智能平衡数据加载器需要进一步优化") + + # 保存最终模型 + final = Path(f"checkpoints/{args.model_name}/{args.run_name}") / "final" + final.mkdir(parents=True, exist_ok=True) + accelerator.unwrap_model(model).save_pretrained(final, safe_serialization=True) + tokenizer.save_pretrained(final) + print(f"💾 最终模型已保存: {final}") + + wandb.finish() + +if __name__ == "__main__": + main()
\ No newline at end of file |