Merge fc585a9bf9 into 7a6fddd63a

trainer 里的种子此前硬编码为 42（setup_seed(42 + rank) 与每轮的
setup_seed(42 + epoch)），要跑多种子实验只能改源码。新增 --seed，
默认 42，与改动前完全一致。

trainer/train_full_sft.py

@@ -101,6 +101,7 @@ if __name__ == "__main__":
     parser.add_argument('--use_moe', default=0, type=int, choices=[0, 1], help="是否使用MoE架构（0=否，1=是）")
     parser.add_argument('--num_experts', default=4, type=int, help="MoE专家数量（仅use_moe=1时生效）")
     parser.add_argument('--num_experts_per_tok', default=1, type=int, help="每个token路由到的专家数（仅use_moe=1时生效）")
+    parser.add_argument('--seed', default=42, type=int, help="随机种子（DDP下每个rank为seed+rank，每轮为seed+epoch）")
     parser.add_argument("--data_path", type=str, default="../dataset/sft_t2t_mini.jsonl", help="训练数据路径")
     parser.add_argument('--from_weight', default='pretrain', type=str, help="基于哪个权重训练，为none则不基于任何权重训练")
     parser.add_argument('--from_resume', default=0, type=int, choices=[0, 1], help="是否自动检测&续训（0=否，1=是）")
@@ -112,7 +113,7 @@ if __name__ == "__main__":
     # ========== 1. 初始化环境和随机种子 ==========
     local_rank = init_distributed_mode()
     if dist.is_initialized(): args.device = f"cuda:{local_rank}"
-    setup_seed(42 + (dist.get_rank() if dist.is_initialized() else 0))
+    setup_seed(args.seed + (dist.get_rank() if dist.is_initialized() else 0))
     
     # ========== 2. 配置目录、模型参数、检查ckp ==========
     os.makedirs(args.save_dir, exist_ok=True)
@@ -160,7 +161,7 @@ if __name__ == "__main__":
     # ========== 8. 开始训练 ==========
     for epoch in range(start_epoch, args.epochs):
         train_sampler and train_sampler.set_epoch(epoch)
-        setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()
+        setup_seed(args.seed + epoch); indices = torch.randperm(len(train_ds)).tolist()
         skip = start_step if (epoch == start_epoch and start_step > 0) else 0
         batch_sampler = SkipBatchSampler(train_sampler or indices, args.batch_size, skip)
         loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)

trainer/train_pretrain.py

@@ -100,6 +100,7 @@ if __name__ == "__main__":
     parser.add_argument('--use_moe', default=0, type=int, choices=[0, 1], help="是否使用MoE架构（0=否，1=是）")
     parser.add_argument('--num_experts', default=4, type=int, help="MoE专家数量（仅use_moe=1时生效）")
     parser.add_argument('--num_experts_per_tok', default=1, type=int, help="每个token路由到的专家数（仅use_moe=1时生效）")
+    parser.add_argument('--seed', default=42, type=int, help="随机种子（DDP下每个rank为seed+rank，每轮为seed+epoch）")
     parser.add_argument("--data_path", type=str, default="../dataset/pretrain_t2t_mini.jsonl", help="预训练数据路径")
     parser.add_argument('--from_weight', default='none', type=str, help="基于哪个权重训练，为none则从头开始")
     parser.add_argument('--from_resume', default=0, type=int, choices=[0, 1], help="是否自动检测&续训（0=否，1=是）")
@@ -111,7 +112,7 @@ if __name__ == "__main__":
     # ========== 1. 初始化环境和随机种子 ==========
     local_rank = init_distributed_mode()
     if dist.is_initialized(): args.device = f"cuda:{local_rank}"
-    setup_seed(42 + (dist.get_rank() if dist.is_initialized() else 0))
+    setup_seed(args.seed + (dist.get_rank() if dist.is_initialized() else 0))
     
     # ========== 2. 配置目录、模型参数、检查ckp ==========
     os.makedirs(args.save_dir, exist_ok=True)
@@ -159,7 +160,7 @@ if __name__ == "__main__":
     # ========== 8. 开始训练 ==========
     for epoch in range(start_epoch, args.epochs):
         train_sampler and train_sampler.set_epoch(epoch)
-        setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()
+        setup_seed(args.seed + epoch); indices = torch.randperm(len(train_ds)).tolist()
         skip = start_step if (epoch == start_epoch and start_step > 0) else 0
         batch_sampler = SkipBatchSampler(train_sampler or indices, args.batch_size, skip)
         loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)