🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
[update] shuffle data
jingyaogongf7ffdf1fdb4f
diff --git a/trainer/train_distillation.py b/trainer/train_distillation.pyindex 2b8afc3..2ec6fbe 100644--- a/trainer/train_distillation.py+++ b/trainer/train_distillation.py@@ -221,13 +221,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- train_epoch(epoch, loader, len(loader) + start_step + 1, teacher_model, lm_config_student, start_step, wandb, args.alpha, args.temperature)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None), sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ train_epoch(epoch, loader, len(loader) + skip, teacher_model, lm_config_student, start_step, wandb, args.alpha, args.temperature)+ else:train_epoch(epoch, loader, len(loader), teacher_model, lm_config_student, 0, wandb, args.alpha, args.temperature)# ========== 9. 清理分布进程 ==========diff --git a/trainer/train_dpo.py b/trainer/train_dpo.pyindex 2f37851..31102cf 100644--- a/trainer/train_dpo.py+++ b/trainer/train_dpo.py@@ -205,13 +205,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- train_epoch(epoch, loader, len(loader) + start_step + 1, ref_model, lm_config, start_step, wandb, args.beta)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None), sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ train_epoch(epoch, loader, len(loader) + skip, ref_model, lm_config, start_step, wandb, args.beta)+ else:train_epoch(epoch, loader, len(loader), ref_model, lm_config, 0, wandb, args.beta)# ========== 9. 清理分布进程 ==========diff --git a/trainer/train_full_sft.py b/trainer/train_full_sft.pyindex fc68cd7..7b64a23 100644--- a/trainer/train_full_sft.py+++ b/trainer/train_full_sft.py@@ -148,13 +148,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- train_epoch(epoch, loader, len(loader) + start_step + 1, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None), sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ train_epoch(epoch, loader, len(loader) + skip, start_step, wandb)+ else:train_epoch(epoch, loader, len(loader), 0, wandb)# ========== 9. 清理分布进程 ==========diff --git a/trainer/train_grpo.py b/trainer/train_grpo.pyindex 593780f..bab4223 100755--- a/trainer/train_grpo.py+++ b/trainer/train_grpo.py@@ -290,15 +290,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- grpo_train_epoch(epoch, loader, len(loader) + start_step + 1, ref_model, reward_model, reward_tokenizer, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, pin_memory=True,- drop_last=False, shuffle=(train_sampler is None),- num_workers=args.num_workers, sampler=train_sampler)+ grpo_train_epoch(epoch, loader, len(loader) + skip, ref_model, reward_model, reward_tokenizer, start_step, wandb)+ else:grpo_train_epoch(epoch, loader, len(loader), ref_model, reward_model, reward_tokenizer, 0, wandb)# ========== 9. 清理分布进程 ==========diff --git a/trainer/train_lora.py b/trainer/train_lora.pyindex fdf83bc..83d54c6 100644--- a/trainer/train_lora.py+++ b/trainer/train_lora.py@@ -161,13 +161,14 @@ if __name__ == "__main__":# ========== 9. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- train_epoch(epoch, loader, len(loader) + start_step + 1, lora_params, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None), sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ train_epoch(epoch, loader, len(loader) + skip, lora_params, start_step, wandb)+ else:train_epoch(epoch, loader, len(loader), lora_params, 0, wandb)# ========== 10. 清理分布进程 ==========diff --git a/trainer/train_ppo.py b/trainer/train_ppo.pyindex 02e9e0c..47af75a 100644--- a/trainer/train_ppo.py+++ b/trainer/train_ppo.py@@ -365,15 +365,15 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- ppo_train_epoch(epoch, loader, len(loader) + start_step + 1, old_actor_model, ref_model,+ ppo_train_epoch(epoch, loader, len(loader) + skip, old_actor_model, ref_model,actor_scheduler, critic_scheduler, reward_model, reward_tokenizer, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None),- sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ else:ppo_train_epoch(epoch, loader, len(loader), old_actor_model, ref_model,actor_scheduler, critic_scheduler, reward_model, reward_tokenizer, 0, wandb)diff --git a/trainer/train_pretrain.py b/trainer/train_pretrain.pyindex 7f2a584..ea0ca21 100644--- a/trainer/train_pretrain.py+++ b/trainer/train_pretrain.py@@ -147,13 +147,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- train_epoch(epoch, loader, len(loader) + start_step + 1, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None), sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ train_epoch(epoch, loader, len(loader) + skip, start_step, wandb)+ else:train_epoch(epoch, loader, len(loader), 0, wandb)# ========== 9. 清理分布进程 ==========diff --git a/trainer/train_reason.py b/trainer/train_reason.pyindex 80c6879..eaae014 100644--- a/trainer/train_reason.py+++ b/trainer/train_reason.py@@ -165,13 +165,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- train_epoch(epoch, loader, len(loader) + start_step + 1, tokenizer, lm_config, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=(train_sampler is None), sampler=train_sampler, num_workers=args.num_workers, pin_memory=True)+ train_epoch(epoch, loader, len(loader) + skip, tokenizer, lm_config, start_step, wandb)+ else:train_epoch(epoch, loader, len(loader), tokenizer, lm_config, 0, wandb)# ========== 9. 清理分布进程 ==========diff --git a/trainer/train_spo.py b/trainer/train_spo.pyindex d449491..b6308d5 100755--- a/trainer/train_spo.py+++ b/trainer/train_spo.py@@ -340,15 +340,14 @@ if __name__ == "__main__":# ========== 8. 开始训练 ==========for epoch in range(start_epoch, args.epochs):train_sampler and train_sampler.set_epoch(epoch)- if epoch == start_epoch and start_step > 0: # 第一个epoch且存在检查点- batch_sampler = SkipBatchSampler(train_sampler or range(len(train_ds)), args.batch_size, start_step + 1)- loader = DataLoader(train_ds, batch_sampler=batch_sampler, num_workers=args.num_workers, pin_memory=True)+ setup_seed(42 + epoch); indices = torch.randperm(len(train_ds)).tolist()+ skip = start_step + 1 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)+ if skip > 0:Logger(f'Epoch [{epoch + 1}/{args.epochs}]: 跳过前{start_step}个step,从step {start_step + 1}开始')- spo_train_epoch(epoch, loader, len(loader) + start_step + 1, ref_model, reward_model, reward_tokenizer, value_tracker, start_step, wandb)- else: # 默认从头开始- loader = DataLoader(train_ds, batch_size=args.batch_size, pin_memory=True,- drop_last=False, shuffle=(train_sampler is None),- num_workers=args.num_workers, sampler=train_sampler)+ spo_train_epoch(epoch, loader, len(loader) + skip, ref_model, reward_model, reward_tokenizer, value_tracker, start_step, wandb)+ else:spo_train_epoch(epoch, loader, len(loader), ref_model, reward_model, reward_tokenizer, value_tracker, 0, wandb)# ========== 9. 清理分布进程 ==========