🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
update sft
gongjyf16991d7ec92
diff --git a/3-full_sft.py b/3-full_sft.pyindex 6bee3c1..fdf673a 100644--- a/3-full_sft.py+++ b/3-full_sft.py@@ -14,7 +14,7 @@ from contextlib import nullcontextfrom torch import optimfrom torch.nn.parallel import DistributedDataParallelfrom torch.utils.data import DataLoader, DistributedSampler-from transformers import AutoTokenizer, AutoModel+from transformers import AutoTokenizer, AutoModelForCausalLMfrom model.model import Transformerfrom model.LMConfig import LMConfigfrom model.dataset import SFTDataset@@ -118,7 +118,7 @@ def init_model():state_dict[k[len(unwanted_prefix):]] = state_dict.pop(k)model.load_state_dict(state_dict, strict=False)else:- model = AutoModel.from_pretrained('./minimind', trust_remote_code=True)+ model = AutoModelForCausalLM.from_pretrained('./minimind-v1-small', trust_remote_code=True)Logger(f'LLM总参数量:{count_parameters(model) / 1e6:.3f} 百万')model = model.to(args.device)@@ -143,7 +143,7 @@ if __name__ == "__main__":parser.add_argument("--out_dir", type=str, default="out", help="Output directory")parser.add_argument("--epochs", type=int, default=19, help="Number of epochs")parser.add_argument("--batch_size", type=int, default=32, help="Batch size")- parser.add_argument("--learning_rate", type=float, default=1e-4, help="Learning rate")+ parser.add_argument("--learning_rate", type=float, default=1.5e-4, help="Learning rate")parser.add_argument("--device", type=str, default="cuda:0" if torch.cuda.is_available() else "cpu", help="Device to use")parser.add_argument("--dtype", type=str, default="bfloat16", help="Data type")parser.add_argument("--use_wandb", action="store_true", help="Use Weights & Biases")