🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
update convert script
jingyaogong2e118e9d3d2e
diff --git a/scripts/convert_model.py b/scripts/convert_model.pyindex 685cf9e..a6e22f8 100644--- a/scripts/convert_model.py+++ b/scripts/convert_model.py@@ -62,11 +62,11 @@ def convert_transformers2torch(transformers_path, torch_path):if __name__ == '__main__':- lm_config = MiniMindConfig(hidden_size=768, num_hidden_layers=16, max_seq_len=8192, use_moe=True)+ lm_config = MiniMindConfig(hidden_size=768, num_hidden_layers=16, max_seq_len=8192, use_moe=False)torch_path = f"../out/full_sft_{lm_config.hidden_size}{'_moe' if lm_config.use_moe else ''}.pth"- transformers_path = '../MiniMind2-MoE'+ transformers_path = '../MiniMind2'convert_torch2transformers_minimind(torch_path, transformers_path)