From c8972f7e89e6a78d90d2ad09598d44d5a74c17b6 Mon Sep 17 00:00:00 2001 From: Wenyu Zhang Date: Sat, 26 Apr 2025 11:50:23 -0400 Subject: [PATCH] sft should use pretrain model --- trainer/train_full_sft.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/trainer/train_full_sft.py b/trainer/train_full_sft.py index 9b04d5e..2409ba3 100644 --- a/trainer/train_full_sft.py +++ b/trainer/train_full_sft.py @@ -97,7 +97,7 @@ def init_model(lm_config): tokenizer = AutoTokenizer.from_pretrained('../model') model = MiniMindForCausalLM(lm_config) moe_path = '_moe' if lm_config.use_moe else '' - ckp = f'{args.save_dir}/full_sft_{lm_config.hidden_size}{moe_path}.pth' + ckp = f'./out/pretrain_{lm_config.hidden_size}{moe_path}.pth' state_dict = torch.load(ckp, map_location=args.device) model.load_state_dict(state_dict, strict=False)