🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
[fix] decide native-vs-transformers loading by directory name, not substring
YoungjuKimcd36b4a01374
diff --git a/eval_llm.py b/eval_llm.pyindex 6339711..2ef2e12 100755--- a/eval_llm.py+++ b/eval_llm.py@@ -1,3 +1,4 @@+import osimport timeimport argparseimport random@@ -11,7 +12,7 @@ warnings.filterwarnings('ignore')def init_model(args):tokenizer = AutoTokenizer.from_pretrained(args.load_from)- if 'model' in args.load_from:+ if os.path.basename(os.path.normpath(args.load_from)) == 'model':model = MiniMindForCausalLM(MiniMindConfig(hidden_size=args.hidden_size,num_hidden_layers=args.num_hidden_layers,diff --git a/scripts/eval_toolcall.py b/scripts/eval_toolcall.pyindex d159eff..a35e2d5 100644--- a/scripts/eval_toolcall.py+++ b/scripts/eval_toolcall.py@@ -56,7 +56,7 @@ TEST_CASES = [def init_model(args):tokenizer = AutoTokenizer.from_pretrained(args.load_from)- if 'model' in args.load_from:+ if os.path.basename(os.path.normpath(args.load_from)) == 'model':model = MiniMindForCausalLM(MiniMindConfig(hidden_size=args.hidden_size, num_hidden_layers=args.num_hidden_layers, use_moe=bool(args.use_moe)))moe_suffix = '_moe' if args.use_moe else ''ckp = f'./{args.save_dir}/{args.weight}_{args.hidden_size}{moe_suffix}.pth'diff --git a/scripts/serve_openai_api.py b/scripts/serve_openai_api.pyindex 09e7fc1..0da37bc 100644--- a/scripts/serve_openai_api.py+++ b/scripts/serve_openai_api.py@@ -27,7 +27,7 @@ app = FastAPI()def init_model(args):tokenizer = AutoTokenizer.from_pretrained(args.load_from)- if 'model' in args.load_from:+ if os.path.basename(os.path.normpath(args.load_from)) == 'model':moe_suffix = '_moe' if args.use_moe else ''ckp = f'../{args.save_dir}/{args.weight}_{args.hidden_size}{moe_suffix}.pth'model = MiniMindForCausalLM(MiniMindConfig(