🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
update 0-eval_pretrain.py
gongjya8405b08a27f
diff --git a/0-eval_pretrain.py b/0-eval_pretrain.pyindex 32f1dda..edb3ad8 100644--- a/0-eval_pretrain.py+++ b/0-eval_pretrain.py@@ -16,8 +16,7 @@ def count_parameters(model):def init_model(lm_config):- tokenizer = AutoTokenizer.from_pretrained('./model/minimind_tokenizer',- trust_remote_code=True, use_fast=False)+ tokenizer = AutoTokenizer.from_pretrained('./model/minimind_tokenizer')model_from = 1 # 1从权重,2用transformersif model_from == 1: