🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
update model/dataset.py
gongjy3068e5efcc46
diff --git a/model/dataset.py b/model/dataset.pyindex f3cc897..ef58956 100644--- a/model/dataset.py+++ b/model/dataset.py@@ -80,18 +80,18 @@ class SFTDataset(Dataset):#sample = self.df.iloc[index]history = self.safe_eval(sample['history'])- q = sample['q']- a = sample['a']+ q = str(sample['q'])+ a = str(sample['a'])messages = []for history_message in history:if len(history_message) <= 1:continuemessages.append(- {"role": 'user', "content": history_message[0][:self.max_length // 2]}+ {"role": 'user', "content": str(history_message[0])[:self.max_length // 2]})messages.append(- {"role": 'assistant', "content": history_message[1][:self.max_length // 2]}+ {"role": 'assistant', "content": str(history_message[1])[:self.max_length // 2]})messages += [