fix: load dataset error

llama-fine-tuning-QLoRA.py

@@ -5,7 +5,8 @@ os.chdir(os.path.dirname(__file__))
 
 # 导入必要的库
 import torch
-from datasets import load_dataset, Dataset
+import json
+from datasets import Dataset
 from transformers import (
     AutoModelForCausalLM,  # 用于加载预训练的语言模型
     AutoTokenizer,  # 用于加载与模型相匹配的分词器
@@ -15,7 +16,8 @@ from transformers import (
     pipeline,  # 用于创建模型的pipeline
     logging,  # 用于记录日志
 )
-from peft import LoraConfig, PeftModel  # 用于配置和加载QLoRA模型
+from huggingface_hub import hf_hub_download
+from peft import LoraConfig # 用于配置和加载QLoRA模型
 from trl import SFTTrainer  # 用于执行监督式微调的Trainer
 
 # 设置预训练模型的名称
@@ -108,12 +110,16 @@ packing = False
 device_map = {"": 0}
 
 # 加载数据集
-dataset = load_dataset(path="json", data_dir="./num_list", data_files="num_list_500_per_sample_100_length.json")
-fine_tune_dataset = []
 print("Loading dataset...")
-for instance in dataset["train"]:
-    prompt = instance["system_prompt"] + "\n\n" + instance["description"] + "\nQuestion: " + instance["data"]["question"] + "\nData: " + instance["data"]["struct_data"]
-    answer = instance["data"]["answer"]
+REPO_ID = "TreeAILab/NumericBench"
+dataset_name = 'num_list/num_list_500_per_sample_100_length.json'
+with open(hf_hub_download(repo_id=REPO_ID, filename=dataset_name, repo_type="dataset")) as f:
+    dataset = json.load(f)
+fine_tune_dataset = []
+
+for instance in dataset["data"]:
+    prompt = dataset["system_prompt"] + "\n\n" + dataset["description"] + "\nQuestion: " + instance["question"] + "\nData: " + instance["struct_data"]
+    answer = instance["answer"]
     completion = f"The answer is {answer}."
     fine_tune_dataset.append({"prompt": prompt, "completion": completion})