在autodl上微调llama3_8b_instruct #347

New Issue

18800352607 · 2024-11-04T19:05:58+08:00

18800352607 commented

2024-11-04 19:05:58 +08:00

注意：1.如果你的json文件做成了如下所示：

[{
    "conversation":[
        {
            "system": "xxx",
            "input": "xxx",
            "output": "xxx"
        }
    ]
},
{
    "conversation":[
        {
            "system": "xxx",
            "input": "xxx",
            "output": "xxx"
        }
    ]
}]

则在llama3_8b_instruct_qlora_alpaca_e3_copy.py文件里做如下修改：
dataset=dict(type=load_dataset, path=alpaca_en_path), 改为
dataset=dict(type=load_dataset, path='json', data_files=dict(train=alpaca_en_path)),

dataset_map_fn=alpaca_map_fn,改为
dataset_map_fn=None,
2.mpi4py安装时，要用 conda install mpi4py，不然会报错
微调结果：
输入问题：A man is 28 yrs old and his semen counting is 64%, he wants a child so what can he do?，输出：

输入问题：There is this lump towards the left center side of my stomach, what could it be？，输出：

注意：1.如果你的json文件做成了如下所示： ``` [{ "conversation":[ { "system": "xxx", "input": "xxx", "output": "xxx" } ] }, { "conversation":[ { "system": "xxx", "input": "xxx", "output": "xxx" } ] }] ``` 则在llama3_8b_instruct_qlora_alpaca_e3_copy.py文件里做如下修改： `dataset=dict(type=load_dataset, path=alpaca_en_path), ` 改为 `dataset=dict(type=load_dataset, path='json', data_files=dict(train=alpaca_en_path)),` `dataset_map_fn=alpaca_map_fn,`改为 `dataset_map_fn=None,` 2.mpi4py安装时，要用 `conda install mpi4py`，不然会报错微调结果：输入问题：A man is 28 yrs old and his semen counting is 64%, he wants a child so what can he do?，输出： <img width="1502" alt="image" src="/attachments/f554fe5e-6bc7-422c-946c-1408a607bf37"> 输入问题：There is this lump towards the left center side of my stomach, what could it be？，输出： <img width="1513" alt="image" src="/attachments/e06422f8-7867-4bd3-93ea-c6a4598bb9c0">

image.png

36 KiB

image.png

31 KiB

Sign in to join this conversation.