Replies: 1 comment 1 reply
-
Check the documentation before creating an issue: https://qwen.readthedocs.io/zh-cn/latest/framework/function_call.html Qwen2.5-7B-Instruct supports tool calls and you could get better results just by using the original tool calling capabilites instead of ReAct. How to finetune the model depending on how you would like to use the model. If you would like to use the ReAct format, organize the examples in that format. If you would like to use the original format, just use |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
以下是我项目的一些基本信息
agent框架:langgraph使用langgraph.prebuilt.chat_agent_executor.create_react_agent创建的react agent
微调框架:LLama-factory
我主要问题在于:
1.使用langgraph.prebuilt.chat_agent_executor.create_react_agent创建agent是否是一种合适的做法?或者有没有更好的做法?
2.当我自己微调的时候,我该如何构建数据集的内容?具体内容格式是什么样的?我不确定是否正确,以下内容为我当前使用的格式,内容经过处理仅还原我的数据集的格式。
[
{
"conversations": [
{
"from": "human",
"value": "执行操作:在屏幕上播放变形金刚"
},
{
"from": "function_call",
"value": "{"name": "GetVideos", "arguments": {}}"
},
{
"from": "observation",
"value": "[[{"id": "bd919d8dbeef4e308d4e8cb19b2e03c8", "name": "变形金刚"}], 200, null]"
},
{
"from": "function_call",
"value": "{"name": "StartPlay", "arguments": {"signal_id": "bd919d8dbeef4e308d4e8cb19b2e03c8"}}"
},
{
"from": "observation",
"value": "["bd919d8dbeef4e308d4e8cb19b2e03c8", 200, null]"
},
{
"from": "gpt",
"value": "已为您播放变形金刚"
}
],
"tools": "[{"name": "GetVideos", "description": "获取可播放的视频", "returns": "所有视频列表", "parameters": {"type": "object", "properties": {}}},{"name": "StartPlay", "description": "播放视频", "returns": "无", "parameters": {"type": "object", "properties": {"signal_id": {"type": "string", "description": "信号源ID"}}}]"
}
]
Beta Was this translation helpful? Give feedback.
All reactions