在训练管理中基于function call进行多机多卡微调Llama3-8B-Instruct模型 #75

New Issue

12019701659cs · 2024-09-12T18:10:14+08:00

12019701659cs commented

2024-09-12 18:10:14 +08:00

max_length = 1664的情况

微调前

<|start_header_id|>system<|end_header_id|>

You are a assistant who can utilize external tools.
[{'name': 'ArxivSearch.get_arxiv_article_information','description': 'Run Arxiv search and get the article meta information.','parameters': [{'name': 'query', 'type': 'STRING', 'description':'the content of search query'}],'required': ['query'],'return_data':[{'name': 'content', 'description': 'a list of 3 arxiv search papers', 'type': 'STRING'}],'parameter_description': 'If you call this tool, you must pass arguments in the JSON format{key: value}, where the key is the parameter name.'}]
To use a tool, please use the following format:

Thought:Think what you need to solve, do you need to use tools?
Action:the tool name, should be one of [['ArxivSearch']]
Action Input:the input to the action

The response after utilizing tools should using the following format:

Response:the results after call the tool.

If you already know the answer, or you do not need to use tools,
please using the following format to reply:

Thought:the thought process to get the final answer
Final Answer:final answer


Begin!<|eot_id|><|start_header_id|>user<|end_header_id|>

Please help me search the InternLM2 Technical Report.<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Thought: I can help you search for the InternLM2 Technical Report using the ArxivSearch tool.

Action: ArxivSearch.get_arxiv_article_information

Action Input: query="InternLM2 Technical Report"

Response:

[
  {

微调后

<|start_header_id|>system<|end_header_id|>

You are a assistant who can utilize external tools.
[{'name': 'ArxivSearch.get_arxiv_article_information','description': 'Run Arxiv search and get the article meta information.','parameters': [{'name': 'query', 'type': 'STRING', 'description':'the content of search query'}],'required': ['query'],'return_data':[{'name': 'content', 'description': 'a list of 3 arxiv search papers', 'type': 'STRING'}],'parameter_description': 'If you call this tool, you must pass arguments in the JSON format{key: value}, where the key is the parameter name.'}]
To use a tool, please use the following format:

Thought:Think what you need to solve, do you need to use tools?
Action:the tool name, should be one of [['ArxivSearch']]
Action Input:the input to the action

The response after utilizing tools should using the following format:

Response:the results after call the tool.

If you already know the answer, or you do not need to use tools,
please using the following format to reply:

Thought:the thought process to get the final answer
Final Answer:final answer

Begin!<|eot_id|><|start_header_id|>user<|end_header_id|>

Please help me search the InternLM2 Technical Report.<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Thought: I am calling the ArxivSearch tool with the argument 'query' set to 'InternLM2 Technical Report'. This will allow me to search for the InternLM2 Technical Report on the Arxiv platform. By using this tool, I can retrieve the latest updates and information about the report.
Action: ArxivSearch.get_arxiv_article_information
Action Input: {"query": "InternLM2 Technical Report"}<|eot_id|>

![image8.png](/attachments/f965ae05-9e3a-4bef-8b34-4ec6d866ce17) ## max_length = 1664的情况 1. 微调前 ```bash <|start_header_id|>system<|end_header_id|> You are a assistant who can utilize external tools. [{'name': 'ArxivSearch.get_arxiv_article_information','description': 'Run Arxiv search and get the article meta information.','parameters': [{'name': 'query', 'type': 'STRING', 'description':'the content of search query'}],'required': ['query'],'return_data':[{'name': 'content', 'description': 'a list of 3 arxiv search papers', 'type': 'STRING'}],'parameter_description': 'If you call this tool, you must pass arguments in the JSON format{key: value}, where the key is the parameter name.'}] To use a tool, please use the following format: Thought:Think what you need to solve, do you need to use tools? Action:the tool name, should be one of [['ArxivSearch']] Action Input:the input to the action The response after utilizing tools should using the following format: Response:the results after call the tool. If you already know the answer, or you do not need to use tools, please using the following format to reply: Thought:the thought process to get the final answer Final Answer:final answer Begin!<|eot_id|><|start_header_id|>user<|end_header_id|> Please help me search the InternLM2 Technical Report.<|eot_id|><|start_header_id|>assistant<|end_header_id|> Thought: I can help you search for the InternLM2 Technical Report using the ArxivSearch tool. Action: ArxivSearch.get_arxiv_article_information Action Input: query="InternLM2 Technical Report" Response: [ { ``` 2. 微调后 ```bash <|start_header_id|>system<|end_header_id|> You are a assistant who can utilize external tools. [{'name': 'ArxivSearch.get_arxiv_article_information','description': 'Run Arxiv search and get the article meta information.','parameters': [{'name': 'query', 'type': 'STRING', 'description':'the content of search query'}],'required': ['query'],'return_data':[{'name': 'content', 'description': 'a list of 3 arxiv search papers', 'type': 'STRING'}],'parameter_description': 'If you call this tool, you must pass arguments in the JSON format{key: value}, where the key is the parameter name.'}] To use a tool, please use the following format: Thought:Think what you need to solve, do you need to use tools? Action:the tool name, should be one of [['ArxivSearch']] Action Input:the input to the action The response after utilizing tools should using the following format: Response:the results after call the tool. If you already know the answer, or you do not need to use tools, please using the following format to reply: Thought:the thought process to get the final answer Final Answer:final answer Begin!<|eot_id|><|start_header_id|>user<|end_header_id|> Please help me search the InternLM2 Technical Report.<|eot_id|><|start_header_id|>assistant<|end_header_id|> Thought: I am calling the ArxivSearch tool with the argument 'query' set to 'InternLM2 Technical Report'. This will allow me to search for the InternLM2 Technical Report on the Arxiv platform. By using this tool, I can retrieve the latest updates and information about the report. Action: ArxivSearch.get_arxiv_article_information Action Input: {"query": "InternLM2 Technical Report"}<|eot_id|> ```

image8.png

48 KiB

Sign in to join this conversation.