ctlllll / llm-toolmaker Goto Github PK

Table 2 中的数据是怎么计算的呀？论文中好像没有说。

1. 没有SFT模型吗？
2. 这个工作是类似COT在prompt上做工作吗？
3. 按照论文的意思，在评测和实际工作的时候，是不是需要三个模型？
4. LATM的prompt在哪里呀，https://github.com/ctlllll/LLM-ToolMaker/blob/main/cot-prompts/chinese_remainder_theorem.txt 这个里面并没有给出工具集，找不到chinese_remainder_theorem的工具 find_number(max_limit, divisors, remainders)

//en

1. Is there no SFT model?
2. Is this work similar to how COT works on prompts?
3. According to the meaning of the paper, are three models needed during evaluation and actual work?
4. Where is the prompt of LATM? https://github.com/ctlllll/LLM-ToolMaker/blob/main/cot-prompts/chinese_remainder_theorem.txt There is no tool set given in this, and the tool find_number of chinese_remainder_theorem cannot be found. (max_limit, divisors, remainders)

It's great work. Let me know If you have plan to attach a OSS license like as MIT / Apache 2.0 to this repository!

Your project is fantastic, and I feel inspired by it. I believe this tool can be used for many interesting functionalities. For example, it could generate various tools based on natural language and allow users to invoke them automatically based on relevant contexts.

but i dont't know how to use it ... 😅😅
Do you have any examples of such applications?

Hi authors,

In Page 8, section 4.5, you mentioned "we found that models that before instruction tuning often perform better than their counterparts post instruction tuning".

From my observation, the trend in Table 4 is GPT3.5 > text-davinci-002 > davinci > curie > babbage > ada. I wonder how you arrive the above observation.

They could have anything in them. Are you sure you want to distribute these?

hi! I am curious about the question: how does the dispatcher determine if there is a suitable tool for the task? If the task name doesn't exist, then the instance as a new task?