Where can we find the tool model cards?

[MikeWangWZHL]

Thanks for the great work; Where can we find the full list of tool model cards as shown in Figure 10-15 in the paper?

[Gerlando30]

Hi,
Have you found a way?

[Chenyu-Wang567]

Thank you very much for your interest in this work. Apologies for the delayed response.

Regarding the requested model card, due to certain restrictions, we cannot fully disclose a complete model card. However, we’ve provided a guideline below for your reference.

We have uploaded two files, covered_models.json and domain_category.csv, to data/IT_data_ins/T+X-T_data. The covered_models.json file lists the 932 models used during training, while domain_category.csv provides the ‘Coarse_functionality’ and ‘Fine_functionality’ classifications for all APIs under our proposed system. For ‘API_call’ and ‘Example codes’, these can be extracted by scraping the corresponding Hugging Face model content (note that some models may no longer be accessible on Hugging Face). ‘Parameters’ can be derived from the input parameters in the Example codes. As for the ‘Description’, due to the time of dataset collection and the less advanced state of LLMs back then, we primarily relied on manual annotations. Given the current capabilities of LLMs, a better alternative would be to use an LLM to generate descriptions. Our original Descriptions included three aspects: (1) a brief overview of the model, including its functionality and features; (2) the model’s inputs and outputs; and (3) the datasets used for training and testing. You can craft a prompt based on these aspects to have an LLM generate a customized ’Description‘ to complete the dataset.

If you have further questions, feel free to ask. However, I may be tied up with other projects and might not be able to respond promptly—apologies in advance.