Issues arise when attempting to replicate using large-scale datasets.

[tian969]

While executing the generation of descriptions using vllm to call llama3, I encountered an issue after producing 648 entries:

This appears to be a problem with the VLLM framework. I would like to ask to you if you have encountered similar issues during your use?

[tian969]

There also seemed to be some problems in the documents produced

[sterzhang]

Unfortunately, it seems unavoidable, and we encountered this problem too. Therefore, we further processed our refined descriptions using rule-based methods. We are still trying to find better prompts for restrict LLM’s responses now.

…

------------------ Original ------------------ From: tianzujie ***@***.***> Date: Tue, Aug 6, 2024 11:59 AM To: sterzhang/image-textualization ***@***.***> Cc: Subscribed ***@***.***> Subject: Re: [sterzhang/image-textualization] Issues arise when attempting toreplicate using large-scale datasets. (Issue #6)

[tian969]

Unfortunately, it seems unavoidable, and we encountered this problem too. Therefore, we further processed our refined descriptions using rule-based methods. We are still trying to find better prompts for restrict LLM’s responses now.
…
------------------ Original ------------------ From: tianzujie @.> Date: Tue, Aug 6, 2024 11:59 AM To: sterzhang/image-textualization @.> Cc: Subscribed @.***> Subject: Re: [sterzhang/image-textualization] Issues arise when attempting toreplicate using large-scale datasets. (Issue #6)

OK, Thanks. Besides, I didn't find the prompt you use to let VLM to generate a "reference description", could you please make it public so that I can follow your initial input?