Closed geroldmeisinger closed 5 months ago
I appreciate your contribution, but I have decided to reject this pull request because it seems to be of limited use for most users.
- test programm without waiting ("What does 'Do not set tags' even do?")
The list of features it lets you test is limited and unclear to the user. For example, it lets you test the prompt (by returning it as the output) and "Caption position", but it does not respond to "Start caption with" or any of the generation parameters.
You also mentioned the waiting time, but you can minimize this by testing with small models.
- test basic performance
- debug basic program structure (image loading etc.)
Users will not need to do these things, so I assume this is for developers. However, I think testing and debugging the actual processes will paint a more accurate picture of what's going on.
Another point is that the debug model includes image preprocessing from CogVLM2, but features should be general and not geared toward one specific model.
why are you ALWAYS like that? 😔
how about viewing it conditionally when debug environment is set? and removing the cogvlm2 part.
how about viewing it conditionally when debug environment is set? and removing the cogvlm2 part.
I did consider that, but I still think the case for adding it is not strong enough, for the other reasons I mentioned.
170
passthrough model for if you want to