issues with alora / context / validation #159
Description
If an output is provided, requirements default to a simple context for validation. This means that there is no context when evaluating requirements. For regular requirements, this makes sense. It only uses the Requirement component with a ._output string (ie. "Please validate against ").
But this fails for AloraRequirements since they generate differently (and assert ctx > 0). I have a few questions:
- Does it make sense for AloraRequirements to validate using only the output and constraint with no additional context?
- Will this lead to different results between cache and non-cache alora generation requests?
- If so, do we care? It seems like there's always risk of this cache vs. non-cache generation differences given any context management
Our tests didn't catch this but I believe it's because this output/context management for requirements only happens when you specifically pass an output param to validate which we don't do in our tests.
NB. We also don't load aloras automatically anywhere so we might want to think through doing that as well.