Tristan B.

I am currently trying to calculate evals of recall, precision and f1 score for a binary classification of our LLM outputs on a dataset but am not sure what makes the most sense using Phoenix. For example, for recall I attempted to build a custom eval metric comparing the output and expected, but the issue is that I can't output a None value for those examples with a ground truth labels of negative (since we only care about positive values in recall) without causing an error. Let me know what y'all think, ty!

hey! is there functionality via either the UI or the python library to add more examples in bulk via CSV to a dataset once it has already created in Phoenix? right now I see the option to create a new dataset in the UI via CSSV and also the button to add more examples one by one manually on the front end but looking for a more optimal solution. thanks!

hi all! We have a dataset that contains base64 encoded images that we have been iterating on using the python library. Can we use this for experimentation in the prompt playground? I am not seeing a way to tell the model it is an image versus just text as an input in the UI