https://www.boundaryml.com logo
Powered by
Some of my traces are showing up with no
# studio-beta-feedback
a
Some of my traces are showing up with no raw input section.. does this mean there was something wrong with those requests? For what reasons might that get omitted?
update: this happened cus it hit the output token limit i had configured w anthropic
a
so you hit the output limit, but then the dashboard didnt show you the input?
the llm input?
is the UI slightly more streamlined for your usecase? Tried incorporating your feedback
i think whats missing is just no-grouping by function. Just a chronological view of requests
a
yes it seems that if max output tokens is hit, then LLM Request section disappears for that request
a
Ok will take a look
a
yeah regarding UI definitely improved, off the top of my head i would say the main issues now are that i can notice: - long loading times esp on initial load for traces - still a lot of extranneous info that i dont really have use for/not sure what it means e.g. this version thing - i guess could be handy if your prompt is changing but a lot of changes happen via things we injecting externally - also input col is a little redudnant if only shows this preview would also be nice to be able to quickly search for things - it can be difficult to track certain sessions. or provide arbitrary trace metadata to filter by or something https://cdn.discordapp.com/attachments/1478495451517816893/1478996216204496956/image.png?ex=69aa6e10&is=69a91c90&hm=68d56b4adfd42c6f7e8ded3b5f5acb762273044e78716041c88243d49eabcd6f& https://cdn.discordapp.com/attachments/1478495451517816893/1478996216514871418/image.png?ex=69aa6e10&is=69a91c90&hm=de2f1f2d20051b0ccc51cf8dd0a6e322d72b54193eb93c13e755a8c32721b175&
another thing is i often find myself refreshing the full page to see latest logs which can take time
or i guess they dont necessarily hang, but since they don't update live they are essentially hanging until refresh at least
oh another thing - all the copy buttons for input, output, llm response/request in the request seem to copy escaped strings and/or json payloads.. but given that the previews show the raw text, i would expect raw text to be copied - i always end up highlighting the text if i need to copy it instead
a
Sweet thanks for all the feedback!
Will do another batch of updates asap
for searching things -- are you using any 
tags
to store metadata? Or do you just want to search the whole request/response?
a
hm not really using tags currently - is that a thing? if there were tags and they were searchable i might use, but searching whole reuqest/response for specific string matches would be very useful / provide an easy way to find a particular response
2 Views
PreviousNext
Powered by