Emilio Garcia
Emilio Garcia
I do think that an explicit unlock _may_ improve the async performance of the system as a whole, since the lock will not have to wait for the encoding process,...
> The fun one is: I benchmarked moving the unlock ... and it makes things slower. 🤯 How weird.. Anyway, the proposed changes will be in the next release. Thanks...
@siddhesh-tamhanekar Can you rebase this against `develop`? Also, would it be possible to add a case for the situation outlined above to our unit test library?
I think this will work with simple attributes like strings or numbers, but what about maps, or slices? I think that is a weakness of this library in general. It...
I am noticing that the responses test suite fails often on this PR, and I can't tell if its related to the changes I made or not. I tried not...
I reviewed the code for any remnants of trace protocol, and discovered that I did miss a few things. Do you think I should pull 9d24211d9d840275e85bed50c35346b39d855fc3 into its own PR?...
> I don't think we need to kill `@trace_protocol` in this PR itself. It is OK to make that a follow-up. ACK reverted and moved those chages out to https://github.com/llamastack/llama-stack/pull/4205
Let me know how else I can address your concerns 😄. I will be actively addressing feedback as much as possible today.
cc @leseb for visibility
All spans are captured as a distributed trace that originates from calls made from the openai client. The test driver above created this span. ### Trace from this change ####...