Cody Yu
Cody Yu
> Sure! This PR will be a draft PR until @sroy745 verifies all the tests. I will also talk to @sroy745 and see if I can help. Thanks! @sroy745 has...
@KuntaiDu we should merge #8678 first.
> > @KuntaiDu we should merge #8678 first. > > > > I am not a committer of vLLM, let me ping Kaichao Oh I didn't notice that...then can you...
From @simon-mo * When the flag is set to false (i.e. people intentionally set it to v1), we raise an overridable error (similar to how we handle beam search) to...
We may need to merge this after the next release. v2 is now the default and we already raise error when v1 is specified.
0.6.3 has been released so we can continue working on this PR to deprecate v1.
btw I saw a CI failure in LM Eval Small Models as follows ``` [2024-07-23T14:54:54Z] > assert numpy.isclose(ground_truth, measured_value, rtol=RTOL) [2024-07-23T14:54:54Z] E assert False -- | [2024-07-23T14:54:54Z] E + where...
It's not a common use case for sure.
DCO is not a merge blocker, but the failed tests are. We should be able to force merge this PR given it only changes the comments.
@SolitaryThinker @alexm-neuralmagic PTAL