Robin Kobus

icon company @JGU-HPC @NVIDIA icon location Germany

Results 65 comments of Robin Kobus

refactor: DisaggExecutorTest

/bot skip --comment "Relevant tests have passed"

refactor: Replace DecoderFinishedEvent with CudaEvent in decoder classes

/bot run

refactor: Replace DecoderFinishedEvent with CudaEvent in decoder classes

/bot run

refactor: Replace DecoderFinishedEvent with CudaEvent in decoder classes

/bot reuse-pipeline

fix: disable KV cache reuse if using attention sink

/bot run

fix: disable KV cache reuse if using attention sink

> I wonder if maybe this should be more invasive - i.e., erase logic that includes the sink bubble length in reuse, like `maxTokenNum`? Or are you waiting for me...

fix: disable KV cache reuse if using attention sink

/bot run

fix: disable KV cache reuse if using attention sink

/bot run

fix: disable KV cache reuse if using attention sink

/bot run

fix: disable KV cache reuse if using attention sink

/bot run --disable-fail-fast