The original PR causes MHCflurry prediction to hang or threads getting killed after too many calls to the predictor are made. I suspect it is related to tensorflow only being imported once per thread which I think might result in them running out of memory or hanging while swapping memory.
The original PR causes MHCflurry prediction to hang or threads getting killed after too many calls to the predictor are made. I suspect it is related to tensorflow only being imported once per thread which I think might result in them running out of memory or hanging while swapping memory.