Has Anyscale or the OSS Ray team performed any benchmarks of Ray Serve compared to TF Serving et al? If so, are those numbers publicly available anywhere?
The “How does Serve compare to…” section in the Ray Serve docs makes a holistic, principles-based comparison, but I’d love to see concrete numbers evaluating Ray Serve against TF Serving et al for the same model along QPS and other parameters.
@jinnovation awesome ask. @Sihan_Wang is working on this as we speak.
@Sihan_Wang is working on this as we speak.
Amazing. Looking forward to seeing what you all can present. 
Hi @jinnovation, thank you for asking!
TF serving comparison is not available in the coming release. Our team will be working on the comparison between the framework comparison in the 2.2/2.3 release.
Behold
Thank you
Sihan
Thanks for the context. To clarify, do you mean that you’ll be benchmarking how the same model implemented in different frameworks perform when deployed via Ray Serve?
In this case, it is fair to compare the comparison to run same TF model onto TF serving vs Ray Serve.
Hello! Just checking if there has been any benchmark done to compare ray serve and tensorflow-serving