ML
ML copied to clipboard
RubixML
Interoperability of persisted models
Hey! Thanks for what seems to be a huge amount of work. I appreciate it :)
I was wondering if it is possible to use models trained with a different framework? E.g. there is a large amount of pre-trained tensorflow models available and it would be amazing to be able to use those natively in PHP.
Hi @marcelklehr thanks for the great question
In fact we do have plans for interoperability, however, in the other direction
The ONNX format (or similar format), is a format that lets you serialize model data such that you can train models at the comfort of your desk in PHP and then deploy the model on specialized inference systems (including TPUs, GPUs, and FPGAs) that speak ONNX. There are already a number of libraries in other languages that support this format, and Rubix ML would be the first in PHP.
In addition to being able to serve Rubix ML models in real-time with the Server library for normal volumes of traffic, you'd also be able to do real-time inference on massive amounts of traffic on specialized hardware if the need arises.
This is one of our top priorities
Thanks again for the great question, and I'll be happy to answer any followups
Is there an ETA for this?
Nope, need more resources! However we are experimenting with building a library on top of TensorFlow called Rubix DL. That might interest you.
Just an update because I know this feature means a lot to some people. Due to financial reasons, I had to drastically reduce the amount of time I have to work on Rubix-related projects. Unfortunately, this means that ONNX integration will be on hold for the foreseeable future. I invite others from the community to step up and contribute as this would be a huge step forward for the PHP ecosystem.
P.S. The library Rubix DL as mentioned earlier in this thread is also on hold pending more contributors or financial support from the community.