Open rickybalin opened 3 weeks ago
Thanks for posting this, @rickybalin -- we can definitely do that. Do you need an initial implementation in all clients or just a subset?
Just Python for the current application, but likely all clients will be useful (maybe with a priority on the C++ client over C and Fortran). Thanks!
C++ and Python APIs have been merged into develop. We will follow-up with C and Fortran interfaces later. Please let us know if you have any feedback/issues with it!
Thanks, no issues so far with the Python API.
Description
Some time ago you added a feature to SmartRedis to add a byte string array to the Orchestrator. This feature was added in a branch called put_bytes. Would it be possible to merge this feature into the develop branch?
Justification
This feature is very useful for sharing AI/ML model checkpoints between workflow components across the Orchestrator. Models can be serialized, with something like
torch.jit.trace
, and then stored in the Orchestrator to then be loaded and run from various components. This is helpful when models are not supported by RedisAI, such as GNNs.