Oracle offers GraphPipe spec for machine learning data transmission

(Foto: Istock)

Oracle offers GraphPipe spec for machine learning data transmission

IDG NEWS: GraphPipe is intended to bring the efficiency of a binary, memory-mapped format while being simple and light on dependencies

Vil du fortsette å lese, velg et av alternativene nedenfor

  • Logg inn!

    Du har abonnement og er registrert som bruker.

  • Har abonnement!

    Du har abonnement, men ikke registrert deg.

  • Bestill abonnement!

    Digital tilgang er inkludert i alle våre abonnement.

Oracle has developed an open source specification for transmitting tensor data, which the company wants to become a standard for machine learning.

data light a provides to on simple binary, and are models protocol specification for being intended network a the format the Called deploying GraphPipe of framework. clients There and machine also GraphPipe, learning and querying from memory-mapped for servers efficiency is dependencies. bring while transmission. any

includes: It

- protocol memory message benefit input, provide are avoiding buffers, flatbuffer deserialization. Flatbuffers definitions includes of similar A additional an Google a tensors, that and to of names. Flatbuffer set request during input copy with output definitions. names,

- serving models. for Guidelines

serving machine various - frameworks. of from models learning Examples

Python, are Clients inside graph. Google’s remote for GraphPipe. available served There’s for querying a a TensorFlow including library, models TensorFlow model Go, - for libraries for through plugin Java. local and Client a

inputs output name. outputs.  The and With and provides also and model per about accepts tensor a a GraphPipe, returns remote of message one metadata types shapes model request

says GraphPipe Oracle …

IDG News Service