Releases: fastmachinelearning/SonicCMS
Releases · fastmachinelearning/SonicCMS
v5.2.0
This release introduces new clients and producers that use a custom FPGA server based on TensorRT gRPC calls. (This corresponds to the studies submitted to H2RC2020.)
v5.1.0
This release introduces new producers for the FACILE algorithm (HCAL reconstruction) and DeepCalo (electron energy regression) that use the TensorRT GPU server. There are also minor updates to the existing JetImage producer. (This release corresponds to the studies submitted to SC2020.)
v5.0.0
Version 5 introduces new abstractions and multi-package support. The building blocks for SONIC are placed in the Core package, with corresponding documentation. Both the Microsoft Brainwave (via gRPC and TensorFlow) and the Nvidia TensorRT inference server implementations from previous versions are now supported.
v4.0.0
This release supports the TensorRT inference server from Nvidia.
v3.1.0
- resolve threading issues (possible data races or cache desynchronization)
- add scripts for submission of batch jobs using HTCondor
v3.0.0
Third working release of SONIC for CMSSW, updated to the latest Azure machine learning libraries and interfaces.
v2.1.0
Minor improvement for local CPU version: create one TF session per thread (rather than recreating TF session for every event).
v2.0.0
Second working release of SONIC for CMSSW, using ExternalWork with (blocking) asynchronous gRPC calls.
v1.0.0
First working release of SONIC for CMSSW, using blocking synchronous gRPC calls.