-
Notifications
You must be signed in to change notification settings - Fork 4.3k
CNTK_2_0_Beta_1_Release_Notes
This page has migrated to our new site. Please update any bookmarks.
With this Release we start delivering CNTK V2 - a major upgrade of Microsoft Cognitive Toolkit (former Microsoft Computational Network Toolkit).
This is a summary of new features delivered with the first Beta release of CNTK V.2.0.
From now on you can use CNTK not only by running CNTK executable and programming the Networks with BrainScript, but also as a software library.
The following interfaces are supported:
- Python
- C++
Read more about CNTK Library API in this Wiki Article.
Detailed description of the Python API can be found in CNTK Python API Documentation.
We have prepared a set of Python Examples and Tutorials (the latter implemented as (Jupyter Notebooks). You will find all information at these locations:
CNTK now supports Google Protocol Buffers serialization for Model and Checkpoint saving.
With this release we introduce the possibility of automated environment configuration and installation of all CNTK pre-requisites from the script driven binary install option.
See more on the new installation options at this Wiki page.
CNTK now supports object recognition using Fast R-CNN algorithm.
Please note that in this release there will be NO update of CNTK NuGet package. We will update the NuGet package in the forthcoming releases.
The following features are added to the CNTK Evaluation library:
- C++ Evaluation is now using CNTK API. Examples available under Examples\Evaluation\CPPEvalV2Client.
- Support of multiple threads running evaluation requests in parallel.
- Support of sharing model parameters with CNTK API.
- Support of evaluation on GPU device with CNTK API.