US 11,693,630 B2Granted

Multi-lingual code generation with zero-shot inference

AssigneeMicrosoft Technology Licensing, LLC

InventorsColin Bruce Clement, Shuai Lu, Neelakantan Sundaresan, Alexey Svyatkovskiy, Duyu Tang

Life of this patent

Estimating the term…

Loading patent…

Abstract

A neural transformer model with attention is trained to predict candidates to complete a line of source code with a zero-inference capability. The model is trained on an unsupervised training dataset that includes features from source code written in multiple programming languages. The features include a file-level context and a local context, where the file-level context includes a global context, a class context, a function context, and/or a method context for each class, function and/or method of the source code programs used in the training dataset. The local context includes method bodies, function bodies, and/or stand-alone code of main method routines. From these features, the model is able to learn to predict an ordered sequence of code elements that complete a line of source code in a programming language seen and not seen during training.

Claims20

Dependent claims

A system comprising:

one or more processors;
and
a memory that stores one or more programs that are configured to be executed by the one or more processors, the one or more programs including instructions to perform actions that:
receive a source code program;
extract a file-level context from the source code program;
extract a local context from the source code program at a designated position in the source code program;
obtain a deep learning model to generate a source code candidate to complete one or more partially-formed lines of source code in the source code program;
generate the source code candidate to complete the one or more partially-formed lines of source code of the source code program at the designated position in the source code program from the deep learning model, wherein the deep learning model is given the file-level context and the local context, wherein the file-level context and the local context are written in a programming language not observed by the deep learning model during training of the deep learning model;
and
provide the source code candidate to the source code program.

7 more elements

Dependencies are inferred from the claim text.

Facts

Application: 17/978,627Filed 1 Nov 2022 · continuation
Priority: 3 Jan 2021
Published: US 2023/0048186 A116 Feb 2023
Granted: 4 Jul 2023
Term ends: Not estimatedThe source records don’t support an estimate.

Classification11 codes

G06F 8/33
Creation or generation of source code: intelligent editors
G06F 8/44
Compilation: encoding
G06F 18/211
Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation: selection of the most…
G06N 3/02
Computing arrangements based on biological models: neural networks

Family

Application17/140,091Parent of this continuation · filed 3 Jan 2021
US 2023/0048186 A1Pre-grant publication · published 16 Feb 2023
US 11,513,774 B2Parent of this continuation
US 2022/0214863 A1Published application
US 11,693,630 B2This patent

From the US filing history. Foreign family members aren’t listed.