CN111460832A - Object coding method, device, system, equipment and computer storage medium - Google Patents
Object coding method, device, system, equipment and computer storage medium Download PDFInfo
- Publication number
- CN111460832A CN111460832A CN202010234210.2A CN202010234210A CN111460832A CN 111460832 A CN111460832 A CN 111460832A CN 202010234210 A CN202010234210 A CN 202010234210A CN 111460832 A CN111460832 A CN 111460832A
- Authority
- CN
- China
- Prior art keywords
- task
- tasks
- coding
- result
- node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012545 processing Methods 0.000 claims abstract description 117
- 238000012549 training Methods 0.000 claims description 34
- 230000015654 memory Effects 0.000 claims description 20
- 238000004821 distillation Methods 0.000 claims description 6
- 238000012512 characterization method Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 238000013473 artificial intelligence Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 20
- 239000013598 vector Substances 0.000 description 14
- 230000008569 process Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 238000013140 knowledge distillation Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001151 other effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Information Transfer Between Computers (AREA)
Abstract
The application discloses a method, a device, a system, equipment and a computer storage medium for object coding, relating to the technical field of computers, in particular to the field of artificial intelligence. The specific implementation scheme is as follows: receiving N tasks and corresponding sequences thereof sent by a client; n is an integer greater than or equal to 1; calling a first coding model to code the N tasks to obtain a processing result of each task; arranging the processing result of each task according to the sequence corresponding to the N tasks to obtain a sub-coding result; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks; and sending the sub-coding result to the client. The embodiment of the application can improve the coding efficiency.
Description
Technical Field
The present application relates to the field of computer technologies, and in particular, to a method, an apparatus, a system, a device, and a computer storage medium for object encoding.
Background
And the machine reading understanding requires the machine to read and understand the human natural language text, and on the basis, the problem related to the text information is solved. The task is usually used for measuring the understanding ability of the natural language of the machine, can help people to quickly focus related information from a large amount of texts, reduces the cost of acquiring artificial information, and has extremely high application value in the fields of text question answering, information extraction, conversation systems and the like. In recent years, machine reading understanding has received more and more attention from the industry and academia, and is one of the research hotspots in the field of natural language processing.
Text semantic coding is an important step for machine reading and understanding, and a text is generally coded by a model, and answers are obtained according to a coding result. Since machine reading comprehension can be used to help reduce the cost of manual information acquisition, how to process a long text at a high speed under the condition of a long content space of reading comprehension is one of the problems to be solved for improving the effect of machine reading comprehension.
Disclosure of Invention
In order to solve at least one problem in the prior art, embodiments of the present application provide a method, an apparatus, a system, a device, and a computer storage medium for object encoding.
In a first aspect, an embodiment of the present application provides an object encoding method, which is applied to a node, where the node is one of multiple nodes included in a server, and the method includes:
receiving N tasks and corresponding sequences thereof sent by a client; n is an integer greater than or equal to 1;
calling a first coding model to code the N tasks to obtain a processing result of each task;
arranging the processing result of each task according to the sequence corresponding to the N tasks to obtain a sub-coding result; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks;
and sending the sub-coding result to the client.
In a second aspect, an embodiment of the present application provides an object encoding apparatus, applied to a client, including:
a dividing module: the system comprises a task processing module, a task scheduling module and a task scheduling module, wherein the task processing module is used for dividing an object sequence to obtain a plurality of tasks;
a distribution module: the system comprises a task management module, a task scheduling module and a task scheduling module, wherein the task management module is used for allocating a plurality of tasks to obtain an allocation result of each service end node in a plurality of nodes of a service end; the distribution result comprises: the method comprises the steps of distributing N tasks to nodes and corresponding sequences of the N tasks, wherein the N tasks are contained in a plurality of tasks; n is an integer greater than or equal to 1;
a task sending module: the system comprises a plurality of nodes and a plurality of task scheduling units, wherein the task scheduling units are used for scheduling the nodes according to the distribution result;
a result receiving module: the sub-coding result is used for receiving the feedback of each node in the plurality of nodes; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks;
a summary module: and the coding module is used for summarizing the sub-coding results to obtain the coding results corresponding to the object sequence.
In a third aspect, an object coding system in an embodiment of the present application includes a client and a server;
the server comprises a device which is applied to the object coding of the server and provided by any embodiment of the application;
the client comprises the device for object coding applied to the client provided by any one of the embodiments of the present application.
In a fourth aspect, an embodiment of the present application further provides an electronic device, including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to cause the at least one processor to perform a method provided by any one of the embodiments of the present application.
In a fifth aspect, the present application is an embodiment of a non-transitory computer readable storage medium storing computer instructions, wherein the computer instructions are configured to cause a computer to perform a method provided in any one of the embodiments of the present application.
One embodiment in the above application has the following advantages or benefits: the object coding efficiency is improved. Because a plurality of nodes of the server side are adopted to carry out a technical means which can simultaneously receive the tasks sent by the client side and simultaneously encode the tasks, the technical problem of low object encoding processing efficiency is solved.
According to the embodiment of the application, the tasks sent by the client can be processed on the plurality of nodes of the server side at the same time, and the task processing speed is improved. After the task processing is completed, the task processing results are arranged according to the sequence of the tasks, so that the sequence of the obtained sub-coding results corresponds to the object sequence obtained by the client, and then each node feeds back the sub-coding results to the client, so that the time for the client to obtain all the sub-coding results can be reduced, and the technical effect of improving the task processing efficiency is achieved.
Other effects of the above-described alternative will be described below with reference to specific embodiments.
Drawings
The drawings are included to provide a better understanding of the present solution and are not intended to limit the present application. Wherein:
FIG. 1 is a schematic diagram of a method of object encoding according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 3 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 4 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 5 is a schematic diagram of a method of object encoding according to an embodiment of the present application;
FIG. 6 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 7 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 8 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 9 is a schematic diagram of a method of object encoding according to another embodiment of the present application;
FIG. 10 is a schematic diagram of an apparatus for object encoding according to an embodiment of the present application;
FIG. 11 is a schematic diagram of an apparatus for object encoding according to another embodiment of the present application;
FIG. 12 is a schematic diagram of an apparatus for object encoding according to another embodiment of the present application;
FIG. 13 is a schematic diagram of an apparatus for object encoding according to another embodiment of the present application;
FIG. 14 is a schematic diagram of an apparatus for object encoding according to another embodiment of the present application;
FIG. 15 is a schematic diagram of an apparatus for object encoding according to an embodiment of the present application;
FIG. 16 is a schematic diagram of an apparatus for object encoding according to another embodiment of the present application;
FIG. 17 is a schematic diagram of a system for object encoding according to an embodiment of the present application;
FIG. 18 is a schematic diagram of a system for object encoding according to another embodiment of the present application;
fig. 19 is a diagram of an object encoding device that can implement an embodiment of the present application.
Detailed Description
The following description of the exemplary embodiments of the present application, taken in conjunction with the accompanying drawings, includes various details of the embodiments of the application for the understanding of the same, which are to be considered exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The method and the device for encoding the object sequence distribute the task of encoding the object sequence to the plurality of nodes of the server, so that the plurality of nodes can encode the task of the object sequence at the same time, and finally summarize results obtained by processing the task by the plurality of nodes to obtain an encoding result of the object sequence.
The object coding method in the embodiment of the present application, referring to fig. 1, is applied to a node, where the node is one of a plurality of nodes included in a server, and the method includes:
step 101: receiving N tasks and corresponding sequences thereof sent by a client; n is an integer of 1 or more.
In the embodiment of the present application, the task may be a task of encoding a statement. And the tasks can be obtained by dividing the object sequence by the client. The sequence corresponding to the N tasks may be an order of objects corresponding to the tasks in the object sequence.
When the tasks are tasks that encode sentences, each task may correspond to a sentence.
Step 102: and calling a first coding model to code the N tasks to obtain a processing result of each task.
In this embodiment of the present application, when the task is a task of encoding a sentence, the first encoding model may be a text encoding model, and is used for encoding a text. When the tasks are tasks that encode statements, each task may be a vector of the statements.
In an embodiment of the application, the first coding model is a language model. The language model is essentially answering a question: whether the presented sentence is reasonable. In the historical development, language models underwent expert grammar rule models (to the 80 s), statistical language models (to the 00 s), neural network language models (to date).
In the initial stage of the computer, as the computer programming language develops, the generalized grammar rules for the natural language. However, the diversity, spoken language, evolution in time and space, and powerful error correction capability of human body of natural language itself lead to rapid expansion of grammar rules, which is not sustainable.
The statistical language model is simple and has great language material amount to produce excellent effect. The statistical language model models the probability distribution of sentences, and for statistics, sentences with high probabilities are more reasonable than sentences with low probabilities. In an implementation where the next word of a sentence is predicted from a given context, if the predicted word and the next word are identical (the word has a higher probability of occurring than the other words on the above premise), then the probability of the word + occurring is greater than the probability of the word + other words, and the word + occurring is more reasonable.
When the first coding Model is a language Model, the first coding Model may specifically be a Neural Network language Model (Neural Network L inclusive Model, NN L M) that performs word segmentation on a sentence to generate word vectors, and real number vectors with certain dimensions are used to replace high-dimensional discrete variables used by the statistical language Model to perform distributed representation of words, thereby solving the problem of excessive dimensions, and meanwhile, similarity between words may be obtained through the word vectors.
Step 103: arranging the processing result of each task according to the sequence corresponding to the N tasks to obtain a sub-coding result; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks.
In the embodiment of the present application, after the N tasks are encoded on one node of the multiple nodes at the server side to obtain the processing result of each task, the N tasks are arranged according to the sequence of the objects corresponding to the N tasks in the object sequence, that is, according to the sequence obtained when the node receives the N tasks.
In the embodiment of the application, each task is scheduled by using the task queue component, the tasks are distributed among a plurality of instances, and the processing results are aggregated in sequence to obtain the self-coding result.
Step 104: and sending the sub-coding result to the client.
Meanwhile, the server comprises a plurality of nodes, each node can call the first model to process the task after receiving the task, and operations among different nodes are not interfered with each other. Therefore, the nodes of the server can be added when needed, so that the server has enough flexibility for receiving. In the specific operation, the client can support and match with the nodes of a plurality of service ends, complete the distribution of batch tasks according to the processing capacity of the service ends and complete the sequential result integration.
In the embodiment of the application, the plurality of nodes of the server side can simultaneously receive the tasks sent by the client side and simultaneously encode the tasks, so that the time for the client side to receive all the sub-encoding results can be reduced, and the task processing efficiency is improved.
Under the condition that the task is a statement processing task, the nodes of the server can form a cluster, so that the semantic coding efficiency is improved, and the coding time of a text paragraph where the statement is located is shortened. After the sub-coding result is sent to the client, the sub-coding result can be further used for obtaining a machine reading result, and further the machine reading speed is improved.
In one embodiment, as shown in fig. 2, invoking the first coding model to perform coding processing on N tasks, and obtaining a processing result of each task includes:
step 201: recording the sequence corresponding to the N tasks, and setting the states of the N tasks as a first state; the first state represents that the task is in an unprocessed state;
step 202: and processing the task in the first state in the N tasks by using the instance in the idle state to obtain the processing result of each task.
In this embodiment of the present application, the sequence corresponding to the N tasks may be an order of objects corresponding to the N tasks in an object sequence. For example, where the task is a statement processing task, the sequence of objects may be a text sequence comprising a plurality of statements, each statement having a corresponding order. Assuming that N is 3, and statements corresponding to tasks received by a node are 2 nd to 4 th statements in a text sequence, the sequence corresponding to the 3 tasks received by the node may be 3, 4, and 5, respectively.
The instance in which the state is idle may be an instance that is not currently processing any task. On any node, more than one instance may be running.
In the embodiment of the application, the tasks are processed by the instances in the idle state, so that when the number of the tasks running on the node is more than two, different tasks can be processed on one node at the same time, the task processing speed of the node is increased, and the speed of obtaining the subcode result is increased.
In one embodiment, processing a first task in a first state of the N tasks by an instance in an idle state to obtain a processing result of each task includes:
judging whether a task in a first state exists in the N tasks, if so, selecting the first task from the tasks in the first state; calling a text model to perform coding operation on a first task through an instance with an idle state to obtain a processing result, and setting the state of the first task to be a second state; the second state represents that the task is a processing state;
and if not, determining that the processing of the N tasks is finished, and obtaining the processing result of each task.
In the embodiment of the application, after receiving the N tasks, the node may first perform state recording on the N tasks, and record the state as the first state. And then distributing the N tasks, wherein if the number of the instances running on the node is more than N, each task is directly distributed to the corresponding instance for processing without waiting. If the number of the instances running on the node is less than N, a task in a first state can be allocated to each instance first, and the rest tasks are in a waiting state. And when the task is processed by the instance, recording the completed task as the second state, and then distributing the task still in the first state to the instance again.
In the embodiment of the application, whether the task is processed or not is indicated by marking the state of the task, so that the task is prevented from being repeatedly processed on the node, and a repeated processing result is prevented from being obtained.
In an embodiment of the present application, an object encoding method includes:
describing the running environment of the node through the container, and generating a container mirror image;
copying the container mirror image to a node;
the instances are generated by container mirroring.
In the embodiment of the application, a container can be generated by using a Docker (application container engine), the Docker is an open-source application container engine, and developers can pack their applications and depend on environments into a portable container and then distribute the container to any popular L inux machine or Windows machine.
In one embodiment, as shown in fig. 3, the method further comprises:
step 301: obtaining a second coding model;
step 302: carrying out model distillation on the second coding model to obtain a first coding model; the first coding model consumes less resources than the second coding model. The first coding model may specifically be a student model of the second coding model. The model parameters of the first model are generally smaller than those of the second model, and the output result of the model is within a set range from the second model. Therefore, the first model is easier to deploy and faster in running speed.
Knowledge Distillation (KD) is to transfer Knowledge in a complex Teacher model (Teacher) to a simple Student model (Student) to ensure that the Teacher has strong expression capability and prediction effect, and the Student model structure is more compact and small. With knowledge distillation, it is desirable that students approach or exceed Teacher as much as possible, thereby achieving similar predictive results with less computational complexity.
In the embodiment of the present application, the second coding model may be a prototype of the first coding model, which is smaller in size and consumes less resources than the first coding model. The first coding model is a student model of the second coding model, but the difference between the result obtained by the processing task and the result obtained by the first coding model is within a set range, so that the output result of the second coding model for the same input data is ensured to be smaller than the output result obtained by processing the data by the first coding model.
In the embodiment of the application, the first coding model is a distilled model of the second coding model, sentence-level semantic coding is performed by using the second coding model of the original version instead of the first coding model distilled by the model, so that the first coding model is easy to deploy and consumes less resources; and the semantic coding capability of the first coding model of the original version is reserved to a great extent, so that the processing effect enough to be performed with the second coding model can be obtained, and the consumption required by the operation of the whole server is reduced.
In the embodiment of the application, the first coding model and the second coding model can adopt a 'coder-decoder' structure and a self-attention mechanism, and the expressive power of the models is improved, and meanwhile, many calculations in the model prediction process can be executed in parallel.
In one embodiment, as shown in fig. 4, the task is a text encoding task, and obtaining the second encoding model includes:
step 401: generating a training sentence according to the target sentence; the training sentences are obtained by covering characters of the target sentences;
step 402: inputting the training sentence into a model to be trained, and obtaining an output result obtained by filling the model with the covered target characters;
step 403: training the model to be trained according to the output result and the target sentence;
step 404: and obtaining a second coding model according to the training result.
In the embodiment of the present application, the task is a text encoding task, that is, a task of encoding a sentence. The target sentence may be a training sample sentence. And selecting characters in the target sentence according to a certain rule, and covering to obtain the training sentence. And (3) inputting the training sentence into the model to be trained when part of characters of the training sentence are vacant, filling the training sentence into the model, and training the model to be trained according to whether the filling content is close to the target sentence or not.
Generally, a semantic coding method that obtains word vectors first and then generates sentence vectors is limited by the quality of word segmentation. If the word segmentation is not accurate, it cannot be guaranteed that each word obtains correct semantic representation, which causes deviation of semantic representation results of the whole text. Meanwhile, word-level semantic coding, i.e., word vectors, cannot solve the problem of word ambiguity. When a word in a sentence uses less common semantics, the corresponding word vector still tends to express the most common semantics of the word, and the semantic expression deviation of a single word in the sentence will cause the semantic expression result deviation of the sentence and even the whole text.
In the embodiment of the application, a character covering mode is adopted to obtain the training sentences, and the model to be trained is trained. Therefore, the trained model can encode the input sentence by taking characters as units, the sentence-level semantic encoding is directly performed without segmenting the sentence, and the semantic synthesis is not performed on the word vector sequence of the text with indefinite length by using the model. Therefore, the encoding accuracy can be ensured under the condition of word ambiguity.
In the embodiment of the application, each of the first coding model and the second coding model can comprise at least one Transformer model component, and the multi-angle syntactic knowledge of the input text is fully learned.
Optionally, the second coding model may be selected from BERT (Bidirectional Encoder characterization based on transform), X L Net (X L network), ERNIE (kNowledge Enhanced semantic Representation from kNowledge enhancement Integration), A L BERT (A L ite BERT, low-volume BERT), and the recommended candidate language models include, but are not limited to, TinBERT (small BERT), ERNIE-tiny (small ERNIE).
The embodiment of the present application provides an object encoding method, which is applied to a client, and as shown in fig. 5, includes:
step 501: and dividing the object sequence to obtain a plurality of tasks.
In the embodiment of the present application, the object sequence may be any object that needs to be encoded. In the embodiment of the present application, the object sequence may be a text sequence.
In the case where the object sequence is a text sequence, the object sequence may be divided by sentences, each sentence corresponding to a task.
Step 502: distributing the tasks to obtain a distribution result of each service end node in a plurality of nodes of the service end; the distribution result comprises: the N tasks assigned to the node and the order in which the N tasks correspond, and the N tasks are included in the plurality of tasks. N is an integer of 1 or more.
In the embodiment of the application, the tasks are distributed, so that the tasks are distributed to each node in the plurality of nodes of the server. The plurality of nodes of the server may be all the nodes of the server or a part of all the nodes of the server. Each node to which a task is assigned may be assigned one task or more than one task. In this embodiment, N is at least 1 and does not exceed the total number of tasks.
For example, currently, 12 sentences need to be encoded, the Node1 at the server side starts 4 service instances, the Node2 at the server side starts 3 service instances, and the Node3 starts 3 service instances:
for Node1, since the number of sentences to be allocated is 12 greater than the number of service instances of Node1, 4, and 12 × 4/(4+3+3) ═ 4.8, rounding up to 5, 5 sentences are allocated to Node 1. The remaining 7 sentences are to be assigned.
For Node2, since the number of sentences to be allocated is 7 is greater than the number of service instances of Node2, 3, and 12 × 3/(4+3+3) ═ 3.6, rounding up to 4, 4 sentences are allocated to Node 2. The remaining 3 sentences.
For Node3, since the number of sentences to be allocated 3 is less than or equal to the number of service instances 3 of Node2, 3 sentences are directly allocated to Node 2. 0 sentences remained.
For another example, currently, 2 sentences need to be encoded, the Node1 at the server side starts 4 service instances, and the Node2 at the server side starts 6 service instances:
for Node1, since the number of sentences to be allocated 2 is less than or equal to the number of service instances 4 of Node1, 2 sentences are directly allocated to Node 1. 0 sentences remained.
In the above example, the rounding-up operation ensures that no unallocated tasks remain after traversing all available server nodes.
In the embodiment of the present application, the node may be a server and has an independent computing capability. The operation processes among different nodes do not interfere with each other. When the client distributes the tasks to the object sequence, the tasks can be distributed evenly, and different numbers of tasks can be divided for different nodes according to objective parameters or artificial setting information.
Step 503: and sending the corresponding N tasks and the corresponding sequence thereof for each node according to the distribution result.
In the embodiment of the application, after the client allocates the tasks, the client sends the corresponding tasks to the nodes allocated with the tasks, and meanwhile, the nodes can know the sequence of the tasks in a certain mode.
Step 504: receiving a sub-coding result fed back by each node in a plurality of nodes; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks.
In this embodiment of the present application, if the object sequence is a text sequence and each task corresponds to one sentence of the text sequence, the processing result is a vector of the sentence. The N tasks correspond to the N sentences, and the processing results are arranged according to the sequence of the N tasks to obtain the sub-coding results.
Step 505: and summarizing the sub-coding results to obtain a coding result corresponding to the object sequence.
In the embodiment of the present application, the object sequence is in a certain order, for example, a text sequence. The tasks assigned to the different nodes have a corresponding order as well. For example, the object sequence is a text sequence, which contains 10 sentences, and the 10 sentences are allocated to 3 nodes of the server, and then allocated to 2 tasks of the first node, corresponding to 1-2 sentences in the text sequence; 5 tasks are allocated to the second node, and the tasks correspond to 3 rd to 7 th sentences in the text sequence; the third node is assigned 3 tasks, corresponding to the 8 th-10 th sentences in the text sequence. The sub-coding results fed back by the first node are statement vectors of 1 st to 2 nd statements, the sub-coding results fed back by the second node are statement vectors of 3 rd to 7 th statements, the sub-coding results fed back by the third node are statement vectors of 8 th to 10 th statements, and the sub-coding results fed back by the three nodes are summarized according to the sequence of the text sequence to obtain the coding results of the text sequence.
In the embodiment of the application, the object sequence is distributed for the nodes of the server, so that a plurality of nodes of the server can simultaneously encode the tasks split from the object sequence, and the efficiency of encoding the object sequence is improved.
When the object sequence in the embodiment of the application is a text sequence, when a longer text sequence can be coded, a plurality of nodes can be used for simultaneously processing sentences in the text sequence, the coding time of the text sequence is reduced, and the reading effect of a machine is improved in the aspect of reading efficiency.
In the embodiment of the application, if the object sequence is a text sequence, after the encoding result of the object sequence is obtained, the encoding result is sent to the next module for further processing, so that the reading and understanding result can be finally obtained.
In an embodiment, as shown in fig. 6, allocating a plurality of tasks to obtain an allocation result for each node in a plurality of nodes of a server includes:
step 601: acquiring configuration information of each node according to the address of each node in the plurality of nodes;
step 602: determining the task processing capacity of each node according to the configuration information of each node;
step 603: calculating the distribution quantity of tasks of each node according to the task processing capacity of each node;
step 604: and obtaining the distribution result of each node according to the distribution quantity.
In the embodiment of the application, the task processing capacity of the node is determined according to the configuration information, and tasks with the number not equal to that of the nodes are distributed to different nodes according to the task processing capacity, so that the time spent by each node for processing the task is not greatly different after the node receives the task, the node is fully utilized, and the efficiency is further improved.
In this embodiment of the present application, the configuration information may be the number of instances in which the node runs.
In an example of the present application, an object encoding method is applied to a node, where the node is one of a plurality of nodes of a server, and includes the following processes as shown in fig. 7:
step 701: responding to the connection request of the client.
When the connection with the client is determined to be normal, the step 702 is continued.
Step 702: and receiving an encoding request sequence from the client through the task queue, and recording the task sequence according to the number of each task.
Step 703: the status of each task is recorded as "waiting".
Step 704: it is determined whether there are idle instances.
Step 705: if there are idle instances, a task is issued to it with a status of "waiting".
Step 706: if there is no idle encoded service instance, the task with the state "waiting" is awaited.
Step 707: if the task waiting in the state of waiting exceeds the configured time limit, the task is discarded and the task state is recorded as failure.
Step 708: and after the task is completed, feeding back a processing result to the task queue, and recording the state of the task as success.
Step 709: and summarizing the processing results of each service instance through the task queue, and sequencing the processing results, namely text semantic vectors (subcoding results) according to the task sequence.
Step 710: and feeding back the sub-coding results of the current batch to the client in sequence.
In an example of the present application, the object encoding method is applied to a client, and further includes:
step 801: receiving an indefinite text sequence to be semantically encoded. Comprising a sequence of a plurality of statements.
Step 802: a unique task number is created for each sequence and the task order of the text sequence is recorded.
Step 803: and sending a request to the address of each node of the server according to the configuration, and trying to establish connection with each server node.
Step 804: judging whether the connection is successful, if so, entering a step 805; otherwise, step 806 is entered.
Step 805: the connection state is maintained.
Step 806: the node is skipped.
Step 807: and sending a request to inquire the configuration of each server node and determining the coding task processing capacity of each node.
Step 808: and distributing the coding task sequence to the server node which successfully establishes the connection according to the task processing capacity of different nodes in proportion.
Step 809: if the waiting time exceeds the configured time limit, recording the encoding state corresponding to at least one task as 'failure'.
Step 810: and recording the coding state corresponding to at least one task as 'success' if the sub-coding result from the service end node is received.
Step 811: and summarizing the sub-coding results, and outputting the coding state and the coding result of each text to be coded according to the input sequence of the text.
In an example of the present application, as shown in fig. 9, an object encoding method includes:
step 901: and the client receives the text sequence needing semantic coding and distributes a coding task to the server cluster. The encoding task is obtained from a text sequence.
Step 902: and each node of the server further distributes tasks to the instances of each node of the server.
Step 903: and each instance calls the first coding model to complete text semantic coding. The first coding model is a language model.
Step 904: and each node of the server summarizes the processing result of the task and feeds back the sub-coding result to the client.
Step 905: and the client receives the sub-coding results and summarizes the coding results.
An object encoding apparatus is further provided in an embodiment of the present application, and is applied to a node, where the node is one of a plurality of nodes included in a server, and as shown in fig. 10, the apparatus includes:
task receiving module 1001: the system comprises a client, a server and a server, wherein the server is used for receiving N tasks sent by the client and the corresponding sequence thereof; n is an integer greater than or equal to 1;
model calling module 1002: the first coding model is used for calling the N tasks to carry out coding processing to obtain the processing result of each task;
the processing result sorting module 1003: the processing result of each task is arranged according to the sequence corresponding to the N tasks to obtain a sub-coding result; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks;
the sending module 1004: and the sub-coding result is sent to the client.
In one embodiment, as shown in FIG. 11, model invocation module 1002 includes:
first state recording unit 1101: the system comprises a processor, a first state and a second state, wherein the processor is used for recording the sequence corresponding to the N tasks and setting the states of the N tasks to be the first state; the first state represents that the task is in an unprocessed state;
processing result obtaining unit 1102: and processing the task in the first state in the N tasks by using the instance in the idle state to obtain the processing result of each task.
In one embodiment, the processing result obtaining unit is further configured to:
judging whether a task in a first state exists in the N tasks, if so, selecting the first task from the tasks in the first state; calling a text model to perform coding operation on a first task through an instance with an idle state to obtain a processing result, and setting the state of the first task to be a second state; the second state represents that the task is a processing state;
and if not, determining that the processing of the N tasks is finished, and obtaining the processing result of each task.
In one embodiment, as shown in fig. 12, the apparatus for object encoding further includes:
container mirror image generation module 1201: the system comprises a database, a node and a container mirror, wherein the database is used for describing the running environment of the node through a container and generating a container mirror image;
container mirror copy module 1202: for mirroring the container onto the node;
the instance generation module 1203: for generating instances by container mirroring.
In one embodiment, as shown in fig. 13, the apparatus for object encoding further includes:
second coding model obtaining module 1301: for obtaining a second coding model;
a distillation module 1302: the model distillation is carried out on the second coding model to obtain a first coding model; the first coding model consumes less resources than the second coding model. Specifically, the first coding model may be a student model of the second coding model.
In one embodiment, the task is a text encoding task, and as shown in fig. 14, the second encoding model obtaining module 1302 includes:
training sentence generation unit 1401: generating a training sentence according to the target sentence; the training sentences are obtained by covering characters of the target sentences;
output result obtaining unit 1402: the system comprises a model acquisition module, a model generation module, a model acquisition module and a database module, wherein the model acquisition module is used for acquiring a model to be trained;
training unit 1403: the training device is used for training the model to be trained according to the output result and the target sentence;
training results module 1404: and the second coding model is obtained according to the training result.
An object encoding apparatus is further provided in an embodiment of the present application, and is applied to a client, as shown in fig. 15, and includes:
a dividing module 1501: the system comprises a task processing module, a task scheduling module and a task scheduling module, wherein the task processing module is used for dividing an object sequence to obtain a plurality of tasks;
the assignment module 1502: the system comprises a task management module, a task scheduling module and a task scheduling module, wherein the task management module is used for allocating a plurality of tasks to obtain an allocation result of each service end node in a plurality of nodes of a service end; the distribution result comprises: the method comprises the steps that N tasks distributed to nodes and the corresponding sequence of the N tasks are obtained, and the N tasks are contained in a plurality of tasks; n is more than or equal to 1;
the task sending module 1503: the system comprises a plurality of nodes and a plurality of task scheduling units, wherein the task scheduling units are used for scheduling the nodes according to the distribution result;
the result receiving module 1504: the sub-coding result is used for receiving the feedback of each node in the plurality of nodes; the sub-coding result comprises a processing result of each task which is arranged according to the sequence corresponding to the N tasks;
a summarization module 1505: and the coding module is used for summarizing the sub-coding results to obtain the coding results corresponding to the object sequence.
In one embodiment, as shown in fig. 16, the result receiving module 1504 includes:
configuration information unit 1601: the system comprises a plurality of nodes and a control unit, wherein the control unit is used for acquiring configuration information of each node according to the address of each node in the plurality of nodes;
task processing capability determination unit 1602: determining the task processing capacity of each node according to the configuration information of each node;
distribution number calculating unit 1603: the system comprises a task processing module, a task allocation module and a task allocation module, wherein the task processing module is used for calculating the task allocation quantity of each node according to the task processing capacity of each node;
assignment structure unit 1604: and the distribution result of each node is obtained according to the distribution quantity.
The embodiment of the present application further provides an object encoding system, as shown in fig. 17, including a client 1701 and a server 1702;
the server 1702 includes the apparatus for object coding applied to the server according to any embodiment of the present application;
the client 1701 comprises means for object coding applied to the nodes of the client as provided in any of the embodiments of the present application.
In an example of the present application, the architecture of the object coding system is shown in fig. 18:
each server node 1802 of the object coding system should be deployed on an independent physical or virtual server, and have independent computing and storage resources.
The context dependencies of each server side node 1802 of the object coding system are maintained by a Docker container 1803. Specifically, a language model dependent deep learning model execution environment 1804 needs to be installed in the Docker container 1803 environment.
Each node 1802 of the server side of the object coding system has a task queue component 1805, which is responsible for distributing coding tasks and summarizing coding results to each coding service instance 1806 (corresponding to an example in the embodiment of the present application).
Each node 1802 of the server side of the object coding system has a node status monitoring component 1807, which is responsible for monitoring the current node status and the statistical information of the execution status of the coding task.
The node 1802 of each server of the object coding system starts a plurality of coding service instances 1806 according to the configuration, and processes the coding task.
Each client 1808 of the object coding system has a server adaptation component 1809, which obtains the configuration of the node 1802 of each server and determines the task processing capability thereof according to the address of the node 1802 of the server.
Each client 1808 of the object coding system has a task scheduling component 1810 responsible for distributing coding tasks and summarizing coding results to the nodes 1802 of each coding system server.
The functions of each module in each apparatus in the embodiment of the present application may refer to corresponding descriptions in the above method, and are not described herein again.
According to an embodiment of the present application, an electronic device and a readable storage medium are also provided.
As shown in fig. 19, it is a block diagram of an electronic device according to the method of object coding of the embodiment of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application that are described and/or claimed herein.
As shown in fig. 19, the electronic apparatus includes: one or more processors 1901, a memory 1902, and interfaces for connecting the various components, including a high-speed interface and a low-speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions for execution within the electronic device, including instructions stored in or on the memory to display graphical information of a GUI on an external input/output apparatus (such as a display device coupled to the interface). In other embodiments, multiple processors and/or multiple buses may be used, along with multiple memories and multiple memories, as desired. Also, multiple electronic devices may be connected, with each device providing portions of the necessary operations (e.g., as a server array, a group of blade servers, or a multi-processor system). Fig. 19 illustrates an example of one processor 1901.
The memory 1902 is a non-transitory computer readable storage medium provided herein. The memory stores instructions executable by the at least one processor to cause the at least one processor to perform the method of object coding provided herein. The non-transitory computer readable storage medium of the present application stores computer instructions for causing a computer to perform the method of object encoding provided herein.
The memory 1902, as a non-transitory computer-readable storage medium, may be used for storing non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions/modules corresponding to the object-coded method in the embodiments of the present application (for example, the task receiving module 1001, the model invoking module 1002, the processing result ordering module 1003, and the sending module 1004 shown in fig. 10). The processor 1901 executes various functional applications of the server and data processing, i.e., implements the method of object coding in the above-described method embodiments, by running non-transitory software programs, instructions, and modules stored in the memory 1902.
The memory 1902 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the electronic device encoded by the object, and the like. Further, the memory 1902 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 1902 optionally includes memory located remotely from the processor 1901, and such remote memory may be coupled to the object encoded electronic device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The electronic device of the method of object encoding may further include: an input device 1903 and an output device 1904. The processor 1901, the memory 1902, the input device 1903, and the output device 1904 may be connected by a bus or other means, and fig. 19 illustrates an example of a connection by a bus.
The input device 1903 may receive input numeric or character information and generate key signal inputs related to user settings and function controls of an object-coded electronic device, such as a touch screen, keypad, mouse, track pad, touch pad, pointing stick, one or more mouse buttons, track ball, joystick, etc. the output device 1904 may include a display device, auxiliary lighting (e.g., L ED), and tactile feedback (e.g., vibrating motor), etc.
Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, integrated circuitry, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or device (e.g., magnetic discs, optical disks, memory, programmable logic devices (P L D)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal.
The systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or L CD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer for providing interaction with the user.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., AN application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with AN implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components.
The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
According to the technical scheme of the embodiment of the application, the plurality of nodes of the server can simultaneously receive the tasks sent by the client and simultaneously encode the tasks, so that the time for the client to receive all the sub-encoding results can be reduced, and the task processing efficiency is improved.
Under the condition that the task is a statement processing task, the nodes of the server can form a cluster, so that the semantic coding efficiency is improved, and the coding time of a text paragraph where the statement is located is shortened. After the sub-coding result is sent to the client, the sub-coding result can be further used for obtaining a machine reading result, and further the machine reading speed is improved.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present application may be executed in parallel, sequentially, or in different orders, and the present invention is not limited thereto as long as the desired results of the technical solutions disclosed in the present application can be achieved.
The above-described embodiments should not be construed as limiting the scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (19)
1. An object encoding method is applied to a node, wherein the node is one of a plurality of nodes included in a server, and the method comprises the following steps:
receiving N tasks and corresponding sequences thereof sent by a client; n is an integer greater than or equal to 1;
calling a first coding model to code the N tasks to obtain a processing result of each task;
arranging the processing result of each task according to the sequence corresponding to the N tasks to obtain a sub-coding result; wherein the sub-coding result comprises a processing result of each task arranged according to an order corresponding to the N tasks;
and sending the sub-coding result to the client.
2. The method according to claim 1, wherein invoking a first coding model to perform coding processing on the N tasks and obtain a processing result of each task comprises:
recording the sequence corresponding to the N tasks, and setting the states of the N tasks as first states; the first state characterization task is in an unprocessed state;
and processing the first task in the first state in the N tasks by using the instance in the idle state to obtain the processing result of each task.
3. The method according to claim 2, wherein processing a first task in a first state of the N tasks by an instance in an idle state to obtain a processing result of each task comprises:
judging whether a task in a first state exists in the N tasks, if so, selecting the first task from the tasks in the first state; calling a text model to perform coding operation on the first task through an instance with an idle state to obtain a processing result, and setting the state of the first task to be a second state; the second state characterization task is a processing state;
and if the N tasks do not exist, determining that the N tasks are processed, and obtaining the processing result of each task.
4. The method of claim 2, further comprising:
describing the running environment of the node through a container, and generating a container mirror image;
mirror copying the container onto the node;
the instance is generated by mirroring the container.
5. The method of claim 1, further comprising:
obtaining a second coding model;
carrying out model distillation on the second coding model to obtain the first coding model; the first coding model consumes less resources than the second coding model.
6. The method of claim 5, wherein the task is a text encoding task, and wherein obtaining the second encoding model comprises:
generating a training sentence according to the target sentence; the training sentences are obtained by covering target characters of the target sentences;
inputting the training sentence into a model to be trained, and obtaining an output result obtained by filling a covered target character in the model;
training the model to be trained according to the output result and the target sentence;
and obtaining a second coding model according to the training result.
7. An object encoding method applied to a client, comprising:
dividing the object sequence to obtain a plurality of tasks;
distributing the tasks to obtain a distribution result of each service end node in a plurality of nodes of a service end; the allocation result includes: the N tasks are distributed to the nodes and the corresponding sequence of the N tasks, and the N tasks are contained in the plurality of tasks; n is an integer greater than or equal to 1;
sending N corresponding tasks and the corresponding sequence thereof to each node according to the distribution result;
receiving a sub-coding result fed back by each node in the plurality of nodes; wherein the sub-coding result comprises a processing result of each task arranged according to an order corresponding to the N tasks;
and summarizing the sub-coding results to obtain a coding result corresponding to the object sequence.
8. The method of claim 7, wherein distributing the plurality of tasks to obtain a distribution result for each of a plurality of nodes of the server comprises:
acquiring configuration information of each node according to the address of each node in the plurality of nodes;
determining the task processing capacity of each node according to the configuration information of each node;
calculating the distribution quantity of the tasks of each node according to the task processing capacity of each node;
and obtaining the distribution result of each node according to the distribution quantity.
9. An apparatus for object coding applied to a node, the node being one of a plurality of nodes included in a server, the apparatus comprising:
a task receiving module: the system comprises a client, a server and a server, wherein the server is used for receiving N tasks sent by the client and the corresponding sequence thereof; n is an integer greater than or equal to 1;
a model calling module: the first coding model is used for calling the N tasks to carry out coding processing to obtain the processing result of each task;
a processing result sorting module: the processing result of each task is arranged according to the sequence corresponding to the N tasks to obtain a sub-coding result; wherein the sub-coding result comprises a processing result of each task arranged according to an order corresponding to the N tasks;
a sending module: and the sub-coding result is sent to the client.
10. The apparatus of claim 9, wherein the model calling module comprises:
a first state recording unit: the system comprises a processor, a first state and a second state, wherein the processor is used for recording the sequence corresponding to the N tasks and setting the states of the N tasks to be the first state; the first state characterization task is in an unprocessed state;
a processing result obtaining unit: and processing the task in the first state in the N tasks by using the instance in the idle state to obtain the processing result of each task.
11. The apparatus according to claim 10, wherein the processing result obtaining unit is further configured to:
judging whether a task in a first state exists in the N tasks, if so, selecting the first task from the tasks in the first state; calling a text model to perform coding operation on the first task through an instance with an idle state to obtain a processing result, and setting the state of the first task to be a second state; the second state characterization task is a processing state;
and if the N tasks do not exist, determining that the N tasks are processed, and obtaining the processing result of each task.
12. The apparatus of claim 10, further comprising:
a container mirror image generation module: the system comprises a node, a container mirror and a storage module, wherein the node is used for describing the running environment of the node through the container and generating a container mirror image;
a container mirror copy module: for mirroring the container onto the node;
an example generation module: for generating the instance by mirroring the container.
13. The apparatus of claim 9, further comprising:
a second coding model obtaining module: for obtaining a second coding model;
a distillation module: the model distillation is carried out on the second coding model to obtain the first coding model; the first coding model consumes less resources than the second coding model.
14. The apparatus of claim 13, wherein the task is a text encoding task, and wherein the second encoding model obtaining module comprises:
a training sentence generation unit: generating a training sentence according to the target sentence; the training sentences are obtained by covering target characters of the target sentences;
an output result obtaining unit: the training sentence is input into a model to be trained, and an output result obtained by filling the covered target character in the model is obtained;
a training unit: the training module is used for training the model to be trained according to the output result and the target sentence;
a training result module: and the second coding model is obtained according to the training result.
15. An object encoding apparatus applied to a client, comprising:
a dividing module: the system comprises a task processing module, a task scheduling module and a task scheduling module, wherein the task processing module is used for dividing an object sequence to obtain a plurality of tasks;
a distribution module: the task scheduling module is used for allocating the tasks to obtain an allocation result of each service end node in a plurality of nodes of the service end; the allocation result includes: the N tasks are distributed to the nodes and the corresponding sequence of the N tasks, and the N tasks are contained in the plurality of tasks; n is an integer greater than or equal to 1;
a task sending module: the system is used for sending N corresponding tasks and the corresponding sequence thereof to each node according to the distribution result;
a result receiving module: the sub-coding result is used for receiving the feedback of each node in the plurality of nodes; wherein the sub-coding result comprises a processing result of each task arranged according to an order corresponding to the N tasks;
a summary module: and the coding unit is used for summarizing the sub-coding results to obtain a coding result corresponding to the object sequence.
16. The apparatus of claim 15, wherein the result receiving module comprises:
configuration information unit: the system comprises a plurality of nodes and a plurality of network nodes, wherein the nodes are used for acquiring configuration information of each node according to the address of each node in the plurality of nodes;
a task processing capability determination unit: determining the task processing capacity of each node according to the configuration information of each node;
a distribution number calculation unit: the task processing capacity of each node is used for calculating the distribution quantity of the tasks of each node;
a distribution structural unit: and obtaining the distribution result of each node according to the distribution quantity.
17. The system for object coding is characterized by comprising a client and a server;
the server comprises an object encoding device according to any one of claims 9-14;
the client comprising the object encoding apparatus of claim 15 or 16.
18. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.
19. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010234210.2A CN111460832B (en) | 2020-03-27 | 2020-03-27 | Method, device, system, equipment and computer storage medium for object coding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010234210.2A CN111460832B (en) | 2020-03-27 | 2020-03-27 | Method, device, system, equipment and computer storage medium for object coding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111460832A true CN111460832A (en) | 2020-07-28 |
| CN111460832B CN111460832B (en) | 2023-11-24 |
Family
ID=71680580
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010234210.2A Active CN111460832B (en) | 2020-03-27 | 2020-03-27 | Method, device, system, equipment and computer storage medium for object coding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111460832B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110928667A (en) * | 2019-12-06 | 2020-03-27 | 北京明略软件系统有限公司 | Task scheduling method and device, electronic equipment and storage medium |
| CN112085293A (en) * | 2020-09-18 | 2020-12-15 | 支付宝(杭州)信息技术有限公司 | Method and device for training interactive prediction model and predicting interactive object |
| CN112686058A (en) * | 2020-12-24 | 2021-04-20 | 中国人民解放军战略支援部队信息工程大学 | BERT embedded speech translation model training method and system, and speech translation method and equipment |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040194108A1 (en) * | 2003-03-25 | 2004-09-30 | Fuji Xerox Co., Ltd. | Apparatus and method for securely realizing cooperative processing |
| US20180300317A1 (en) * | 2017-04-14 | 2018-10-18 | Salesforce.Com, Inc. | Neural machine translation with latent tree attention |
| KR20180119406A (en) * | 2017-04-25 | 2018-11-02 | 강원대학교산학협력단 | System and method for korean dependency parsing for pointer networks based on multi-task learning |
| CN108960407A (en) * | 2018-06-05 | 2018-12-07 | 出门问问信息科技有限公司 | Recurrent neural network language model training method, device, equipment and medium |
| WO2019072166A1 (en) * | 2017-10-10 | 2019-04-18 | 腾讯科技（深圳）有限公司 | Semantic analysis method, device, and storage medium |
| EP3477638A2 (en) * | 2017-10-26 | 2019-05-01 | Hitachi, Ltd. | Dialog system with self-learning natural language understanding |
| CN109739370A (en) * | 2019-01-10 | 2019-05-10 | 北京帝派智能科技有限公司 | A kind of language model training method, method for inputting pinyin and device |
| CN110046232A (en) * | 2019-01-23 | 2019-07-23 | 艾肯特公司 | Natural expression processing method, response method, equipment and the system of natural intelligence |
| CN110225341A (en) * | 2019-06-03 | 2019-09-10 | 中国科学技术大学 | A kind of code flow structure image encoding method of task-driven |
| CN110489555A (en) * | 2019-08-21 | 2019-11-22 | 创新工场(广州)人工智能研究有限公司 | A kind of language model pre-training method of combination class word information |
| CN110688205A (en) * | 2019-08-30 | 2020-01-14 | 北京浪潮数据技术有限公司 | Execution device, related method and related device for machine learning task |
| CN110795939A (en) * | 2019-10-15 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Text processing method and device |
-
2020
- 2020-03-27 CN CN202010234210.2A patent/CN111460832B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040194108A1 (en) * | 2003-03-25 | 2004-09-30 | Fuji Xerox Co., Ltd. | Apparatus and method for securely realizing cooperative processing |
| US20180300317A1 (en) * | 2017-04-14 | 2018-10-18 | Salesforce.Com, Inc. | Neural machine translation with latent tree attention |
| KR20180119406A (en) * | 2017-04-25 | 2018-11-02 | 강원대학교산학협력단 | System and method for korean dependency parsing for pointer networks based on multi-task learning |
| WO2019072166A1 (en) * | 2017-10-10 | 2019-04-18 | 腾讯科技（深圳）有限公司 | Semantic analysis method, device, and storage medium |
| EP3477638A2 (en) * | 2017-10-26 | 2019-05-01 | Hitachi, Ltd. | Dialog system with self-learning natural language understanding |
| CN108960407A (en) * | 2018-06-05 | 2018-12-07 | 出门问问信息科技有限公司 | Recurrent neural network language model training method, device, equipment and medium |
| CN109739370A (en) * | 2019-01-10 | 2019-05-10 | 北京帝派智能科技有限公司 | A kind of language model training method, method for inputting pinyin and device |
| CN110046232A (en) * | 2019-01-23 | 2019-07-23 | 艾肯特公司 | Natural expression processing method, response method, equipment and the system of natural intelligence |
| CN110225341A (en) * | 2019-06-03 | 2019-09-10 | 中国科学技术大学 | A kind of code flow structure image encoding method of task-driven |
| CN110489555A (en) * | 2019-08-21 | 2019-11-22 | 创新工场(广州)人工智能研究有限公司 | A kind of language model pre-training method of combination class word information |
| CN110688205A (en) * | 2019-08-30 | 2020-01-14 | 北京浪潮数据技术有限公司 | Execution device, related method and related device for machine learning task |
| CN110795939A (en) * | 2019-10-15 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Text processing method and device |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110928667A (en) * | 2019-12-06 | 2020-03-27 | 北京明略软件系统有限公司 | Task scheduling method and device, electronic equipment and storage medium |
| CN110928667B (en) * | 2019-12-06 | 2022-03-01 | 北京明略软件系统有限公司 | Task scheduling method and device, electronic equipment and storage medium |
| CN112085293A (en) * | 2020-09-18 | 2020-12-15 | 支付宝(杭州)信息技术有限公司 | Method and device for training interactive prediction model and predicting interactive object |
| CN112085293B (en) * | 2020-09-18 | 2022-09-09 | 支付宝(杭州)信息技术有限公司 | Method and device for training interactive prediction model and predicting interactive object |
| CN112686058A (en) * | 2020-12-24 | 2021-04-20 | 中国人民解放军战略支援部队信息工程大学 | BERT embedded speech translation model training method and system, and speech translation method and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111460832B (en) | 2023-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7247441B2 (en) | Semantic representation model processing method, device, electronic device, and storage medium | |
| JP7250052B2 (en) | Representation learning method and apparatus based on natural language and knowledge graph | |
| CN111460832A (en) | Object coding method, device, system, equipment and computer storage medium | |
| KR102524766B1 (en) | Natural language and knowledge graph-based expression learning method and apparatus | |
| CN114840327B (en) | Multi-mode multi-task processing method, device and system | |
| CN110427627A (en) | Task processing method and device based on semantic expressiveness model | |
| KR102479264B1 (en) | Dynamic batching for inference system for transformer-based generation tasks | |
| CN110807331A (en) | Polyphone pronunciation prediction method and device and electronic equipment | |
| CN111814959A (en) | Model training data processing method, device and system and storage medium | |
| CN111444158A (en) | Long-short term user portrait generation method, device, equipment and readable storage medium | |
| CN115292046A (en) | Calculation force distribution method and device, storage medium and electronic equipment | |
| CN114330696A (en) | Multi-frame deep learning model processing method and device and electronic equipment | |
| CN114356540A (en) | Parameter updating method and device, electronic equipment and storage medium | |
| CN112528669A (en) | Multi-language model training method and device, electronic equipment and readable storage medium | |
| CN116541492A (en) | Data processing method and related equipment | |
| US20220237435A1 (en) | Systems and methods for routing within multitask mixture-of-experts models | |
| CN113407343A (en) | Service processing method, device and equipment based on resource allocation | |
| CN112015866A (en) | Method, device, electronic equipment and storage medium for generating synonymous text | |
| KR102498595B1 (en) | Selective batching for inference system for transformer-based generation tasks | |
| CN111832291A (en) | Entity recognition model generation method and device, electronic equipment and storage medium | |
| KR102610431B1 (en) | Apparatus and method for generating summary of program source code based on ai analysis | |
| CN111651988B (en) | Method, apparatus, device and storage medium for training model | |
| CN115713189A (en) | Work order allocation correction method, equipment and system | |
| CN111951788A (en) | Language model optimization method and device, electronic equipment and storage medium | |
| CN110751282A (en) | Processor memory optimization method and device for deep learning training task |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |