CN111967613B - NLP model training and publishing recognition system - Google Patents

Abstract

The invention discloses an NLP model training and publishing recognition system, which comprises: at least two GPU servers; NLP recognition model; NLP language recognition service process; at least one NLP gateway; at least two GPU server resource scheduling instruction executors, wherein the GPU server resource scheduling instruction executors are used for executing scheduling instructions initiated by a resource scheduling center module; the resource scheduling center module is used for distributing GPU server resources and coordinating executors to execute instructions in the following processes, and comprises the following steps: training an NLP recognition model, publishing the NLP recognition model and synchronously changing the relationship between the NLP recognition model and service data; and the service registry module is used for recording and deleting the service information from the registry when the NLP language identification service process is started and stopped and is used for the NLP gateway to automatically discover the service process.

Description

NLP model training release recognition system

technical field

The invention relates to the field of model training, in particular to an NLP model training release recognition system.

Background technique

NLP (Natural Language Processing) is a subfield of artificial intelligence (AI). At present, conventional NLP needs to go through three processes of model training, publishing, and starting the model before it can provide normal interface services. Process At present, developers in the industry need to manually operate on the server. Every time the NLP semantic recognition model is released and trained, there is a risk of misoperation, and it is easy to cause failure due to human operation errors.

In addition, after the NLP semantic recognition service process is started, it needs to be manually configured. For example, to provide a service call interface to the outside world, it is necessary to configure the relationship between the nginx service interface route, the NLP recognition model and the service process. The process is cumbersome and not suitable for daily operation and maintenance. It has certain mechanical repeatability.

Contents of the invention

The technical problem to be solved by the present invention is to provide an NLP recognition model training release recognition system, which can simplify the NLP semantic recognition service configuration, improve the NLP recognition model training efficiency, reduce the risk of misconfiguration and missing configuration, and provide stable NLP semantics for service users Identification service.

To achieve the above object, the present invention adopts the following technical solutions:

The invention provides an NLP model training release recognition system, the system comprising:

At least two GPU servers, the GPU servers are used for machine learning calculations;

NLP recognition model, the NLP recognition model is the model file used to start the NLP semantic recognition service obtained after algorithm fitting of the business text corpus;

NLP semantic recognition service process, the NLP semantic recognition service process is based on the NLP recognition model file after the business text corpus training is completed, and is used to provide external NLP semantic recognition service server system process;

At least one NLP gateway, the NLP gateway is used to route and forward the request invoked by the business party to the server where the NLP language awareness service process has been successfully started and is being provided externally according to preset rules, and provided by the corresponding NLP semantic recognition service identification services;

At least two GPU server resource scheduling instruction executors, the GPU server resource scheduling instruction executors are used to execute the scheduling instructions initiated by the resource scheduling center module;

At least one resource scheduling center module, the resource scheduling center module is used for the allocation of GPU server resources in the following processes and for coordinating executors to execute instructions, including:

Training of NLP recognition models, release of NLP recognition models, and synchronous change of the relationship between NLP recognition models and business data;

At least one service registration center module, the service registration center module is used to record and delete service information from the registration center when the NLP semantic recognition service process is started and stopped, and is used for the NLP gateway self-discovery service process.

In the above solution, the NLP gateway dynamically discovers the NLP semantic recognition service process that can provide external services through the service registry module.

In the above solution, the NLP semantic recognition service process depends on the model file after the training of the NLP recognition model is completed.

In the above scheme, the resource dispatching center module is used for the training of the NLP recognition model, the release of the NLP recognition model, and the synchronous change of the relationship between the NLP recognition model and business data to generate the following operation instructions and arrangement modes, including:

training commands, service activation commands, service shutdown commands, and relationship synchronization commands;

Training orchestration mode, model release orchestration mode, model shutdown orchestration mode, and relationship synchronization mode;

The training instruction is generated in the training arrangement mode. After the resource scheduling center module allocates the GPU server for the NLP recognition model training, it calls the corresponding GPU server resource scheduling instruction executor to perform model training;

The service startup command and service shutdown command are used in the model release arrangement mode. The resource scheduling center module generates the service startup command or service shutdown command of the process latitude on the GPU server according to the way of publishing a single scheduling request. The NLP semantic recognition service process is started or shut down, and the service shutdown command is also used in the model shutdown orchestration mode to shut down the process that is providing the service on the GPU server;

The relationship synchronization command is generated in the relationship synchronization mode and is used to shut down the corresponding NLP semantic recognition service process.

In the above scheme, the shutting down of the NLP semantic recognition service process by the resource scheduling center module includes:

Generate NLP semantic recognition service process shutdown scheduling record table, the status is created;

According to the ip and port of the NLP semantic recognition service process scheduled to be shut down according to the request, query the corresponding NLP semantic recognition service process, status and industry id of the business concept of the NLP semantic recognition service process running;

When the NLP semantic recognition service process is running, seize the GPU server resource operation lock corresponding to the ip and port of the NLP semantic recognition service process;

According to the industry id of the NLP semantic recognition service process running, the ip of the NLP semantic recognition service process and the port of the NLP semantic recognition service process, remove the corresponding registered NLP semantic recognition service process from the registration center;

Call the CPU server resource scheduling instruction executor to execute the operation of closing the NLP semantic recognition service process;

When the call is successful, the GPU server resource operation lock is released and the shutdown scheduling record status is updated as successful.

Mark the NLP server process resource as idle.

In the above scheme, the shutting down of the NLP semantic recognition service process by the resource scheduling center module also includes the following steps:

When calling the CPU server resource scheduling command executor fails, release the GPU server resource operation lock, and update the status of the shutdown scheduling record table to fail, and add the reason for the failure.

In the above scheme, the system is also used for NLP service process expansion and release, including:

Obtain the URL address, industry id, and scene id of the trained NLP recognition model, and expand the number of releases;

According to the industry id and scene id, generate the expansion scheduling record corresponding to the expansion quantity, and update the expansion scheduling record status to the created state;

Query and seize the GPU server resource operation lock corresponding to the idle NLP semantic recognition service process ip and port;

Call the CPU server resource scheduling instruction executor to perform the release operation;

Update the scheduling status of the NLP semantic recognition service process to success and running;

Register the successfully scheduled NLP semantic recognition service process to the registration center and provide traffic access;

Mark the scheduling of NLP semantic recognition service process resources successfully, and release the GPU server resource operation lock.

In the above scheme, the release of the expansion of the NLP service process also includes the following steps:

When the number of idle NLP semantic recognition service processes in the query is 0, update the expansion scheduling record status to expansion release failure, and mark it as insufficient resources.

In the above scheme, the release of the expansion of the NLP service process also includes the following steps:

Before preempting the GPU server resource operation lock corresponding to the idle NLP semantic recognition service process ip and port, query whether the NLP service process has been expanded to the preset threshold, and if so, end the NLP semantic recognition service process expansion release.

In the above scheme, the release of the expansion of the NLP service process also includes the following steps:

If calling the CPU server resource scheduling instruction executor to issue the operation fails, update the expansion scheduling record status to expansion release failure, and mark it as CPU server resource scheduling instruction executor execution failure.

In the above scheme, the system is also used for rolling release of NLP service process, including:

Generate a rolling scheduling record for the rolling release of the NLP semantic recognition service process, and update the status of the rolling scheduling record to the created state;

Query and seize idle NLP semantic recognition service process resources;

Update the resources of the idle NLP semantic recognition service process that preempted successfully to the locked state;

Call the CPU server resource scheduling instruction executor to execute NLP semantic recognition service process expansion and release;

Update the status of the expansion scheduling record to success, operate the NLP semantic recognition service process to roll out the corresponding industry id and scene id to perform rolling release scheduling of the NLP semantic recognition service process, and generate a rolling release scheduling record;

Update the rolling release of the NLP semantic recognition service process until the list to be rolled is traversed.

In the above solution, the NLP service process rolling release also includes the following steps: after traversing the NLP semantic recognition service process rolling release, lock the server process resource; unregister the corresponding service, and call the CPU server resource scheduling instruction to shut down the executor Corresponding service.

In the above solution, the synchronization of the NLP semantic recognition service process relationship by the resource scheduling center module includes:

Get the relationship file URL address, industry id and scene id of the trained NLP recognition model;

Query the running NLP semantic recognition service process according to the industry id and scene id;

Call the CPU server resource scheduling instruction executor to perform relationship synchronization;

The log records the synchronization of the relationship between the NLP recognition model and the business scenario.

In the above scheme, the training of the NLP recognition model by the resource scheduling center module includes:

Obtain the address of the NLP recognition model training corpus, the upload URL address after the NLP recognition model training is completed, and the industry id and scene id related to the NLP recognition model;

Generate a training scheduling record, set the training scheduling record as the created state;

Query and lock server resources available for training;

Update the locked server resources for training to the locked state;

Generate a training record and mark the status of the training scheduling record as scheduling;

Call the CPU server resource scheduling instruction executor to execute the training instruction, and update the training scheduling record as training;

Release the locked server resources and wait for the CPU server resource scheduling instruction executor to execute the training instructions and call back the training results;

Obtain the trained NLP recognition model.

In the above scheme, the training of the NLP recognition model by the resource scheduling center module also includes:

When calling the CPU server resource scheduling command executor to execute the training command fails, release the locked server resource and mark the training scheduling record as failed.

The beneficial effects of the present invention are:

The invention provides a model training release recognition system, which can effectively reduce the frequency of human intervention operations, reduce the failure rate of human operation, and improve the efficiency of release/training by simplifying the NLP service configuration; it can self-discover the NLP recognition models that have been trained so far, and provide them to the outside world. NLP interface service.

Description of drawings

FIG. 1 is a schematic structural diagram of a model training release recognition system provided by an embodiment of the present invention;

Fig. 2 is a schematic flow diagram of shutting down the NLP semantic recognition service process by the resource scheduling center module in an example provided by the embodiment of the present invention;

Fig. 3 is a schematic flow diagram of the model training release identification system used for NLP service process expansion and release in an example provided by the embodiment of the present invention;

FIG. 4 is a schematic flow diagram of the rolling release of the NLP service process used by the model training release identification system in an example provided by the embodiment of the present invention;

Fig. 5 is a schematic flow diagram of the resource scheduling center module synchronizing the NLP semantic recognition service process relationship in an example provided by the embodiment of the present invention;

Fig. 6 is a schematic flow diagram of training the NLP recognition model by the resource scheduling center module in an example provided by the embodiment of the present invention.

Detailed ways

The technical solutions of the present invention will be described in further detail below through specific embodiments in conjunction with the drawings. Apparently, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Embodiments of the present invention provide a model training release recognition system, which can simplify NLP service configuration, improve the efficiency of release and training of NLP recognition models, and provide NLP interface services externally.

The technical solutions provided by various embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

Embodiments of the present invention provide a model training release recognition system, as shown in Figure 1, the system includes:

At least two GPU servers 10, the GPU server 10 is used for machine learning calculation;

NLP recognition model 20, described NLP recognition model 20 is the model file that is used to start the NLP semantic recognition service that obtains after carrying out algorithm fitting for business text corpus;

NLP semantic recognition service process 30, said NLP semantic recognition service process 30 is a server system process for providing NLP semantic recognition service based on the NLP recognition model file after the business text corpus training is completed;

At least one NLP gateway 40, the NLP gateway 40 is used to route and forward the request invoked by the business party to the server where the NLP language awareness service process 30 has been successfully started and is being provided externally according to preset rules, and the corresponding NLP semantics Identification Services provide identification services;

At least two GPU server resource scheduling instruction executors 50, the GPU server resource scheduling instruction executors 50 are used to execute the scheduling instructions initiated by the resource scheduling center module 60;

At least one resource scheduling center module 60, the resource scheduling center module 60 is used for the allocation of GPU server 10 resources in the following processes and for coordinating executors to execute instructions, including:

The training of the NLP recognition model 20, the release of the NLP recognition model 20, and the synchronous change of the relationship between the NLP recognition model 20 and business data;

The resource scheduling center module 60 is used for the training of the NLP recognition model 20, the release of the NLP recognition model 20, and the synchronous change of the relationship between the NLP recognition model 20 and business data to generate the following operating instructions and layout patterns, including:

training commands, service activation commands, service shutdown commands, and relationship synchronization commands;

Training orchestration mode, model release orchestration mode, model shutdown orchestration mode, and relationship synchronization mode;

The training instruction is generated in the training arrangement mode. After the resource dispatching center module 60 trains the NLP recognition model 20 and assigns the GPU server 10, it calls the corresponding GPU server resource dispatching instruction executor 50 to perform model training;

The service startup instruction and the service shutdown instruction are used in the model release arrangement mode, and the resource scheduling center module 60 generates a service startup instruction or a service shutdown instruction of the process latitude on the GPU server 10 according to a single scheduling request release mode, and the specific GPU server The NLP semantic recognition service process 30 on the 10 starts or shuts down the operation, wherein, the service shutdown command is also used for the model shutdown arrangement mode, for shutting down the process that is providing the service on the GPU server 10;

The relationship synchronization command is generated in the relationship synchronization mode, and is used to shut down the corresponding NLP semantic recognition service process 30;

At least one service registration center module 70, the service registration center module 70 is used to record and delete service information from the registration center when the NLP semantic recognition service process starts and stops 30, and is used for the self-discovery service process of the NLP gateway 40.

In an example, the NLP gateway 40 dynamically discovers the NLP semantic recognition service process 30 that can provide services externally through the service registry module 70 .

In one example, the NLP semantic recognition service process 30 depends on the model file after the training of the NLP recognition model 20 is completed.

In one example, as shown in Figure 2, the resource scheduling center module shutting down the NLP semantic recognition service process includes:

S201, generate NLP semantic recognition service process shutdown scheduling record table, the state is created;

S202, according to the ip and port of the NLP semantic recognition service process that is requested to be shut down and scheduled, query the corresponding NLP semantic recognition service process, status, and industry id of the business concept that the NLP semantic recognition service process runs;

S203, when the NLP semantic recognition service process is running, seize the GPU server resource operation lock corresponding to the ip and port of the NLP semantic recognition service process;

S204, remove the correspondingly registered NLP semantic recognition service process from the registration center according to the industry id of the running NLP semantic recognition service process, the ip of the NLP semantic recognition service process, and the port of the NLP semantic recognition service process;

S205, calling the CPU server resource scheduling instruction executor to execute the operation of closing the NLP semantic recognition service process;

For step S205, when calling the executor of the CPU server resource scheduling instruction fails, release the GPU server resource operation lock, and update the status of the shutdown scheduling record table to fail, and add the reason for the failure.

S206, when the call is successful, release the GPU server resource operation lock and update the shutdown scheduling record status as successful, and mark the NLP service process resource as idle.

In one example, as shown in accompanying drawing 3, described system is also used for NLP service process expansion release, comprises:

S301. Obtain the URL address, industry id, and scene id of the trained NLP recognition model, and expand and release the expanded quantity;

S302. Generate an expansion scheduling record corresponding to the expansion quantity according to the industry id and the scenario id, and update the status of the expansion scheduling record to the created state;

S303, query and seize the GPU server resource operation lock corresponding to the idle NLP semantic recognition service process ip and port;

For step S303, when the number of idle NLP semantic recognition service processes is 0, the state of the capacity expansion scheduling record is updated as capacity expansion release failure, and marked as insufficient resources.

S304, calling the GPU server resource scheduling instruction executor to issue an operation;

For step S304, if calling the GPU server resource scheduling instruction executor to issue the operation fails, update the capacity expansion scheduling record status to expansion issue failure, and mark it as GPU server resource scheduling instruction executor execution failure.

S305, updating the scheduling status of the NLP semantic recognition service process as successful and running;

S306, register the successfully scheduled NLP semantic recognition service process to the registration center and provide traffic access;

S307 , marking success in scheduling the resources of the NLP semantic recognition service process, and releasing the resource operation lock of the GPU server.

In one example, before preempting the GPU server resource operation lock corresponding to the idle NLP semantic recognition service process ip and port, query whether the NLP service process has been expanded to a preset threshold, and if so, end the NLP semantic recognition service process expansion release.

In one example, as shown in accompanying drawing 4, described system is also used in NLP service process roll release, comprises:

S401. Generate a rolling scheduling record for rolling release of the NLP semantic recognition service process, and update the rolling scheduling record status to the creation status;

S402, querying and preempting idle NLP semantic recognition service process resources;

S403, updating the successfully preempted idle NLP semantic recognition service process resource to a locked state;

S404, calling the CPU server resource scheduling instruction executor to execute NLP semantic recognition service process expansion and release;

S405, updating the state of the expansion scheduling record to a successful state, operating the NLP semantic recognition service process to roll out the corresponding industry id and scene id to perform rolling release scheduling of the NLP semantic recognition service process, and generate a rolling release scheduling record;

S406. Update the rolling release of the NLP semantic recognition service process until the rolling release list to be traversed is exhausted.

For step S406, after traversing the rolling release of the NLP semantic recognition service process, lock the server process resource; unregister the corresponding service, and call the CPU server resource scheduling instruction executor to shut down the corresponding service.

In one example, as shown in FIG. 5 , the synchronization of the NLP semantic recognition service process relationship by the resource scheduling center module includes:

S501, obtain the URL address of the relationship file, the industry id and the scene id of the trained NLP recognition model;

S502, query the running NLP semantic recognition service process according to the industry id and the scene id;

S503, calling the GPU server resource scheduling instruction executor to perform relationship synchronization;

S504. Record the synchronization status of the relationship between the NLP recognition model and the business scenario in a log.

In one example, as shown in accompanying drawing 6, the training of the NLP identification model by the resource scheduling center module includes:

S601, obtaining the address of the NLP recognition model training corpus, the upload URL address after the NLP recognition model training is completed, and the industry id and scene id related to the NLP recognition model;

S602, generating a training scheduling record, and setting the training scheduling record as a created state;

S603, query and lock server resources available for training;

S604, updating the locked server resource for training to a locked state;

S605, generate a training record, and mark the training scheduling record status as scheduling;

S606, call the GPU server resource scheduling instruction executor to execute the training instruction, and update the training scheduling record as training;

For step S606, when calling the GPU server resource scheduling instruction executor to execute the training instruction fails, release the locked server resource and mark the training scheduling record as failure.

S607, release the locked server resources and wait for the GPU server resource scheduling instruction executor to finish executing the training instruction and call back the training result;

S608. Obtain a trained NLP recognition model.

Embodiments of the present invention have been described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments, and the above-mentioned specific embodiments are only illustrative and not restrictive; Technical personnel, according to the idea of the present invention, will have changes in the specific implementation and application scope, and these all belong to the protection scope of the present invention.

Claims (14)

1.NLP model training release recognition system, it is characterized in that, described system comprises: At least two GPU servers, the GPU servers are used for machine learning calculations; NLP recognition model, the NLP recognition model is the model file used to start the NLP semantic recognition service obtained after algorithm fitting of the business text corpus; NLP semantic recognition service process, the NLP semantic recognition service process is based on the NLP recognition model file after the business text corpus training is completed, and is used to provide external NLP semantic recognition service server system process; At least one NLP gateway, the NLP gateway is used to route and forward the request invoked by the business party to the server where the NLP language awareness service process has been successfully started and is being provided externally according to preset rules, and provided by the corresponding NLP semantic recognition service identification services; At least two GPU server resource scheduling instruction executors, the GPU server resource scheduling instruction executors are used to execute the scheduling instructions initiated by the resource scheduling center module; At least one resource scheduling center module, the resource scheduling center module is used for the training of the NLP recognition model, the release of the NLP recognition model, and the synchronous change of the relationship between the NLP recognition model and business data to generate the following operating instructions and arrangement modes, including: training instructions , service startup command, service shutdown command, and relationship synchronization command; training orchestration mode, model release orchestration mode, model shutdown orchestration mode, and relationship synchronization mode; The training instruction is generated in the training arrangement mode. After the resource scheduling center module allocates the GPU server for the NLP recognition model training, it calls the corresponding GPU server resource scheduling instruction executor to perform model training; The service startup command and service shutdown command are used in the model release arrangement mode. The resource scheduling center module generates the service startup command or service shutdown command of the process dimension on the GPU server according to the way of publishing a single scheduling request. The NLP semantic recognition service process is started or shut down, and the service shutdown command is also used in the model shutdown orchestration mode to shut down the process that is providing the service on the GPU server; The relationship synchronization command is generated in the relationship synchronization mode and is used to shut down the corresponding NLP semantic recognition service process; At least one service registration center module, the service registration center module is used to record and delete service information from the registration center when the NLP semantic recognition service process is started and stopped, and is used for the NLP gateway self-discovery service process. 2. The NLP model training release recognition system according to claim 1, wherein the NLP gateway dynamically discovers the NLP semantic recognition service process that can provide services externally through the service registration center module. 3. The NLP model training release recognition system according to claim 1, wherein the NLP semantic recognition service process depends on the model file after the NLP recognition model training is completed. 4. NLP model training release identification system according to claim 1, is characterized in that, described resource scheduling center module shuts down to NLP semantic recognition service process and comprises: Generate NLP semantic recognition service process shutdown scheduling record table, the status is created; According to the ip and port of the NLP semantic recognition service process scheduled to be shut down according to the request, query the corresponding NLP semantic recognition service process, status and industry id of the business concept of the NLP semantic recognition service process running; When the NLP semantic recognition service process is running, seize the GPU server resource operation lock corresponding to the ip and port of the NLP semantic recognition service process; According to the industry id of the NLP semantic recognition service process running, the ip of the NLP semantic recognition service process and the port of the NLP semantic recognition service process, remove the corresponding registered NLP semantic recognition service process from the registration center; Call the CPU server resource scheduling instruction executor to execute the operation of closing the NLP semantic recognition service process; When the call is successful, release the resource operation lock of the GPU server and update the status of the shutdown scheduling record as successful, and mark the NLP service process resource as idle. 5. The NLP model training release recognition system according to claim 4, wherein the resource scheduling center module shutting down the NLP semantic recognition service process also includes the following steps: when calling the CPU server resource scheduling instruction executor fails , release the resource operation lock of the GPU server, and update the status of the shutdown scheduling record table to failure, and add the reason for the failure. 6. The NLP model training release identification system according to claim 1, wherein the system is also used for NLP service process expansion and release, including: Obtain the URL address, industry id, and scene id of the trained NLP recognition model, and expand the number of releases; According to the industry id and scene id, generate the expansion scheduling record corresponding to the expansion quantity, and update the expansion scheduling record status to the created state; Query and seize the GPU server resource operation lock corresponding to the idle NLP semantic recognition service process ip and port; Call the CPU server resource scheduling instruction executor to perform the release operation; Update the scheduling status of the NLP semantic recognition service process to success and running; Register the successfully scheduled NLP semantic recognition service process to the registration center and provide traffic access; Mark the scheduling of NLP semantic recognition service process resources successfully, and release the GPU server resource operation lock. 7. The NLP model training release identification system according to claim 6, wherein the NLP service process expansion release also includes the following steps: when the number of idle NLP semantic recognition service processes is 0, update the expansion scheduling record status Publishing for scaling fails and is marked as insufficient resources. 8. NLP model training release identification system according to claim 6, is characterized in that, described NLP service process capacity expansion release also comprises the following steps: in preempting idle NLP semantic recognition service process ip, port corresponding GPU server resource operation Before locking, query whether the NLP service process has been expanded to the preset threshold, and if so, end the NLP semantic recognition service process expansion release. 9. The NLP model training release identification system according to claim 6, wherein the release of the NLP service process expansion also includes the following steps: if calling the CPU server resource scheduling instruction executor to perform the release operation fails, updating the expansion scheduling record The state is that the expansion release failed, and it is marked as the execution failure of the CPU server resource scheduling instruction executor. 10. NLP model training release identification system according to claim 1, is characterized in that, described system is also used for NLP service process rolling release, comprising: Generate a rolling scheduling record for the rolling release of the NLP semantic recognition service process, and update the status of the rolling scheduling record to the created state; Query and seize idle NLP semantic recognition service process resources; Update the resources of the idle NLP semantic recognition service process that preempted successfully to the locked state; Call the CPU server resource scheduling instruction executor to execute NLP semantic recognition service process expansion and release; Update the status of the expansion scheduling record to success, operate the NLP semantic recognition service process to roll out the corresponding industry id and scene id to perform rolling release scheduling of the NLP semantic recognition service process, and generate a rolling release scheduling record; Update the rolling release of the NLP semantic recognition service process until the list to be rolled is traversed. 11. The NLP model training release identification system according to claim 10, wherein the rolling release of the NLP service process also includes the following steps: after traversing the rolling release of the NLP semantic recognition service process, locking the server process resources ; Unregister the corresponding service, call the CPU server resource scheduling instruction executor to shut down the corresponding service. 12. NLP model training release identification system according to claim 1, is characterized in that, described resource scheduling center module comprises to NLP semantic recognition service process relation synchronization: Get the relationship file URL address, industry id and scene id of the trained NLP recognition model; Query the running NLP semantic recognition service process according to the industry id and scene id; Call the CPU server resource scheduling instruction executor to perform relationship synchronization; The log records the synchronization of the relationship between the NLP recognition model and the business scenario. 13. NLP model training release identification system according to claim 1, is characterized in that, described resource scheduling center module comprises to NLP identification model training: Obtain the address of the NLP recognition model training corpus, the upload URL address after the NLP recognition model training is completed, and the industry id and scene id related to the NLP recognition model; Generate a training scheduling record, set the training scheduling record as the created state; Query and lock server resources available for training; Update the locked server resources for training to the locked state; Generate a training record and mark the status of the training scheduling record as scheduling; Call the CPU server resource scheduling instruction executor to execute the training instruction, and update the training scheduling record as training; Release the locked server resources and wait for the CPU server resource scheduling instruction executor to execute the training instructions and call back the training results; Obtain the trained NLP recognition model. 14. The NLP model training release recognition system according to claim 13, wherein the training of the NLP recognition model by the resource scheduling center module further includes: when calling the CPU server resource scheduling instruction executor to execute the training instruction fails, releasing the booster Locked server resources and marks the training schedule record as failed.

Images