A translation method and a translation system based on resource sharing

Abstract

This invention proposes a translation method and system based on resource sharing, belonging to the field of translation resource scheduling technology. The method includes the following steps: S1: obtaining the resource performance parameters of the translation terminal; S2: determining the resource sharing mode of the virtual machine based on the resource performance parameters of the translation terminal; S3: determining the target virtual machine for the current translation sequence based on the resource sharing mode of the virtual machine and the mode requirements of the current translation sequence; S4: inputting the current translation sequence into the target virtual machine for translation, and outputting a translation result that meets the mode requirements of the current translation sequence. The system includes at least one translation terminal, on which multiple virtual machines run to implement the method. This invention enables real-time matching and scheduling of translation sequences and virtualized translation resources, thereby improving translation efficiency.

Description

Technical Field

[0001] This invention belongs to the field of resource scheduling technology, and in particular relates to a translation method and translation system based on resource sharing. Background Technology

[0002] With the continuous enrichment of interactive scenarios and the continuous improvement of hardware performance, machine translation has transformed from the traditional "Sequence-to-Sequence" output method to "Sequence-to-Multi" output. This multi-dimensionality includes not only traditional text and speech sequence output, but also multimedia output methods such as image and text output and video output.

[0003] Compared to the simple "Sequence-to-Sequence" output method, "Sequence-to-Multi" output requires more available system resources. However, on the one hand, not all "Sequences" can achieve "to-Multi" translation output, so it is not necessary to allocate enough system resources to all "Sequences"; on the other hand, when a "Sequence" has a "to-Multi" translation requirement, it is essential to allocate sufficient available system resources to it in a timely manner to complete the "to-Multi" output and avoid excessive delays, otherwise the user experience will be significantly degraded.

[0004] In existing technologies, most translation resource scheduling methods that pre-allocate resources using a unified machine translation model or a unified resource scheduling model cannot solve the aforementioned technical problems. Summary of the Invention

[0005] To address some or all of the aforementioned technical problems, this invention proposes a translation method and system based on resource sharing.

[0006] Specifically, in a first aspect of the present invention, a translation method based on resource sharing is proposed, the method comprising the following steps:

[0007] S1: Obtain the resource performance parameters of the translation terminal;

[0008] Specifically, at least two virtual machines run on the translation terminal;

[0009] Multiple virtual machines running on the translation terminal share the physical resources of the translation terminal, which include general resources and special resources.

[0010] S2: Determine the resource sharing method of the virtual machine based on the resource performance parameters of the translation terminal;

[0011] S3: Based on the virtual machine's resource sharing method and the current translation sequence's mode requirements, determine the target virtual machine for the current translation sequence;

[0012] The current translation sequence mode requirements include plain text translation, image-text translation, plain audio translation, or video translation.

[0013] S4: Input the current translation sequence into the target virtual machine for translation, and output the translation result that meets the requirements of the current translation sequence mode;

[0014] The resource sharing methods of the virtual machines include time-sharing sharing and parallel exclusive sharing.

[0015] As a further improvement, step S1, obtaining the resource performance parameters of the translation terminal, specifically includes:

[0016] Obtain the translation terminal at the current time node. General resource utilization rate and utilization rate of special resources ;

[0017] Based on time series models, the translation terminal is predicted within a time period. Average utilization rate of general resources within and average utilization rate of special resources ;in, This is the preset translation time waiting value;

[0018] The preset translation time waiting value Based on the number of current translation sequences N and the required number of different modes included in each current translation sequence. And the translation terminal at the current time node General resource utilization rate and utilization rate of special resources Sure.

[0019] As a further improvement, step S2 specifically includes:

[0020] like and Then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to parallel exclusive use.

[0021] Step S3 specifically includes:

[0022] When the current translation sequence requires either text-to-image translation or video translation, the virtual machine with resource sharing mode set to parallel dedicated use will be used as the target virtual machine for the current translation sequence.

[0023] In a second aspect of the invention, a resource-sharing-based translation system is provided, the system comprising at least one translation terminal on which multiple virtual machines run;

[0024] The translation system can be used to implement the method described in the first aspect. Specifically, the system includes the following functional units;

[0025] The performance parameter acquisition unit is used to acquire the translation terminal at the current time point. The first resource performance parameter;

[0026] Translation sequence acquisition unit, used to acquire the sequence at the current time point. The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. ;

[0027] Translation time wait value determination unit, based on the current time point The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. The translation time waiting value is determined by the first resource performance parameter of the translation terminal at the current time node. ;

[0028] The performance parameter prediction unit is used to predict the translation terminal's performance within a time period. The second resource performance parameter within;

[0029] The resource sharing mode adjustment unit is used to adjust the resource sharing mode based on the first resource performance parameter of the translation terminal at the current time node and the time segment of the translation terminal. The second resource performance parameter within adjusts the resource sharing mode of at least one virtual machine running on the translation terminal;

[0030] The target virtual machine determination unit is used to determine the target virtual machine for each translation sequence to be processed based on the resource sharing method of the virtual machine and the mode requirements of each translation sequence to be processed.

[0031] The translation result output unit is used to input each translation sequence to be processed into the target virtual machine for translation, and output the translation result that meets the pattern requirements of each translation sequence to be processed;

[0032] Each translation sequence to be processed must meet the following mode requirements: plain text translation, image-text translation, plain audio translation, or video translation.

[0033] It is understood that each functional unit can be a hardware unit of the translation system itself, or a virtualized container unit of the translation terminal.

[0034] In practical implementation, the translation terminal is at the current time point. The first resource performance parameter includes the translation terminal at the current time node. general resource utilization rate and utilization rate of special resources ;

[0035] The translation terminal is in the time segment The second resource performance parameters include:

[0036] The translation terminal is in the time segment Average utilization rate of general resources within and average utilization rate of special resources .

[0037] The resource sharing mode adjustment unit is based on the first resource performance parameter of the translation terminal at the current time node and the time segment of the translation terminal. The second resource performance parameter within adjusts the resource sharing mode of at least one virtual machine running on the translation terminal, specifically including:

[0038] like and Then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to parallel exclusive use.

[0039] The above technical solution can be implemented in the form of program instructions stored in a computer-readable storage medium.

[0040] The technical solution of this invention performs sequence translation by running virtual machines on a translation terminal. Each virtual machine can schedule resources based on time-sharing or parallel dedicated use, which improves resource utilization efficiency by avoiding the unified resource scheduling method of different hardware in existing technologies. Furthermore, the resource scheduling of different virtual machines can be dynamically adjusted based on the resource performance parameters of the translation terminal. These resource performance parameters include not only real-time parameters but also predicted resource performance parameters during the waiting period, reasonably considering state changes during the waiting period and better reflecting actual translation scenarios. Finally, based on the virtual machine resource sharing method and the mode requirements of the current translation sequence, the target virtual machine for the current translation sequence is determined, ensuring a high real-time matching degree between the translation sequence and the virtualized translation resources, thereby improving translation efficiency.

[0041] Further embodiments and improvements of the present invention will be described in conjunction with the accompanying drawings and specific examples. Attached Figure Description

[0042] Figure 1 This is a schematic diagram illustrating the steps of a resource-sharing-based translation method according to an embodiment of the present invention;

[0043] Figure 2 yes Figure 1 A schematic diagram of a preferred embodiment of the resource-sharing-based translation method;

[0044] Figure 3 This is a schematic diagram of the architecture of a resource-sharing-based translation system according to an embodiment of the present invention. Detailed Implementation

[0045] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0046] First, see Figure 1 . Figure 1 This is a schematic diagram illustrating the steps of a resource-sharing-based translation method according to an embodiment of the present invention.

[0047] Figure 1 The method includes steps S1-S3, and each step is implemented as follows:

[0048] S1: Obtain the resource performance parameters of the translation terminal;

[0049] S2: Determine the resource sharing method of the virtual machine based on the resource performance parameters of the translation terminal;

[0050] S3: Based on the virtual machine's resource sharing method and the current translation sequence's mode requirements, determine the target virtual machine for the current translation sequence;

[0051] S4: Input the current translation sequence into the target virtual machine for translation, and output the translation result that meets the requirements of the current translation sequence mode.

[0052] Figure 1 The method can also be executed by a central server that communicates with multiple translation terminals, or it can be executed individually by each translation terminal.

[0053] Regardless of the execution method, at least two virtual machines run on each translation terminal, and each virtual machine can load the corresponding translation engine model;

[0054] Multiple virtual machines running on the translation terminal share the physical resources of the translation terminal, which include general resources and special resources. Therefore, when loading different translation engine models, the corresponding virtual machines schedule different physical resources.

[0055] Next, combined Figure 2 This section details the specific implementation methods for each of the above steps.

[0056] Step S1: Obtain the resource performance parameters of the translation terminal.

[0057] In this step, it is necessary to obtain the current time point of each translation terminal. General resource utilization rate and utilization rate of special resources .

[0058] As an example, the physical resources of a translation terminal include CPU, GPU, graphics card, sound card, FPGA, etc.

[0059] Generally speaking, CPUs and graphics cards can be considered general-purpose resources, while GPUs, sound cards, FPGAs, etc., can be considered dedicated resources.

[0060] Of course, those skilled in the art can also determine general resources and special resources based on the type of physical resources actually contained in the translation terminal and the frequency of resource usage.

[0061] For example, resources with high usage (called by most processes) can be called general resources, while resources with low usage and called by only specific processes can be called special resources.

[0062] For further improvements, see Figure 2 Step S1 further includes:

[0063] Based on time series models, the translation terminal is predicted within a time period. Average utilization rate of general resources within and average utilization rate of special resources ;in, This is the preset translation time waiting value.

[0064] As can be seen, this embodiment not only considers the current real-time resource performance parameters, but also includes the predicted resource performance parameters during the future waiting period. It can reasonably take into account the state changes during the waiting period, which is more in line with the actual translation scenario.

[0065] Specifically, the preset translation time waiting value Based on the number of current translation sequences N and the required number of different modes included in each current translation sequence. And the translation terminal at the current time node General resource utilization rate and utilization rate of special resources Sure.

[0066] Next, step S2 is executed: based on the resource performance parameters of the translation terminal, the resource sharing method of the virtual machine is determined.

[0067] Specifically, the virtual machine's resource sharing methods include time-sharing sharing and parallel dedicated sharing. The differences between these two sharing methods are briefly described below.

[0068] Assume the physical resources of the translation terminal include:

[0069] The system features a 6-core GPU, an 8-core CPU, 1000MB of video memory, and 500MB of sound card resources (5 sound cards). Currently, three virtual machines are running: Vm1, Vm2, and Vm3.

[0070] The resource allocation method for time-sharing sharing of virtual machines can be, for example:

[0071] In the first period, Vm1 can access 6-core GPU, 6-core CPU, and 800MB of video memory resources; at this time, Vm2 and Vm3 obviously cannot access GPU resources; in the second period, Vm1 releases all accessed resources (reclaims Vm1), and then Vm2 and Vm3 can access the corresponding physical resources according to their own needs.

[0072] The resource allocation method for virtual machines to enjoy parallel and exclusive access can be, for example:

[0073] The Vm1 utilizes a 3-core GPU, a 3-core CPU, 800MB of video memory, and 100MB of sound card resources.

[0074] The Vm2 utilizes 3-core GPU, 2-core CPU, and 100MB of video memory resources.

[0075] The Vm3 uses 1 CPU core and 3 sound card resources (300M).

[0076] It is understandable that the resource sharing method of virtual machines can be randomly set at the beginning, such as all virtual machines sharing resources in a time-sharing manner, all virtual machines sharing resources in a parallel manner, or a random part sharing resources in a parallel manner plus another random part sharing resources in a time-sharing manner.

[0077] At this time, see Figure 2 The resource performance parameters of the translation terminal are adjusted, and the resource sharing method of the virtual machine is adjusted as follows:

[0078] like and Then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to parallel exclusive use.

[0079] As can be seen, the above judgment method not only considers the real-time resource utilization rate of different categories, but also the resource utilization rate of different categories during the future waiting period.

[0080] like and This means that the resource idle rate, especially the idle rate of dedicated resources, is gradually increasing. Therefore, the resource sharing mode of at least one virtual machine running on the translation terminal can be adjusted to parallel exclusive use to meet the scheduling needs of the translation engine model of subsequent virtual machines.

[0081] Specifically, see step S3: Determine the target virtual machine for the current translation sequence based on the virtual machine's resource sharing method and the mode requirements of the current translation sequence.

[0082] The current translation sequence mode requirements include plain text translation, image-text translation, plain audio translation, or video translation.

[0083] If the current translation sequence requires plain text translation, it means that the output of the current translation sequence can be in plain text format. In this case, the translation engine model that performs the translation for the current translation sequence only needs to schedule CPU resources (running the model) and GPU resources (outputting the result).

[0084] If the current translation sequence requires image-text translation, it means that if the output of the current translation sequence needs to be in image-text format, including not only text output but also image output, then the translation engine model that performs the translation for the current translation sequence needs to schedule CPU resources (running the model) + GPU resources (image rendering) + graphics card resources (output results).

[0085] If the current translation sequence requires video translation, it means that if the output of the current translation sequence needs to be in video format, the translation engine model that performs the translation for the current translation sequence needs to schedule CPU resources (running the model) + GPU resources (image / video rendering) + graphics card resources (output results).

[0086] If the current translation sequence requires pure audio translation, it means that if the output of the current translation sequence is in pure audio translation format, the translation engine model that performs the translation for the current translation sequence needs to schedule CPU resources (running the model) + sound card resources (outputting the result).

[0087] It is understandable that text or video translation places high demands on system resource scheduling. It requires more dedicated resources and cannot conflict with other processes (virtual machines), otherwise it will lead to significant delays and reduce user experience.

[0088] Therefore, in Figure 2 In this context, step S3 specifically includes:

[0089] When the current translation sequence requires either text-to-image translation or video translation, the virtual machine with resource sharing mode set to parallel dedicated use will be used as the target virtual machine for the current translation sequence.

[0090] Of course, if the mode requirement for all current translation sequences is plain text translation or plain audio translation, then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to time-sharing.

[0091] Figures 1-2 The method can be executed by an electronic device including a memory and a processor, such as a server or cluster, wherein the processor executes the program instructions stored in the memory to implement steps S1-S3 of the method.

[0092] Specifically, when executed using computer program instructions, the preset translation time waiting value... Based on the number N of the current translation sequences and the required number of different modes included in the current translation sequences. And the translation terminal at the current time node General resource utilization rate and utilization rate of special resources The pseudocode for a given algorithm can be represented as follows:

[0093]

[0094] in, , The weighting coefficient is adjustable. This is the preset maximum translation time waiting time.

[0095] The preset upper limit for translation time waiting can be determined based on experience or user surveys; that is, it represents the longest tolerable waiting time. It is understood that in a typical embodiment, this upper limit can be directly set... As The technical solution of this application can also be achieved by using it.

[0096] The current translation sequence includes the required number of different patterns. This refers to the number of plain text translation, image-text translation, plain audio translation, or video translation tags contained in the current translation sequence.

[0097] For example, a certain translation sequence (i=1, 2, ..., N) If both plain text output and plain audio output are required, then... .

[0098] To achieve Figures 1-2 The method described, Figure 3 This diagram illustrates the architecture of a resource-sharing-based translation system according to an embodiment of the present invention.

[0099] exist Figure 3 In the above, the resource-sharing-based translation system includes at least one translation terminal, on which multiple virtual machines run.

[0100] The system also includes:

[0101] The performance parameter acquisition unit is used to acquire the translation terminal at the current time point. The first resource performance parameter;

[0102] Translation sequence acquisition unit, used to acquire the sequence at the current time point. The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. ;

[0103] Translation time wait value determination unit, based on the current time point The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. The translation time waiting value is determined by the first resource performance parameter of the translation terminal at the current time node. ;

[0104] The performance parameter prediction unit is used to predict the translation terminal's performance within a time period. The second resource performance parameter within;

[0105] The resource sharing mode adjustment unit is used to adjust the resource sharing mode based on the first resource performance parameter of the translation terminal at the current time node and the time segment of the translation terminal. The second resource performance parameter within adjusts the resource sharing mode of at least one virtual machine running on the translation terminal;

[0106] The target virtual machine determination unit is used to determine the target virtual machine for each translation sequence to be processed based on the resource sharing method of the virtual machine and the mode requirements of each translation sequence to be processed.

[0107] The translation result output unit is used to input each translation sequence to be processed into the target virtual machine for translation, and output the translation result that meets the pattern requirements of each translation sequence to be processed;

[0108] Each translation sequence to be processed must meet one of the following mode requirements: plain text translation, image-text translation, plain audio translation, or video translation.

[0109] Each unit of the system can execute the steps of the aforementioned method embodiments, specifically including:

[0110] The translation terminal is at the current time point The first resource performance parameter includes the translation terminal at the current time node. general resource utilization rate and utilization rate of special resources ;

[0111] The translation terminal is in the time segment The second resource performance parameters include:

[0112] The translation terminal is in the time segment Average utilization rate of general resources within and average utilization rate of special resources .

[0113] The resource sharing mode adjustment unit is based on the first resource performance parameter of the translation terminal at the current time node and the time segment of the translation terminal. The second resource performance parameter within adjusts the resource sharing mode of at least one virtual machine running on the translation terminal, specifically including:

[0114] like and Then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to parallel exclusive use.

[0115] This invention focuses on the resource shortage or output delay problems that may occur in text-image translation or video translation. Therefore, the above technical solutions are all designed for these two translation modes.

[0116] Understandably, if the mode requirement for all current translation sequences is plain text translation or plain audio translation, then the resource sharing mode of at least one virtual machine running on the translation terminal will be directly adjusted to time-sharing.

[0117] The translation time waiting value determination unit is based on the current time point. The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. The translation time waiting value is determined by the first resource performance parameter of the translation terminal at the current time node. The method is as follows:

[0118]

[0119] in, , The weighting coefficient is adjustable. This is the preset maximum translation time waiting time.

[0120] As can be seen, compared with the prior art, the improvements of the present invention include at least the following:

[0121] (1) Sequence translation is performed by running virtual machines on the translation terminal. Each virtual machine can schedule resources based on time-sharing or parallel exclusive use, without adopting the unified resource scheduling method of different hardware in the existing technology, which can improve resource utilization efficiency.

[0122] (2) The resource scheduling of different virtual machines can be dynamically adjusted, that is, the resource sharing mode of the virtual machine is determined based on the resource performance parameters of the translation terminal. The resource performance parameters here include not only the real-time resource performance parameters, but also the predicted resource performance parameters during the future waiting period. This can reasonably take into account the state changes during the waiting period and is more in line with the actual translation scenario.

[0123] (3) Based on the resource sharing method of the virtual machine and the mode requirements of the current translation sequence, the target virtual machine of the current translation sequence is determined, which can make the real-time matching degree between the translation sequence and the virtualized translation resources higher, thereby improving the translation efficiency.

[0124] Overall, this invention can reduce the possibility of resource scheduling conflicts, especially insufficient dedicated resources, in translation engine models when there are diverse output requirements (text / image output or video output) in the translation sequence, and avoid excessive delays in translation output results that lead to poor user experience.

[0125] Of course, it is understood that each embodiment of the present invention can achieve one of the effects on its own, and the combination of multiple embodiments of the present invention can achieve all the above effects. However, it is not required that each embodiment of the present invention achieve all the above advantages and effects, because each embodiment of the present invention can constitute a separate technical solution and make one or more contributions to the prior art.

[0126] For any module structures not specifically defined in this invention, the existing technical specifications shall prevail. The existing technical specifications mentioned in the foregoing background and specific embodiments sections are considered part of this invention and are used to understand the meaning of certain technical features or parameters. The scope of protection of this invention is determined by the actual contents of the claims.

Claims

1. A resource sharing based translation method, characterized by, The method includes the following steps: S1: Obtain the resource performance parameters of the translation terminal; S2: Determine the resource sharing method of the virtual machine based on the resource performance parameters of the translation terminal; S3: Based on the virtual machine's resource sharing method and the current translation sequence's mode requirements, determine the target virtual machine for the current translation sequence; S4: Input the current translation sequence into the target virtual machine for translation, and output the translation result that meets the requirements of the current translation sequence mode; The resource sharing methods of the virtual machines include time-sharing sharing and parallel dedicated sharing; The specific steps of obtaining the resource performance parameters of the translation terminal in step S1 include: acquiring a current time node of the translation terminal general resource utilization and special resource utilization ; Based on time series models, the translation terminal is predicted within a time period. Average utilization rate of general resources within and average utilization rate of special resources ;in, This is the preset translation time waiting value; Step S2 specifically includes: like and Then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to parallel exclusive use.

2. The translation method based on resource sharing as described in claim 1, characterized in that, At least two virtual machines are running on the translation terminal; Multiple virtual machines running on the translation terminal share the physical resources of the translation terminal, which include general resources and special resources.

3. The translation method based on resource sharing as described in claim 1, characterized in that, The current translation sequence mode requirements include plain text translation, image-text translation, plain audio translation, or video translation.

4. The translation method based on resource sharing as described in claim 3, characterized in that, Step S3 specifically includes: When the current translation sequence requires either text-to-image translation or video translation, the virtual machine with resource sharing mode set to parallel dedicated use will be used as the target virtual machine for the current translation sequence.

5. The translation method based on resource sharing as described in claim 4, characterized in that, The preset translation time waiting value Based on the number of current translation sequences N and the required number of different modes included in each current translation sequence. And the translation terminal at the current time node General resource utilization rate and utilization rate of special resources Sure.

6. A translation system based on resource sharing, the system comprising at least one translation terminal, wherein multiple virtual machines run on the translation terminal; Its features are, The system also includes: The performance parameter acquisition unit is used to acquire the translation terminal at the current time point. The first resource performance parameter; Translation sequence acquisition unit, used to acquire the sequence at the current time point. The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. ; Translation time wait value determination unit, based on the current time point The number of translation sequences N to be processed, and the number of different modes required for each translation sequence. The translation time waiting value is determined by the first resource performance parameter of the translation terminal at the current time node. ; The performance parameter prediction unit is used to predict the translation terminal's performance within a time period. The second resource performance parameter within; The resource sharing mode adjustment unit is used to adjust the resource sharing mode based on the first resource performance parameter of the translation terminal at the current time node and the time segment of the translation terminal. The second resource performance parameter within adjusts the resource sharing mode of at least one virtual machine running on the translation terminal; The target virtual machine determination unit is used to determine the target virtual machine for each translation sequence to be processed based on the resource sharing method of the virtual machine and the mode requirements of each translation sequence to be processed. The translation result output unit is used to input each translation sequence to be processed into the target virtual machine for translation, and output the translation result that meets the pattern requirements of each translation sequence to be processed; Each translation sequence to be processed must meet the following mode requirements: plain text translation, image-text translation, plain audio translation, or video translation.

7. A translation system based on resource sharing as described in claim 6, characterized in that: The translation terminal is at the current time point The first resource performance parameter includes the translation terminal at the current time node. general resource utilization rate and utilization rate of special resources ; The translation terminal is in the time segment The second resource performance parameters include: The translation terminal is in the time segment Average utilization rate of general resources within and average utilization rate of special resources .

8. A translation system based on resource sharing as described in claim 7, characterized in that: The resource sharing mode adjustment unit is based on the first resource performance parameter of the translation terminal at the current time node and the time segment of the translation terminal. The second resource performance parameter within adjusts the resource sharing mode of at least one virtual machine running on the translation terminal, specifically including: like and Then the resource sharing mode of at least one virtual machine running on the translation terminal will be adjusted to parallel exclusive use.

Images