Computer-implemented method for processing a prompt

Abstract

According to an embodiment, a computer-implemented method for processing a prompt comprises: obtaining (101) a prompt comprising a request section and a data section; obtaining (102) a maximum prompt length; in response to the prompt being longer than the maximum prompt length, iteratively performing, until a preconfigured condition (107) is met: during a first iteration, dividing (103) the data section into at least one subsection; during an iteration other than the first iteration, dividing (103) a result data of a previous iteration into at least one subsection; forming (104) at least one subprompt, wherein each sub-prompt in the at least one subprompt comprises the request section and a subsection from the at least one subsection; providing (105) the at least one subprompt to a machine learning model, thus obtaining at least one result from the machine learning model; and combining (106) the at least one result into a result data; and after finishing the iteration, providing (108) the result data of a last iteration.

Description

COMPUTER- IMPLEMENTED METHOD FOR PROCESSING A PROMPTTECHNICAL FIELD

[0001] The present disclosure relates to prompt processing, and more particularly to a computer-implemented method for processing a prompt , a computing device, and a computer program product .BACKGROUND

[0002] Various machine learning models , such as large language models , are typically provided as cloud-based services . A user can use these model s by providing so- called prompts and the models provide results in response to these prompts . The usage of these services is often limited in various ways . For example , the length of the prompts the user can provide to the model is often limited .SUMMARY

[0003] This summary is provided to introduce a selection of concepts in a s implif ied form that are further described below in the detailed description . This summary is not intended to identify key features or essential features of the claimed subj ect matter, nor is it intended to be used to limit the scope of the claimed subj ect matter .

[0004] It is an obj ective to provide a computer-implemented method for processing a prompt , a computingdevice , and a computer program product . The foregoing and other obj ectives are achieved by the features of the independent claims . Further implementation forms are apparent from the dependent claims , the description and the figures .

[0005] According to a first aspect, a computer-implemented method for processing a prompt comprises : obtaining a prompt comprising a request section and a data section; obtaining a maximum prompt length; in response to the prompt being longer than the maximum prompt length, iteratively performing, until a preconfigured condition is met : during a first iteration, dividing the data section into at least one subsection ; during an iteration other than the first iteration, dividing a result data of a previous iteration into at least one subsection ; forming at least one subprompt , wherein each subprompt in the at least one subprompt comprises the request section and a subsection from the at least one subsection ; providing the at least one subprompt to a machine learning model , thus obtaining at least one result from the machine learning model ; and combining the at least one result into a result data ; and after finishing the iteration, providing the result data of a last iteration .

[0006] In an implementation form of the first aspect, the dividing the result data of the previous iteration into the at least one subsection comprises dividing the result data of the previous iteration into the at least one subsection based on the maximum prompt length .

[0007] In another implementation form of the first aspect , the dividing the result data of the previous iteration into the at least one subsection comprises dividing the result data of the previous iteration into the at least one subsection in such a manner that each subprompt in the at least one subprompt is shorter than or of equal length to the maximum prompt length .

[0008] In another implementation form of the first aspect , the preconfigured condition comprises the at least one subprompt being shorter than the maximum prompt length, the at least one result comprising one result , the at least one subprompt compri sing one subprompt , and / or reaching a maximum number of iterations .

[0009] In another implementation form of the first aspect , the request section comprises a request to summari ze the data section .

[0010] In another implementation form of the first aspect , the data section comprises text data .

[0011] In another implementation form of the first aspect , the text data comprises a transcript of an audio data and / or text of a conversation .

[0012] In another implementation form of the first aspect , during at least one iteration, the at least one subsection comprises a plurality of subsections and at least two subsections overlap in the plurality of subsections .

[0013] In another implementation form of the first aspect , the machine learning model comprises a large language model .

[0014] According to a second aspect , a computing device compri ses at least one processor and at least one memory including computer program code , the at least one memory and the computer program code being configured to , with the at least one proces sor, cause the computing device to perform the method according to the first aspect .

[0015] According to a third aspect , a computer program product comprises program code configured to perform the method according to the first aspect when the computer program product is executed on a computer .

[0016] Many of the attendant features wil l be more readily appreciated as they become better understood by reference to the following detailed description considered in connection with the accompanying drawings .DESCRIPTION OF THE DRAWINGS

[0017] In the following, example embodiments are described in more detail with reference to the attached figures and drawings , in which :

[0018] Fig . 1 illustrates a flow chart representation of a method according to an embodiment ;

[0019] Fig . 2 illustrates a schematic representation of a data flow according to an embodiment ;

[0020] Fig . 3 illustrates a flow chart representation of a method according to an embodiment ; and

[0021] Fig . 4 illustrates a schematic representation of a computing device according to an embodiment .

[0022] In the following, like reference numerals are used to des ignate li ke parts in the accompanying drawings .DETAILED DESCRIPTION

[0023] In the following description, reference is made to the accompanying drawings , which form part of the disclosure , and in which are shown, by way of illustration, specific aspects in which the present disclosure may be placed . It is understood that other aspects may be utilised, and structural or logical changes may be made without departing from the scope of the present disclosure . The following detailed description, therefore , is not to be taken in a limiting sense , as the scope of the present disclosure is defined by the appended claims .

[0024] For instance , it is understood that a disclosure in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa . For example , if a specific method step is described, a corresponding device may include a unit to perform the described method step, even if such unit is not explicitly described or il lustrated in the f igures . On the other hand, for example , if a specific apparatus is described based on functional units , a corresponding method may include a step performing the described functionality, even if such step is not explicitly described or illustrated in the figures . Further, it is understood that the featuresof the various example aspects described herein may be combined with each other, unless specifically noted otherwise .

[0025] Fig . 1 illustrates a flow chart representation of a method according to an embodiment .

[0026] According to an embodiment , a computer-implemented method 100 for processing a prompt comprises obtaining 101 a prompt comprising a request section and a data section .

[0027] Herein a prompt may refer to , for example , a request or a question to be provided to a machine learning model , such as a language model . The request section of the prompt may comprise , for example , a request to process the data provided in the data section in some manner . For example , the request section may comprise a request to make the text data comprised in the data section more concise . The request section may comprise the request in a natural language and / or the data section may comprise the data in a natural language .

[0028] For example , the prompt may comprise the question "when was the first GSM network implemented?" . The prompt may further comprise instructions for providing the answer, such as "provide the dates in the answer as numbers" .

[0029] The data section may comprise any data to be used by the machine learning model , such as a language model . For example , if the request section comprises a question or some other type or request , the data section can comprise data based on which the machine learningmodel should provide the answer . The data can be used, for example , for grounding and / or retrieval-augmented generation .

[0030] The method 100 may further comprise obtaining 102 a maximum prompt length .

[0031] The maximum prompt length may refer to a maximum length of a prompt that can be provided to a machine learning model . The maximum prompt length may be due to , for example , a technical limitation of the machine learning model , a limitation due to a subscription plan to the machine learning model , or any other reason .

[0032] The obtaining the maximum prompt length may comprise , for example , obtaining the maximum prompt length from memory . Alternatively, the maximum prompt length may be , for example , set as a constant or as a variable in program code .

[0033] The maximum prompt length may be expressed, for example , in terms of number of characters , number of bytes , number of bits , number of tokens , or similar .

[0034] The method 100 may further comprise , in response to the prompt being longer than the maximum prompt length, iteratively performing iteration steps 110 , until a preconfigured condition 107 is met .

[0035] In some embodiments , in response to the prompt not being longer than the maximum prompt length, the prompt may be provided to a machine learning model without performing the iteration steps .

[0036] The preconfigured condition may comprise , for example , one condition or a plurality of conditions . Forexample , in some embodiments , the preconfigured condition may comprise a plurality of subconditions combined using logical operators , such as OR, AND, and / or XOR . For example , in some embodiments , the preconfigured condition may have to form "f irst subcondition AND second subcondition" . Thus , the preconfigured condition is true if both the first subcondition and the second subcondition is true . Alternatively or additionally, the preconfigured condition may comprise any type of more complex condition implemented using, for example , program code .

[0037] The iteratively performing iteration steps 110 until a preconfigured condition is met may comprise repeatedly performing the same steps / operations until the preconfigured condition is met . During each iteration, the data used for the processing can change but the steps / operations of the processing can remain the same . The iteration can be implemented us ing, for example , a loop in program code .

[0038] The method 100 may further comprise , in the iteration 110 , during a first iteration, dividing 103 the data section into at least one subsection .

[0039] The first iteration may refer to the first time the operations of the iteration 110 are performed for a prompt .

[0040] During some iterations , the at least one subsection may comprise a plurality of subsections . During some iterations , the at least one subsection may comprise one subsection . For example , in some cases , duringthe last iteration, the combined length of the result data of the previous iteration and the request part may be shorter than the maximum prompt length . In such cases , the result data of the previous iteration may be divided into one subsection .

[0041] The method 100 may further comprise , in the iteration 110 , during an iteration other than the first iteration, dividing 103 a result data of a previous iteration into at least one subsection .

[0042] Herein, a previous iteration may also be referred to as a preceding iteration or similar .

[0043] For example , during a second iteration, the result data of the first iteration can be divided into at least one subsection, during a third iteration , the result data of the second iteration can be divided into at least one subsection etc . During an nth iteration, the result data of the n- 1 iteration can be divided into at least one subsection .

[0044] The method 100 may further comprise , in the iteration, forming 104 at least one subprompt , wherein each subprompt in the at least one subprompt compri ses the request section and a subsection from the at least one subsection .

[0045] Each subprompt may comprise a corresponding subsection . Thus , each subprompt may comprise a different subsection and the request section .

[0046] The method 100 may further comprise , in the iteration 110 , providing 105 the at least one subpromptto a machine learning model , thus obtaining at least one result from the machine learning model .

[0047] The machine learning model may comprise any machine learning model capable of processing the prompt .

[0048] Providing the at least one subprompt to the machine learning model may comprise , for example , providing the at least one subprompt to the machine learning model via a telecommunications network, such as over the internet . The machine learning model may be , for example , provided as a cloud service . Alternatively, the machine learning model may be running on a server in a local area network (LAN) or locally, for example on a computing device performing the method 100 .

[0049] The method 100 may further comprise , in the iteration 110 , combining 106 the at least one result into a result data .

[0050] The combining the at least one result into the result data may comprise , for example , copying each result from the at least one result to a data structure corresponding to the result data, such as a buf fer, an array, a list , or similar .

[0051] The method 100 may further comprise , after finishing the iteration 110 , providing 108 the result data of a last iteration .

[0052] The providing the result data of a last iteration may comprise , for example , providing the result data to , for example , a software function / method / sub- program / subroutine , a module , and / or a device . For example , if the method 100 is implemented as a softwarefunction, the function can return result data to the procedure calling the function .

[0053] For example , in the embodiment of Fig . 1 , when the preconfigured condition 107 is not met , the procedure can loop back to operation 103 and the iteration 110 can be repeated . When the preconfigured condition 107 is met , the procedure can move to operation 108 .

[0054] It should be appreciated that at least some of the operations of the method 100 may be performed in various orders . For example , the obtaining a prompt 101 and obtaining 102 the maximum prompt length 102 may be performed in any order .

[0055] At least some embodiments disclosed herein can, for example , make it possible to provide longer prompts to a machine learning model , such as language models .

[0056] For example , in the case of requesting a language model to make text data, such as an email or a chat conversation, more concise , the prompt can easily be longer than the maximum prompt length .

[0057] At least some embodiments disclosed herein can, for example , improve results provided by language models to long prompts .

[0058] At least some embodiments disclosed herein can, for example , enable using language models for larger amounts of data .

[0059] At least some embodiments disclosed herein can, for example , make it possible to use language models for a broader scope of applications .

[0060] Fig . 2 illustrates a schematic representation of a data flow according to an embodiment .

[0061] In the embodiment of Fig . 2 , a prompt 201 comprises a request section 202 and a data section 203 . The data section 203 may be referred to as datal . The prompt201 is longer than the maximum prompt length 250 .

[0062] According to an embodiment , the request section202 comprises a request to summari ze the data section 203 .

[0063] According to an embodiment , the data section203 comprises text data .

[0064] For example , the request section 202 may comprise a request to make text data comprised in the data section 203 more concise .

[0065] According to an embodiment , the text data comprises a transcript of an audio data and / or text of a conversation .

[0066] For example , the text data may comprise a transcript of an audio data of a customer service situation . Alternatively or additionally, the text data may comprise an email conversation or a chat conversation . Such conversations may be from, for example , customer service situations .

[0067] During a first iteration, the data section 203 can be divided into three subsections 204 , referred to as datal_l , data 1_2 , and data 1_3 . Based on the subsections 204 , three subprompts 205 can be formed,wherein each subprompt comprises a corresponding subsection 204 and the request section 202 . Each subprompt 205 may be shorter than the maximum prompt length 250 .

[0068] In some embodiments , each subprompt 205 of an iteration may be of the same length . In other embodiments , the subprompts 205 may have different lengths . For example , in some embodiments , the subsections may be formed in such a fashion that sentences are not split into different subsections . This may cause different subsections to have different lengths .

[0069] Each of the subprompts 205 can be provided to the machine learning model 210 . The machine learning model 210 can provide a result 206 in response to each subprompt 205 .

[0070] According to an embodiment, the machine learning model comprises a large language model .

[0071] The results 206 may be referred to as re- sultl_l , resultl_2 , and resultl_3 . The results 206 can be combined into a result data 207 . The result data 207 of the first iteration can be referred to as data2 .

[0072] According to an embodiment , the dividing the result data of the previous iteration into the at least one subsection comprises dividing the result data of the previous iteration into the at least one subsection based on the maximum prompt length .

[0073] According to an embodiment , the dividing the result data of the previous iteration into the at least one subsection comprises dividing the result data of the previous iteration into the at least one subsection insuch a manner that each subprompt in the at least one subprompt is shorter than or of equal length to the maximum prompt length .

[0074] For example, in the embodiment of Fig . 2 , during a second iteration, the result data 207 of the first iteration can be divided into two subsections 204 , referred to as data2_l and data2_2 . Based on the subsections 204 , two subprompts 205 can be formed, wherein each subprompt comprises a corresponding subsection 204 and the request section 202 . Each subprompt 205 may be shorter than the maximum prompt length 250 .

[0075] Each of the subprompts 205 can be provided to the machine learning model 210 . The machine learning model 210 can provide a result 206 in response to each subprompt 205 . The results 206 may be referred to as result2_l and result2_2 . The results 206 can be combined into a result data 207 . The result data 207 of the second iteration can be referred to as data3 .

[0076] During a third iteration, the result data 207 of the second iteration can be divided into one subsection 204 , referred to as data3 . Based on the subsection 204 , one subprompt 205 can be formed, wherein the subprompt comprises the subsection 204 and the request section 202 .

[0077] It should be appreciated that when the result data 207 is divided into one subsection 204 , this may be implemented in various ways . For example , in some embodiments , it can be checked whether a subprompt formed by the request and the result data 207 of theprevious iteration is shorter than the maximum prompt length and, in response to it being shorter or equal to the maximum prompt length, the subprompt 205 comprising the request and the result data 207 of the previous iteration can be formed . Thus , the dividing the result data 207 into the at least one subsection 204 can be implicit . In other embodiments , the result data 207 of the previous iteration can be divided into the at least one subsection 204 and the at least one subprompt 205 can be formed in the same fashion regardless of how many subsections the at least one subsection comprises .

[0078] The subprompt 205 can be provided to the machine learning model 210 . The machine learning model 210 can provide a result 206 in response to the subprompt 205 . The result 206 may be referred to as result3 .

[0079] The result data of the last iteration can be provided when the iteration ends . For example , in the embodiment of Fig . 2 , the preconfigured condition may comprise the at least one subprompt being shorter than the maximum prompt length or the at least one result comprising one result . The subprompt 205 of the third iteration may be shorted than the maximum prompt length . Thus the iteration may be stopped after the third iteration . Alternatively or additionally, the iteration may be stopped due to the result 206 of the third iteration comprising one result .

[0080] According to an embodiment , during at least one iteration, the at least one subsection comprises a plurality of subsections and at least two subsections overlap in the plurality of subsections .

[0081] For example , in the embodiment of Fig . 2 , the at least one subsection 204 of the first or second iteration may comprise at least two overlapping subsections . Herein, when two subsections overlap, the subsections may comprise at least some common data . For example , the end of a section may comprise the same data, such as text , as the start of a following section . This can improve results provided by the machine learning model .

[0082] Fig . 3 illustrates a flow chart representation of a method according to an embodiment .

[0083] The embodiment of Fig . 3 only illustrates one possible implementation of at least some parts of the method 100 . The method 100 can also be implemented in various other ways .

[0084] In the embodiment of Fig . 3 , the iteration 110 further comprises computing 301 the number of subprompts . For example , the computing 301 the number of subprompts may comprise computing how many subprompts are needed so that each subprompt in the at least one subprompt is shorter than or of equal length to the maximum prompt length .

[0085] According to an embodiment , the preconfigured condition comprises the at least one subprompt being shorter than the maximum prompt length, the at least oneresult comprising one result , the at least one subprompt comprising one subprompt , and / or reaching a maximum number of iterations .

[0086] For example , in the embodiment of Fig . 3 , the preconfigured condition 107 comprises checking 302 whether the at least one prompt comprises more than one prompt . I f the at least one prompt does not comprise more than one prompt , the iteration can be stopped, and the procedure can move to operation 108 . Otherwise , the procedure can proceed to operation 303 .

[0087] In the embodiment of Fig . 3 , the preconfigured condition 107 further comprises checking 303 whether a maximum number of iterations has been reached . I f the maximum number of iterations has been reached, the iteration can be stopped, and the procedure can move to operation 108 . Otherwise , the procedure can proceed to operation 301 and another iteration can be performed .

[0088] The maximum number of iterations can be , for example, preconfigured . For example, the maximum number of iterations may be , for example , set as a constant or as a variable in program code .

[0089] Fig . 4 illustrates a schematic representation of a computing device according to an embodiment .

[0090] According to an embodiment , a computing device 400 comprises at least one processor 401 and at least one memory 402 including computer program code , the at least one memory 402 and the computer program code configured to , with the at least one processor 401 , cause the computing device 400 to perform the method 100 .

[0091] The computing device 400 may comprise at least one processor 401. The at least one processor 401 may comprise, for example, one or more of various processing devices, such as a co-processor, a microprocessor, a digital signal processor (DSP) , a processing circuitry with or without an accompanying DSP, or various other processing devices including integrated circuits such as, for example, an application specific integrated circuit (ASIC) , a field programmable gate array (FPGA) , a microprocessor unit (MCU) , a hardware accelerator, a special-purpose computer chip, or the like.

[0092] The computing device 400 may further comprise a memory 402. The memory 402 may be configured to store, for example, computer programs and the like. The memory 402 may comprise one or more volatile memory devices, one or more non-volatile memory devices, and / or a combination of one or more volatile memory devices and nonvolatile memory devices. For example, the memory 402 may be embodied as magnetic storage devices (such as hard disk drives, magnetic tapes, etc.) , optical magnetic storage devices, and semiconductor memories (such as mask ROM, PROM (programmable ROM) , EPROM (erasable PROM) , flash ROM, RAM (random access memory) , etc.) .

[0093] The computing device 400 may further comprise other components not illustrated in the embodiment of Fig. 4. The computing device 400 may comprise, for example, an input / output bus for connecting the computing device 400 to other devices. Further, a user may control the computing device 400 via the input / output bus.

[0094] When the computing device 400 is configured to implement some functionality, some component and / or components of the computing device 400 , such as the at least one processor 401 and / or the memory 402 , may be configured to implement this functionality . Furthermore , when the at least one processor 401 is configured to implement some functionality, this functionality may be implemented using program code comprised, for example , in the memory .

[0095] The computing device 400 may be implemented at least partially using, for example , a computer, some other computing device , or similar .

[0096] The method 100 and / or the computing device 400 may be utilised in, for example , automatic speech recognition (ASR) applications such as in a so-called voice- bot . A voicebot may be configured to obtain information from users by, for example, phone and convert the voice information into text information using ASR . The information may then be summari zed and / or checked us ing the method 100 and / or the computing device 400 . For example , the method 100 and / or the computing device 400 can be used to check whether the user has provided all needed information . Alternatively or additionally, ASR can be used to transcribe a phone call , such as a customer service phone call , and the phone cal l can be summari zed using the method 100 and / or the computing device 400 . For example , the transcript of the phone call and / or the summari zed transcript can be stored in a customer relationship management (CRM) system .

[0097] Any range or device value given herein may be extended or altered without losing the effect sought . Also any embodiment may be combined with another embodiment unless explicitly disallowed .

[0098] Although the subj ect matter has been described in language specific to structural features and / or acts , it is to be understood that the subj ect matter defined in the appended claims is not necessarily limited to the specific features or acts described above . Rather, the specific features and acts described above are disclosed as examples of implementing the claims and other equivalent features and acts are intended to be within the scope of the claims .

[0099] It will be understood that the benefits and advantages described above may relate to one embodiment or may relate to several embodiments . The embodiments are not limited to those that solve any or all of the stated problems or those that have any or all of the stated benefits and advantages . It wil l further be understood that reference to ' an ' item may refer to one or more of those items .

[0100] The steps of the methods described herein may be carried out in any suitable order, or simultaneously where appropriate . Additionally, individual blocks may be deleted from any of the methods without departing from the spirit and scope of the subj ect matter described herein . Aspects of any of the embodiments described above may be combined with aspects of any of theother embodiments described to form further embodiments without losing the effect sought .

[0101] The term ' comprising ' is used herein to mean including the method, blocks or elements identified, but that such blocks or elements do not comprise an exclusive list and a method or apparatus may contain additional blocks or elements .

[0102] It will be understood that the above description is given by way of example only and that various modif ications may be made by those ski lled in the art . The above specification, examples and data provide a complete description of the structure and use of exemplary embodiments . Although various embodiments have been described above with a certain degree of particularity, or with reference to one or more individual embodiments , those skilled in the art could make numerous alterations to the disclosed embodiments without departing from the spirit or scope of this specification .

Claims

CLAIMS :

1. A computer-implemented method (100) for processing a prompt, the method (100) comprising: obtaining (101) a prompt comprising a request section and a data section; obtaining (102) a maximum prompt length; in response to the prompt being longer than the maximum prompt length, performing an iteration process, until a preconfigured condition (107) is met, wherein the preconfigured condition comprises the at least one subprompt being shorter than the maximum prompt length, the at least one result comprising one result, the at least one subprompt comprising one subprompt, and / or reaching a maximum number of iterations, the iteration process comprising: during a first iteration, dividing (103) the data section into at least one subsection; during an iteration other than the first iteration, dividing (103) a result data of a previous iteration into at least one subsection; forming (104) at least one subprompt, wherein each subprompt in the at least one subprompt comprises the request section and a subsection from the at least one subsection; providing (105) the at least one subprompt to a machine learning model, thus obtaining at least one result from the machine learning model, wherein the machine learning model comprises a large language model; andcombining (106) the at least one result into a result data; and after finishing the iteration process, providing (108) the result data of a last iteration.

2. The computer-implemented method (100) according to claim 1, wherein the dividing the result data of the previous iteration into the at least one subsection comprises dividing the result data of the previous iteration into the at least one subsection based on the maximum prompt length.

3. The computer-implemented method (100) according to claim 1 or claim 2, wherein the dividing the result data of the previous iteration into the at least one subsection comprises dividing the result data of the previous iteration into the at least one subsection in such a manner that each subprompt in the at least one subprompt is shorter than or of equal length to the maximum prompt length.

4. The computer-implemented method (100) according to any preceding claim, wherein the request section comprises a request to summarize the data section.

5. The computer-implemented method (100) according to any preceding claim, wherein the data section comprises text data.

6. The computer-implemented method (100) according to claim 5, wherein the text data comprises a transcript of an audio data and / or text of a conversation.

7. The computer-implemented method (100) according to any preceding claim, wherein, during at least one iteration, the at least one subsection comprises a plurality of subsections and at least two subsections overlap in the plurality of subsections.

8. A computing device (400) , comprising at least one processor (401) and at least one memory (402) including computer program code, the at least one memory (402) and the computer program code configured to, with the at least one processor (401) , cause the computing device (400) to perform the method (100) according to any preceding claim.

9. A computer program product comprising program code configured to perform the method (100) according to any of claims 1 - 7 when the computer program product is executed on a computer.

Images