Path determination method based on large model, electronic equipment and storage medium

Abstract

The invention provides a path determination method based on a large model, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the following steps: obtaining a user question, inputting the user question into a target large model, obtaining an implementation path list and an implementation probability list corresponding to the implementation path list, traversing the implementation path list, and obtaining an implementation probability list corresponding to the implementation path list; and matching the implementation path with the target data list, if the matching succeeds, taking the implementation path as the final path, and if the matching fails, obtaining the related field of the implementation field which is not successfully matched, and determining the final path based on the related field, so that the accuracy of the final path of the user problem is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a path determination method, an electronic device, and a storage medium based on a large model. Background Art

[0002] Large models, especially those in the fields of natural language processing (NLP), computer vision, audio processing, etc., have greatly expanded their problem-solving capabilities as their scale and complexity increase. With the development of large models, their influence is spreading in various fields at an unprecedented speed. With the progress of large model technology, large models are changing the way we solve problems and constantly opening up new possibilities. When we solve problems, we will increasingly rely on large models. However, due to the high complexity and difficulty of interpretation of large models, simply using large models to obtain the implementation path of user problems is often inaccurate. Summary of the Invention

[0003] In view of the above technical problems, the technical solution adopted by the present invention is: a path determination method based on a large model, the method comprising the following steps:

[0004] S001, obtaining a user problem of a target user, inputting the user problem into a target large model, and obtaining an implementation path list AA = {AA 1 , AA 2 , …, AA e , …, AA f} and an implementation probability list AB = {AB 1 , AB 2 , …, AB e , …, AB f}, where AB e is the implementation probability value of AA e , the e-th implementation path AA e = {AA e1 , AA e2 , …, AA eh , …, AA eu}, AA eh is the h-th implementation field of the e-th real path and the implementation label corresponding to the h-th implementation field, the value range of h is from 1 to u, where u is the number of implementation fields, and the value range of e is from 1 to f, where f is the number of implementation paths; wherein, AB e is greater than AB e+1 ;

[0005] S002, traversing AA, for AAe, using AA e1 to AA euMatch with the target data list, where the target data list includes a number of target fields and target tags corresponding to each target field;

[0006] S003, if AA e1 To AA eu Both match successfully with the target data list, and AA e Is used as the final path;

[0007] S004, if the match fails, obtain the list to be matched AC = {AC 1 , AC 2 , …, AC e , …, AC f}, AA e The corresponding data to be matched AC e = {AC e1 , AC e2 , …, AC ev , …, AC ew}, AC ev Is the v-th implementation field that fails to match in AA e And the implementation tag corresponding to the v-th implementation field that fails to match, where the value range of v is from 1 to w, and the number w of implementation fields that fail to match satisfies w ≤ u;

[0008] S005, for AC ev , obtain the relevant field AF ev of AC ev , and obtain the target weighted value AG ev from AC ev to AF ev , where the matching results of the relevant tags corresponding to AF ev and AF ev with the target data list are successful; the target weighted value AG ev Is the sum value of the preset weights of the nodes passed by the implementation path from AC ev to AF ev ;

[0009] S006, determine the final path based on AG ev and AB e ;

[0010] According to another aspect of the present invention, there is provided a non-transitory computer-readable storage medium, in which at least one instruction or at least one program segment is stored, and the at least one instruction or the at least one program segment is loaded and executed by a processor to implement the foregoing method.

[0011] According to yet another aspect of the present invention, there is provided an electronic device, including a processor and the foregoing non-transitory computer-readable storage medium.

[0012] The present invention has at least the following beneficial effects: In summary, obtain the user's question, input the user's question into the target large model, obtain the list of implementation paths and the list of implementation probabilities corresponding to the list of implementation paths, traverse AA, and for AA e , use AA e1 to AA eu match with the target data list. If AA e1 to AA eu both match successfully with the target data list, take AA e as the final path. If the match fails, obtain the list to be matched AC, and for AC ev , obtain AC ev 's relevant field AF ev , and obtain AC ev to AF ev 's target weighted value AG ev , based on AG ev and AB e determine the final path. The present invention obtains the implementation path of the user's question through the target large model, and determines the final path based on the implementation probability and the target weighted value of the relevant field, improving the accuracy of the final path of the user's question. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 is a flowchart of a method for determining a path based on a large model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0016] Embodiment 1

[0017] Embodiment 1 of the present invention provides a method for determining a path based on a large model, as Figure 1 shown, the method includes the following steps:

[0018] S001. Obtain the user's question of the target user, input the user's question into the target large model, and obtain the implementation path list AA = {AA 1 , AA 2 , …, AA e , …, AA f} and the implementation probability list AB corresponding to AA = {AB 1 , AB 2 , …, AB e , …, AB f}, where AB e is the implementation probability value of AA e . The e-th implementation path AA e = {AA e1 , AA e2 , …, AA eh , …, AA eu}, and AA eh is the h-th implementation field of the e-th implementation path and the implementation label corresponding to the h-th implementation field. The value range of h is from 1 to u, where u is the number of implementation fields, and the value range of e is from 1 to f, where f is the number of implementation paths. Among them, AB e is greater than AB e+1 .

[0019] S002. Traverse AA. For AAe, use AA e1 to AA eu to match with the target data list. Among them, the target data list includes several target fields and the target labels corresponding to each target field. Specifically, traverse AA and match with the target data list in the order from AA 1 to AA f .

[0020] Specifically, use the implementation field and the implementation label corresponding to the implementation field and the target field and the target label corresponding to the target field to match. If the similarity between the implementation field and the target field is greater than the preset similarity threshold, and the similarity between the implementation label corresponding to the implementation field and the target label corresponding to the target field is greater than the preset similarity threshold, it is considered that the implementation field and the implementation label corresponding to the implementation field match successfully.

[0021] S003. If AA e1 to AA eu all match successfully with the target data list, take AA e as the final path. And use the final path to solve the user's question.

[0022] S004. If the match fails, obtain the list to be matched AC = {AC 1 , AC 2 , …, AC e, …, AC f},AA e The corresponding data to be matched, AC e ={AC e1 , AC e2 , …, AC ev , …, AC ew}, AC ev is AA e The v-th implementation field that fails to match successfully and the implementation label corresponding to the v-th implementation field that fails to match successfully, where the value range of v is from 1 to w, and the number of implementation fields that fail to match successfully, w ≤ u.

[0023] S005. For AC ev , obtain AC ev 's related field AF ev , and obtain AC ev to AF ev 's target weighted value AG ev , where the matching results of AF ev and the related label corresponding to AF ev with the target data list are successful matches; the target weighted value AG ev is the sum value of the preset weights of the nodes passed by the implementation path from AC ev to AF ev . Among them, C ev to AF ev 's preset weights of the nodes passed by the implementation path can be determined according to the actual situation. In an embodiment of the present invention, C ev to AF ev 's preset weights of the nodes passed by the implementation path increase step by step, that is, C ev to AF ev 's preset weight of the α-th node passed by the implementation path is greater than C ev to AF ev 's preset weight of the (α + 1)-th node passed by the implementation path.

[0024] Furthermore, S005 further includes: if obtaining AC ev 's related field AF ev fails, delete AA e from the implementation path list AA.

[0025] S006. Determine the final path based on AG ev and AB e .

[0026] In summary, obtain the user's question, input the user's question into the target large model, obtain the implementation path list and the corresponding implementation probability list of the implementation path list, traverse AA, for AA e , use AAe1 To AA eu Match with the target data list. If AA e1 To AA eu All match successfully with the target data list, then AA e Is used as the final path. If the match fails, obtain the list AC to be matched, and for AC ev , obtain AC ev Related field AF ev , and obtain AC ev To AF ev Target weighted value AG ev , based on AG ev And AB e Determine the final path. The present invention obtains the implementation path of the user's question through the target large model, and determines the final path based on the implementation probability and the target weighted value of the related field, improving the accuracy of obtaining the final path of the user's question.

[0027] Furthermore, S006 also includes:

[0028] S061, obtain the target sum value AG e0 = ∑ w v＝1 AG ev , and based on the target sum value AG e0 Obtain the final value AH e = β 1 × AG e0 + β 2 × AB e , β 1 Is the first probability factor, β 2 Is the second probability factor. In an embodiment of the present invention, β 1 = β 2 .

[0029] S062, obtain AH 0 = max{AH 1 , AH 2 , …, AH e , …, AH f}, obtain the implementation path AA 0 Corresponding to AH 0 .

[0030] S063, use AA 0 And AC ev To AF ev Related paths as the final path.

[0031] In an embodiment of the present invention, use the A* algorithm to obtain AC ev To AF evThe relevant path. In another embodiment of the present invention, the ant colony algorithm is used to obtain AC ev to AF ev The relevant path

[0032] In summary, obtain the target sum value, obtain AH 0 and obtain AH 0 The corresponding implementation path AA 0 , and use AA 0 and AC ev to AF ev The relevant path as the final path

[0033] Embodiment 2

[0034] Embodiment 2 of the present invention provides a method for determining data tags to obtain a target data list. The method includes the following steps

[0035] S100. Obtain the data list of the target user. The data list of the target user includes several fields

[0036] Specifically, the data list A of the target user = {A 1 , A 2 , …, A i , …, A m}, where A i is the i-th field, and the value range of i is from 1 to m, where m is the number of fields of the target user

[0037] S200. Obtain the single field name corresponding to the field

[0038] S300. Mark any single field name as the target field name and obtain the target tag corresponding to the target field name

[0039] Among them, S300 includes the following steps

[0040] S310. Input the target field name into the target large model to obtain the first use tag list corresponding to the target field name and the initial probability value corresponding to each first use tag of the target field name. The first use tag list includes several first use tags

[0041] Specifically, the target large model is the target large language model. Specifically, obtain the historical field name and the historical use tag corresponding to the historical field name, construct the initial large language model, and use the historical field name and the historical use tag corresponding to the historical field name to train the initial large language model to obtain the target large language model

[0042] S320. Combine the target field name with several other single field names except the target field name to obtain a list of combined field names, and input the combined field names into the target large model to obtain a list of second use tags corresponding to the combined field names; the list of combined field names includes several combined field names, and the list of second use tags includes several second use tags.

[0043] Specifically, the target field name N i The corresponding list of combined field names C i ={C i1 , C i2 , …, C ij , …, C in}, C ij is the jth combined field name containing A i . The value range of j is from 1 to n, and n is the number of combined field names. Input C ij into the target large model to obtain the list of second use tags D ij corresponding to C ij ={D ij1 , D ij2 , …, D ijr , …, D ijs}, D ijr is the rth second use tag corresponding to C ij . The value range of r is from 1 to s, and s is the number of second use tags.

[0044] S330. Obtain a list of quantity weights, and based on the list of quantity weights and the combined field names, obtain the weights corresponding to the second use tags and mark them as second use tag weights; wherein, the list of quantity weights includes several quantity weights, and the quantity weight is the weight between the number of field names in the combined field name and the field. Specifically, obtain the list of second use weights F ij ={F ij1 , F ij2 , …, F ijr , …, F ijs}, F ijr is the weight corresponding to D ijr .

[0045] S340. Input the second use tags into the target large model to obtain a list of field values corresponding to the second use tags, and the list of field values includes the probability values of the second use tags in several preset fields. Input the second use tag D ijr into the target large model to obtain the list of field values E ijr corresponding to the second use tag D ijr ={E ijr1 , E ijr2 , …, E ijrg , …, Eijrz}, H = {H 1 , …, H g , …, H z}, E ijrg is D ijr in H g 's probability value, H g is the g-th preset field.

[0046] S350. Based on the second use label weight and the list of field values corresponding to the second use label, determine the target field corresponding to the target field name.

[0047] S360. Obtain the field probability value of the first use label in the target field, and based on the field probability value of the first use label in the target field and the initial probability value corresponding to the first use label, determine the target label corresponding to the target field name.

[0048] In summary, obtain the data list of the target user, obtain the single field name corresponding to the field, input the target field name into the target large model, obtain the list of first use labels corresponding to the target field name and the initial probability value corresponding to each first use label of the target field name, combine the target field name with several other single field names except the target field name to obtain the combined field name list, and input the combined field name into the target large model to obtain the list of second use labels corresponding to the combined field name, obtain the quantity weight list, and based on the quantity weight list and the combined field name, obtain the weight corresponding to the second use label and mark it as the second use label weight, input the second use label into the target large model to obtain the list of field values corresponding to the second use label, based on the second use label weight and the list of field values corresponding to the second use label, determine the target field corresponding to the target field name, obtain the field probability value of the first use label in the target field, and based on the field probability value of the first use label in the target field and the initial probability value corresponding to the first use label, determine the target label corresponding to the target field name. The present invention determines all possible labels of the field name based on the large model and the field name, and determines the target label based on the field, accurately and comprehensively obtaining the data label.

[0049] Specifically, S200 further includes:

[0050] S210. Obtain the preset field feature list, and the preset field feature list includes several preset field features. The preset field features include the total length of the field, the number of digital characters included in the field, the number of English characters included in the field, etc.

[0051] S220. Obtain the corresponding list of preset field feature values, and the list of preset field feature values includes the preset field feature values corresponding to several preset field features.

[0052] S230. Determine the single field name of a field based on a preset list of field characteristic values.

[0053] In summary, obtain a preset list of field characteristics, obtain the corresponding preset list of field characteristic values, and determine the single field name of a field based on the preset list of field characteristic values. The present invention determines the single field name based on the preset list of field characteristic values, rather than directly obtaining the field name of the filled field, to ensure the accuracy of the field name.

[0054] Specifically, S320 includes:

[0055] S321. Obtain a list of single field names N = {N 1 , N 2 , …, N i , …, N m}, and obtain a list of other field names P = {P i} except for the target field name N 1 , P 2 , …, P p , …, P n-1}, where P p is the p-th other field name, and the value range of p is from 1 to n - 1.

[0056] S322. Take any number of other field names from the list of other field names P and combine them with the target field name N i to obtain a combined field name. The combined field name is obtained by the extraction method of combination numbers to ensure that all combination methods of the target field name N i are combined.

[0057] Specifically, the quantity weight list is obtained through the following steps in S330:

[0058] S331. Obtain a list of historical field quantities R = {R 1 , R 2 , …, R a , …, R b}, where R a = {R a1 , R a2 , …, R ac , …, R ad}, and R ac is the c-th sample of the a-th historical field quantity. The value range of c is from 1 to d, where d is the number of samples, and the value range of a is from 1 to b, where b is the number of historical field quantities; among them, the a-th historical field quantity includes a field names.

[0059] S332. Obtain a set of historical field list T = {T 1 , T 2 , …, T a, …, T b},the a-th historical field list T a＝ {T a1 , T a2 , …, T ac , …, T ad}, T ac is R ac the corresponding field.

[0060] S333. Based on the historical field quantity list R and the historical field list T, obtain the correlation coefficient list U = {U 1 , U 2 , …, U a , …, U b}, where the a-th correlation coefficient U a = [∑ d c＝1 [(R ac - ER a ) × (T ac - ET a )]] / [(∑ d c＝1 ((R ac - ER a )) 2 )] × ∑ 1 / 2 ×∑ d c＝1 ((T ac - ET a ) 2 )]] 1 / 2 , ER a = (∑ d c＝1 R ac ) / d, ET a = (∑ d c＝1 T ac ) / d.

[0061] S334. Normalize U 1 to U b to obtain the quantity weight list.

[0062] Specifically, those skilled in the art know that any method of normalization in the prior art belongs to the protection scope of the present invention and will not be elaborated here.

[0063] In summary, obtain the historical field quantity list, obtain the set of historical field lists, based on the historical field quantity list R and the historical field list T, obtain the correlation coefficient list, and normalize U 1 to U bPerform normalization processing to obtain a list of quantity weights, calculate weights through historical data, and more accurately obtain the weight relationship between the number of field names and the fields.

[0064] Further, in S360, determining the target field corresponding to the target field name based on the second usage label weight and the list of field values corresponding to the second usage label further includes:

[0065] S361, obtain a list of field integral values J i ={J i1 , J i2 ,…, J ig ,…, J iz}, the field integral value J ig =∑ n j＝1 ∑ s r＝1 (F ijr ×E ijrg ), where F ijr is the second usage label weight corresponding to D ijr , D ijr is the r-th second usage label corresponding to C ij , C ij is the j-th combined field name corresponding to N i , the value range of j is from 1 to n, and n is the number of combined field names corresponding to N i , the value range of r is from 1 to s, and s is the number of second usage labels, and the value range of g is from 1 to z, and z is the number of preset fields.

[0066] S362, sort J i1 to J iz in descending order, and obtain the top q preset fields as the target fields corresponding to the target field name N i . Specifically, sort J i1 to J iz in descending order to obtain the sorted list K i ={K i1 , K i2 ,…, K ig ,…, K iz}, and obtain the top q as the target fields TK i ={TK i1 , TK i2 ,…, TK ix ,…, TK iq}, TK ix is the x-th target field corresponding to N i , the value range of x is from 1 to q, and q is the number of target fields. Optionally, q = 1.

[0067] Further, S360 further includes:

[0068] S3601, obtaining a target value corresponding to a first usage label, where the target value is a weighted sum value of a domain probability value and an initial probability value.

[0069] S3602, obtaining the maximum target value and using the first usage label corresponding to the maximum target value as the target label.

[0070] In an embodiment of the present invention, in S3602, when calculating the target value, the weight of the domain probability value is equal to the weight of the initial probability value.

[0071] In summary, obtain the target value corresponding to the first usage label, obtain the maximum target value, and use the first usage label corresponding to the maximum target value as the target label.

[0072] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one segment of a program related to a method in the method embodiment. The at least one instruction or the at least one segment of the program is loaded and executed by the processor to implement the method provided in the above embodiment.

[0073] An embodiment of the present invention also provides an electronic device, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0074] Although some specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and not for limiting the scope of the present invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present invention.

Claims

1. A path determination method based on a large model, characterized in that: The method comprises the following steps: S001, obtain the user problem of the target user, input the user problem into the target large model, and obtain the implementation path list AA = {AA1, AA2, ..., AA e , …, AA f } and the realization probability list AB corresponding to AA = {AB1, AB2, ..., AB e , …, AB f }, A B e Yes AA e The realization probability value of the e-th realization path AA e ={AA e1 , A.A. e2 , …, AA eh , …, AA eu }, AA eh is the h-th implementation field of the e-th reality path and the implementation tag corresponding to the h-th implementation field, the value range of h is 1 to u, u is the number of implementation fields, the value range of e is 1 to f, f is the number of implementation paths; where AB e Greater than AB e+1 ; S002, traverse AA, use AA for AAe e1 To AA eu Matching with a target data list, wherein the target data list includes a plurality of target fields and a target label corresponding to each target field; S003, if AA e1 To AA eu The target data list is successfully matched, and AA e As the final path; S004, if the match fails, obtain the to-be-matched list AC = {AC1, AC2, ..., AC e ,…,AC f }, AA e Corresponding data to be matched AC e ={AC e1 , A.C. e2 ,…,AC ev ,…,AC ew }, AC ev Yes AA e The vth implementation field that was not successfully matched and the implementation tag corresponding to the vth implementation field that was not successfully matched, the value range of v is 1 to w, and the number of unmatched implementation fields w≤u; S005, for AC ev , get AC ev Related fields of AF ev , and obtain AC ev To AF ev The target weighted value AG ev , where AF ev and AF ev The matching result of the corresponding related tag and the target data list is a successful match; the target weighted value AG ev AC ev To AF ev The sum of the preset weights of the nodes that the implementation path passes through; S006, based on AG ev and AB e Determine the final path.

2. The path determination method based on a large model according to claim 1, characterized in that: S006 also includes: S061, Get target and value AG e0 =∑ w v＝1 AG ev , and based on the target and value AG e0 Get the final value AH e =β1×AG e0 +β2×AB e , β1 is the first probability factor, β2 is the second probability factor; S062, obtain AH0=max{AH1, AH2, ..., AH e , …, AH f }, obtain the implementation path AA0 corresponding to AH0; S063, AA0 and AC ev To AF ev The relevant path is taken as the final path.

3. The path determination method based on a large model according to claim 2, characterized in that: β1＝β2。 4. The path determination method based on a large model according to claim 1, characterized in that: S005 also includes: if AC is obtained ev Related fields of AF ev Failure, AA e Delete from the implementation path list AA.

5. The path determination method based on a large model according to claim 2, characterized in that: Using ant colony algorithm to obtain AC ev To AF ev related paths.

6. The path determination method based on a large model according to claim 2, characterized in that: Use the A-star algorithm to obtain AC ev To AF ev related paths.

7. The path determination method based on a large model according to claim 1, characterized in that: Get the target data list by following the steps below: S100, obtaining a data list of a target user, wherein the data list of the target user includes several fields; S200, obtaining a single field name corresponding to the field; S300, marking any single field name as a target field name, and obtaining a target label corresponding to the target field name; Wherein, S300 includes the following steps: S310, inputting the target field name into the target large model, and obtaining a first use label list corresponding to the target field name and an initial probability value corresponding to each first use label of the target field name, wherein the first use label list includes a plurality of first use labels; S320, combining the target field name with several other single field names except the target field name, obtaining a list of combined field names, and inputting the combined field names into the target large model to obtain a list of second purpose labels corresponding to the combined field names; the list of combined field names includes several combined field names, and the second purpose label list includes several second purpose labels; S330, obtaining a quantity weight list, and based on the quantity weight list and the combined field name, obtaining a weight corresponding to the second purpose label and marking it as a second purpose label weight; wherein the quantity weight list includes a plurality of quantity weights, and the quantity weight is a weight between the quantity of field names in the combined field name and the field; S340, inputting the second usage tag into the target large model, obtaining a domain value list corresponding to the second usage tag, wherein the domain value list includes probability values ​​of the second usage tag in a plurality of preset domains; S350, determining a target field corresponding to the target field name based on the second usage tag weight and the field value list corresponding to the second usage tag; S360, obtaining the domain probability value of the first usage tag in the target domain, and determining the target tag corresponding to the target field name based on the domain probability value of the first usage tag in the target domain and the initial probability value corresponding to the first usage tag, thereby obtaining a target data list.

8. The path determination method based on a large model according to claim 7, characterized in that: In S330, the quantity weight list is obtained through the following steps: S331, obtain the history field quantity list R = {R1, R2, ..., R a , …, R b }, R a = {R a1 , R a2 , …, R ac , …, R ad }, R ac is the cth sample of the ath historical field quantity, where c ranges from 1 to d, d is the number of samples, and a ranges from 1 to b, b is the number of historical fields; where the ath historical field quantity includes a field names; S332, obtain the historical domain list set T = {T1, T2, ..., T a , …, T b }, the ath history field list T a＝ {T a1 , T a2 , …, T ac , …, T ad }, T ac YesR ac Corresponding fields; S333, based on the historical field quantity list R and the historical domain list T, obtain the correlation coefficient list U = {U1, U2, ..., U a , …, U b }, where the ath correlation coefficient U a =[∑ d c＝1 [(R ac -ER a )×(T ac -ET a )]] / [(∑ d c＝1 ((R ac -ER a ) 2 )) 1 / 2 ×∑ d c＝1 ((T ac -ET a ) 2 )) 1 / 2 ], ER a =(∑ d c＝1 R ac ) / d,ET a =(∑ d c＝1 T ac ) / d; S334, for U1 to U b Perform normalization processing to obtain a list of quantity weights.

9. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the large model-based path determination method as described in any one of claims 1-8.

10. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 9.

Images