Block chain-based large model collaborative verification method and system
By introducing blockchain technology into large models, generating asymmetric encryption public-private key pairs and smart contracts, and combining hash algorithms with Merkle trees, the problems of large-scale model data sharing and training transparency are solved, and data security sharing, transparent training process and verifiable results are achieved, thereby improving the credibility and sustainability of the system.
Patent Information
- Application Number
- CN202510821224.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-19
- Publication Date
- 2025-10-17
AI Technical Summary
The data silo problem of large models makes it difficult to share data securely, the risk of data privacy leakage is high, the training process is opaque and the results are untraceable, and there is a lack of a full-process collaborative verification mechanism.
By generating asymmetric encryption public and private key pairs, using blockchain smart contracts to implement data interaction rules, combining hash algorithms and Merkle trees to ensure data integrity, using smart contracts to automatically execute rules and record operations throughout the entire life cycle, and introducing token incentive mechanisms to stimulate user participation.
It achieves secure data sharing, transparent training process and verifiable results, ensures end-to-end privacy protection, supports third-party audits, and improves ecological sustainability and anti-attack capabilities.
Smart Images
Figure CN120805196A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of artificial intelligence, in particular to a large model collaborative verification method and system based on a blockchain. BACKGROUND
[0002] With the rapid development of artificial intelligence technology, large models have shown great capabilities in various fields. However, the development of large models faces data islands, and it is difficult to safely share scattered data sources, which restricts the generalization ability of the model; data privacy leakage, centralized training easily exposes sensitive data; the model training process is not transparent, the training process is not transparent, and the results are not traceable. The decentralized, tamper-proof and traceable features of blockchain technology provide a potential solution to the above problems. In the prior art, the combination of blockchain and large models is mostly limited to data storage, and lacks a full-process collaborative verification mechanism.
[0003] In large model applications, how to achieve data security sharing, training process transparency and result verifiability based on blockchain technology is a technical problem that needs to be solved. SUMMARY
[0004] The technical task of the present application is to solve the technical problems of how to achieve data security sharing, training process transparency and result verifiability based on blockchain technology by providing a large model collaborative verification method and system based on a blockchain.
[0005] In a first aspect, the present application provides a large model collaborative verification method based on a blockchain, comprising the following steps:
[0006] Key pair generation: the third-party auditing agency and each user respectively as an applicant submit an identity verification request to the identity authentication center, the identity authentication center verifies the identity of the applicant based on the identity registration information in the identity verification request, and returns a pair of asymmetric encryption public and private key pairs to the applicant who passes the identity verification, and the third-party auditing agency shares its public key to each user;
[0007] Smart contract generation: the user calls the large model to generate a blockchain smart contract, and the blockchain smart contract is used to generate a data interaction rule adapted to the blockchain platform;
[0008] Model training: the user trains, tests and verifies the large model based on the sample data collected by the user, evaluates the sample data and the large model after training based on the model verification result, and obtains a sample data evaluation result and a model evaluation result;
[0009] User evaluation: the user will sample data and the corresponding sample data evaluation results and model evaluation results as plaintext, encrypt the plaintext through the public key of the third-party audit institution, generate evaluation ciphertext, send the evaluation ciphertext to the third-party audit institution, and the third-party audit institution decrypts the evaluation ciphertext based on the private key to obtain the plaintext, audits the user based on the plaintext and calls the blockchain smart contract, and allocates reward coins to the user who passes the audit. The user who passes the audit uploads the reward coins to his blockchain account, encrypts the plaintext through his public key, generates result ciphertext, uploads the hash value of the result ciphertext to the blockchain;
[0010] Audit query: when the user audits the sample data, the user queries the corresponding result ciphertext hash value from the blockchain based on the blockchain smart contract, and decrypts and views the related result ciphertext based on the private key;
[0011] Model calling: for a large model after training, the user pays reward coins to purchase the use permission of the related large model based on the blockchain smart contract.
[0012] As a preferred, when the smart contract is generated, the user writes Prompt input large model, generates the corresponding user demand blockchain smart contract through the large model, and the blockchain smart contract meets the writing specification.
[0013] As a preferred, for the user who passes the audit, the user uploads the hash value of the result ciphertext and the digital signature to the blockchain, and the hash value of the result ciphertext, the digital signature and the timestamp are used as user information, and the user information is stored as node information to the constructed Merkle tree.
[0014] As a preferred, when the third-party audit institution audits the user based on the plaintext and calls the blockchain smart contract, the user is audited based on the data volume of the sample data, the sample data evaluation results and the model evaluation results, the user is scored based on the audit results, and the reward coins are allocated to the user based on the score value.
[0015] In a second aspect, the present application is a large model collaborative verification system based on a blockchain, which comprises a user module, a large model module, a third-party audit institution, an identity authentication center and a blockchain platform.
[0016] The user module is used to support the user as an applicant to submit an identity authentication request to the identity authentication center, and the third-party audit institution is used as an applicant to submit an identity authentication request to the identity authentication center. Correspondingly, the identity authentication center authenticates the applicant based on the identity registration information in the identity authentication request, and returns a pair of asymmetric encryption public and private key pairs to the applicant who passes the identity authentication. The third-party audit institution shares its public key to each user;
[0017] The large model module is configured with multiple multi-modal large models;
[0018] The user module is configured to support a user to call a large model to generate a blockchain smart contract, and support the user to perform model training, model testing and model verification on a large model configured in the large model module based on sample data collected by the user, evaluate the sample data and the large model after training based on a model verification result, obtain a sample data evaluation result and a model evaluation result, and support the user to encrypt the sample data and the corresponding sample data evaluation result and model evaluation result as plaintext through a public key of a third-party auditing agency to generate evaluation ciphertext, and send the evaluation ciphertext to the third-party auditing agency. The blockchain smart contract is configured to generate a data interaction rule adapted to a blockchain platform. Correspondingly, the third-party auditing agency is configured to decrypt the evaluation ciphertext based on a private key to obtain the plaintext, and perform auditing on the user based on the plaintext and the blockchain smart contract, and allocate a reward coin to the user who passes the auditing.
[0019] Correspondingly, the user module is configured to support the user who passes the auditing to upload the reward coin to a blockchain account of the user, and encrypt the plaintext through a public key of the user to generate result ciphertext, and upload a hash value of the result ciphertext to the blockchain.
[0020] The user module supports the user to perform auditing query on the sample data of the user. When the auditing query is performed, the hash value of the corresponding result ciphertext is queried from the blockchain based on the blockchain smart contract, and the relevant result ciphertext is decrypted and viewed based on a private key.
[0021] For the large model after training, the user module supports the user to purchase a use right of the relevant large model based on the blockchain smart contract and payment of the reward coin.
[0022] Preferably, the user module is configured to support the user to write a Prompt input large model, and generate a blockchain smart contract corresponding to a demand of the user through the large model. The blockchain smart contract meets a writing specification.
[0023] Preferably, for the user who passes the auditing, the user module supports the user to upload the hash value of the result ciphertext and a digital signature to the blockchain, and store the hash value of the result ciphertext, the digital signature and a time stamp as user information to a constructed Merkle tree as node information.
[0024] Preferably, when the third-party auditing agency performs auditing on the user based on the plaintext and the blockchain smart contract, the user is audited based on a data volume of the sample data, a sample data evaluation result and a model evaluation result, a score is given to the user based on an auditing result, and a reward coin is allocated to the user based on the score value.
[0025] The large model collaborative verification method and system based on the blockchain have the following advantages:
[0026] 1. Data credibility: ensure data integrity through hash algorithm and Merkel tree, and realize end-to-end privacy protection by combining asymmetric encryption;
[0027] 2. Process transparency: smart contract automatically executes rules, and blockchain records the whole life cycle operation to support third-party audit;
[0028] 3. Ecological sustainability: token incentive mechanism stimulates user participation and promotes data sharing and model iteration optimization;
[0029] 4. Anti-attack ability: distributed storage and consensus mechanism effectively resist single point failure and malicious tampering. BRIEF DESCRIPTION OF DRAWINGS
[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.
[0031] The present application will be further described below in conjunction with the drawings.
[0032] Figure 1 A flowchart of a large model collaborative verification method based on blockchain for embodiment 1. DETAILED DESCRIPTION
[0033] The present application will be further described below in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it. However, the embodiments are not limiting to the present application, and the technical features in the embodiments and the embodiments can be combined with each other without conflict.
[0034] The present application provides a large model collaborative verification method and system based on blockchain, which is used to solve the technical problem of how to realize data security sharing, training process transparency and result verifiability based on blockchain technology.
[0035] Embodiment 1:
[0036] The present application provides a large model collaborative verification method based on blockchain, which includes six steps of key pair generation, smart contract generation, model training, user evaluation, audit query and model calling.
[0037] Step S100 key pair generation: the third-party auditing agency and each user respectively as the applicant submit an identity verification request to the identity authentication center, the identity authentication center verifies the identity of the applicant based on the identity registration information in the identity verification request, and returns a pair of asymmetric encryption public and private key to the applicant who passes the identity verification, the third-party auditing agency shares its public key to each user.
[0038] Step S200 smart contract generation: the user calls the large model to generate a blockchain smart contract, and the blockchain smart contract is used to generate a data interaction rule adapted to the blockchain platform.
[0039] When the smart contract is generated, the user writes a Prompt input large model, and generates a blockchain smart contract corresponding to the user's demand through the large model, which meets the writing specification.
[0040] Step S300 model training: the user trains, tests and verifies the large model based on the sample data collected by the user, evaluates the sample data and the large model after training based on the model verification result, and obtains the sample data evaluation result and the model evaluation result.
[0041] Step S400 user evaluation: the user encrypts the sample data and the corresponding sample data evaluation result and model evaluation result as plaintext through the public key of the third-party auditing agency, generates evaluation ciphertext, sends the evaluation ciphertext to the third-party auditing agency, the third-party auditing agency decrypts the evaluation ciphertext based on its private key to obtain the plaintext, audits the user based on the plaintext and calls the blockchain smart contract, and allocates reward coins to the user who passes the audit, the user who passes the audit uploads the reward coins to the blockchain account, and encrypts the plaintext through the public key to generate result ciphertext, and uploads the hash value of the result ciphertext to the blockchain.
[0042] Among them, considering that there are more hash values uploaded to the blockchain, for the user who passes the audit, the user uploads the hash value of the result ciphertext and the digital signature to the blockchain, and the hash value of the result ciphertext, the digital signature and the timestamp are used as user information, and the user information is stored as node information to the constructed Merkle tree, and when one node is changed, all node data will be changed.
[0043] When the third-party auditing agency audits the user based on the plaintext and calls the blockchain smart contract, the user is audited based on the data volume of the sample data, the sample data evaluation result and the model evaluation result, the user is scored based on the audit result, the reward coins are allocated to the user based on the score, the reward coins can be used to purchase Tokens of the large model after training, and can be used to call the mature large model.
[0044] Step S500 audits the query: when the user audits the query on the sample data, the corresponding result ciphertext hash value is queried from the blockchain based on the blockchain smart contract, and the relevant result ciphertext is decrypted and viewed based on the private key to obtain the original plaintext.
[0045] Step S600 model calling: for the large model after training, the user pays the reward currency to purchase the use permission of the relevant large model based on the blockchain smart contract.
[0046] The method of the embodiment realizes data security sharing, training process transparency and result auditability, improves user participation through a token incentive mechanism, and forms a sustainable model optimization ecology.
[0047] Embodiment 2:
[0048] A large model collaborative verification system based on a blockchain includes a user module, a large model module, a third-party auditing agency, an identity authentication center, and a blockchain platform.
[0049] The user module is configured to support the user as an applicant to submit an identity verification request to the identity authentication center, and the third-party auditing agency is configured to submit an identity authentication request to the identity authentication center as an applicant. Correspondingly, the identity authentication center performs identity verification on the applicant based on the identity registration information in the identity verification request, and returns a pair of asymmetrically encrypted public and private key pairs to the applicant who passes the identity verification. The third-party auditing agency shares its public key to each user.
[0050] The large model module is configured with multiple multi-modal large models.
[0051] The user module is configured to support the user to call the large model to generate a blockchain smart contract, and support the user to perform model training, model testing, and model verification on the large models configured in the large model module based on the sample data collected by the user. Based on the model verification result, the sample data and the large model after training are evaluated to obtain the sample data evaluation result and the model evaluation result, and the user is supported to encrypt the plaintext through the public key of the third-party auditing agency to generate an evaluation ciphertext, send the evaluation ciphertext to the third-party auditing agency, and the blockchain smart contract is configured to generate a data interaction rule adapted to the blockchain platform. Correspondingly, the third-party auditing agency is configured to decrypt the evaluation ciphertext based on the private key to obtain the plaintext, and based on the plaintext, call the blockchain smart contract to audit the user, and allocate reward currency to the user who passes the audit.
[0052] Correspondingly, the user module is configured to support the user who passes the audit to upload the reward currency to the blockchain account, and encrypt the plaintext through the public key to generate a result ciphertext, and upload the hash value of the result ciphertext to the blockchain.
[0053] The user module supports the user to audit and query the sample data of the user, and when the audit and query is performed, the hash value of the corresponding result ciphertext is queried from the blockchain based on the smart contract of the blockchain, and the relevant result ciphertext is decrypted and viewed based on the private key.
[0054] For the large model after training, the user module supports the user to purchase the use permission of the relevant large model based on the reward currency paid by the user based on the smart contract of the blockchain.
[0055] As a specific implementation of the user module, when the smart contract is generated, the user writes the Prompt input large model, and the smart contract corresponding to the user demand is generated through the large model, and the smart contract of the blockchain conforms to the writing specification.
[0056] Considering that the hash value uploaded to the blockchain is more, for the user who passes the audit, the user module supports the user to upload the hash value and the digital signature of the result ciphertext to the blockchain, and the hash value, the digital signature and the timestamp of the result ciphertext are used as the user information, and the user information is stored as the node information to the constructed Merkle tree, wherein when one node is changed, all the node data will be changed.
[0057] As a specific implementation of the third-party audit agency, when the user is audited based on the plaintext and the smart contract of the blockchain, the user is audited based on the data volume of the sample data, the evaluation result of the sample data and the model evaluation result, the user is scored based on the audit result, and the reward currency is allocated to the user based on the score value, the reward currency can be used to purchase the Tokens of the large model after the training is successful, and can be used for the calling of the mature large model.
[0058] The system of the embodiment can execute the method disclosed in embodiment 1 to realize the data collaborative verification of the large model.
[0059] The above describes the method and system for collaborative verification of the large model based on the blockchain provided by the application in detail, and the principle and implementation mode of the application are described by applying specific examples; the above embodiment is only used to help understand the method and the core idea of the application; meanwhile, for the general technical personnel in the art, according to the idea of the application, the specific implementation mode and the application range will be changed, and the above description should not be understood as the limitation of the application.
Claims
1. A large-scale model collaborative verification method based on blockchain, characterized in that: The steps include: Key pair generation: The third-party auditing agency and each user, acting as applicants, submit identity authentication requests to the identity authentication center. The identity authentication center authenticates the applicant based on the identity registration information in the authentication request and returns an asymmetrically encrypted public-private key pair to the authenticated applicant. The third-party auditing agency then shares its public key with each user. Smart contract generation: Users call the big model to generate a blockchain smart contract, which is used to generate data interaction rules that are compatible with the blockchain platform; Model training: Users train, test, and verify the large model based on the sample data they collect. Based on the model verification results, they evaluate the sample data and the trained large model to obtain sample data evaluation results and model evaluation results. User evaluation: The user takes the sample data and the corresponding sample data evaluation results and model evaluation results as plaintext, encrypts the plaintext with the public key of the third-party review agency to generate the evaluation ciphertext, and sends the evaluation ciphertext to the third-party review agency. The third-party review agency decrypts the evaluation ciphertext with its private key to obtain the plaintext, and then calls the blockchain smart contract to review the user based on the plaintext. The user who passes the review is allocated reward coins, and the user who passes the review uploads the reward coins to his blockchain account, encrypts the plaintext with his public key to generate the result ciphertext, and uploads the hash value of the result ciphertext to the blockchain; Audit query: When a user conducts an audit query on their sample data, the hash value of the corresponding result ciphertext is retrieved from the blockchain based on the blockchain smart contract, and the relevant result ciphertext is decrypted and viewed based on their private key; Model call: For trained large models, users pay reward coins based on blockchain smart contracts to purchase the use rights of the relevant large models.
2. The blockchain-based large-scale model collaborative verification method according to claim 1 is characterized in that: When generating a smart contract, the user writes a prompt to input a large model, and the large model generates a blockchain smart contract corresponding to the user's needs. The blockchain smart contract complies with the writing specifications.
3. The blockchain-based large-scale model collaborative verification method according to claim 1 is characterized in that: For users who pass the review, the user uploads the hash value and digital signature of the result ciphertext to the blockchain, uses the hash value, digital signature and timestamp of the result ciphertext as user information, and stores the user information as node information in the constructed Merkle tree.
4. The blockchain-based large-scale model collaborative verification method according to claim 1 is characterized in that: When a third-party auditing agency audits a user based on plain text and calling a blockchain smart contract, it audits the user based on the amount of sample data, the sample data evaluation results, and the model evaluation results, scores the user based on the audit results, and allocates reward coins to the user based on the score.
5. A large-scale model collaborative verification system based on blockchain, characterized in that: Including user module, large model module, third-party audit agency, identity authentication center and blockchain platform; The user module is used to support users as applicants to submit identity authentication requests to the identity authentication center. The third-party audit agency is used to submit identity authentication requests to the identity authentication center as applicants. Correspondingly, the identity authentication center authenticates the applicant based on the identity registration information in the identity authentication request and returns a pair of asymmetric encrypted public and private key pairs to the applicant who passes the authentication. The third-party audit agency shares its public key with each user. The large model module is configured with multiple multimodal large models; The user module is used to support users in calling the big model to generate a blockchain smart contract, and supports users in performing model training, model testing and model verification on the big model configured in the big model module based on the sample data they collected, and evaluating the sample data and the trained big model based on the model verification results to obtain sample data evaluation results and model evaluation results, and supports users in using the sample data and the corresponding sample data evaluation results and model evaluation results as plain text, encrypting the plain text with the public key of a third-party audit agency, generating evaluation ciphertext, and sending the evaluation ciphertext to the third-party audit agency. The blockchain smart contract is used to generate data interaction rules adapted to the blockchain platform; correspondingly, the third-party audit agency is used to decrypt the evaluation ciphertext based on its private key to obtain plain text, audit the user based on the plain text and calling the blockchain smart contract, and allocate reward coins to users who pass the audit; Correspondingly, the user module is used to support users who have passed the review to upload their reward coins to their blockchain accounts, encrypt the plaintext with their public key, generate the resulting ciphertext, and upload the hash value of the resulting ciphertext to the blockchain; The user module supports users to conduct audit queries on their sample data. During the audit query, the hash value of the corresponding result ciphertext is queried from the blockchain based on the blockchain smart contract, and the relevant result ciphertext is decrypted and viewed based on the user's private key; For trained large models, the user module supports users to pay reward coins based on blockchain smart contracts to purchase the use rights of related large models.
6. The blockchain-based large-scale model collaborative verification system according to claim 5 is characterized in that: The user module is used to support users in writing prompt input large models, and generate blockchain smart contracts corresponding to user needs through the large models. The blockchain smart contracts comply with the writing specifications.
7. The blockchain-based large-scale model collaborative verification system according to claim 5 is characterized in that: For users who have passed the review, the user module supports users to upload the hash value and digital signature of the result ciphertext to the blockchain, use the hash value, digital signature and timestamp of the result ciphertext as user information, and store the user information as node information in the constructed Merkle tree.
8. The blockchain-based large-scale model collaborative verification system according to claim 5 is characterized in that: When a third-party auditing agency audits a user based on plain text and calling a blockchain smart contract, it audits the user based on the amount of sample data, the sample data evaluation results, and the model evaluation results, scores the user based on the audit results, and allocates reward coins to the user based on the score.