LoRA fine tuning-based achievement transformation large model task decomposition method and system

Abstract

The invention belongs to the technical field of artificial intelligence task decomposition, and particularly relates to an achievement transformation large model task decomposition method and system based on LoRA fine tuning. According to the method, a LoRA (Low-RankAdaptation) technology is utilized to carry out efficient parameter fine tuning on a large achievement transformation model, and the method comprises four modules of achievement transformation task data set construction, task step decomposition, relation prediction and joint optimization training. A task data set is constructed by collecting multi-step tasks such as technology evaluation and market analysis and marking dependency relationships among steps. Fine tuning is performed on the result transformation large model based on the LoRA technology, and step decomposition loss and relation prediction loss are calculated respectively to realize task step decomposition and inter-step relation prediction; and through joint optimization of the two types of loss functions, the model can generate a reasonable task execution sequence. Through LoRA fine tuning, the method can quickly adapt to a new task under a small amount of data, reduces training time and resource consumption, improves the accuracy and flexibility of task decomposition, and is suitable for various achievement transformation scenes.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence task decomposition technology, and in particular relates to a method and system for task decomposition of a large-scale achievement transformation model based on LoRA fine-tuning. Background Technology

[0002] With the rapid development of artificial intelligence technology, large language models have shown great potential in the field of technology transfer, especially in task decomposition. The technology transfer process involves multiple stages, including technology evaluation, market analysis, patent strategy, and production planning. Each stage contains multiple execution steps with complex dependencies between them. Reasonable task decomposition can improve the efficiency of technology transfer, reduce implementation difficulty, and increase the success rate. Therefore, improving the task decomposition capabilities of large language models in technology transfer scenarios has become a key issue.

[0003] Currently, there are four main technical approaches to decomposing technology transfer tasks: 1) methods based on manual rules, which decompose technology transfer tasks using predefined templates and rules; 2) methods based on search data, which utilize user query behavior during the technology transfer process to uncover task steps; 3) methods based on crowdsourcing, which construct technology transfer task decomposition datasets through manual annotation; and 4) methods based on large language models, which utilize task decomposition knowledge contained in pre-trained models. Among these, methods based on large language models have received widespread attention due to their powerful knowledge acquisition and reasoning capabilities, and have demonstrated unique advantages in the complex scenario of technology transfer.

[0004] However, the inventors discovered that existing methods for decomposing commercialization tasks based on large language models have the following problems: 1) Traditional fine-tuning methods require updating all model parameters, resulting in huge computational resource consumption when adapting to commercialization tasks; 2) Full parameter fine-tuning leads to an excessively large parameter space, resulting in low training efficiency for different types of commercialization tasks; 3) There is a lack of targeted parameter adjustment strategies, making it difficult to accurately grasp the temporal dependencies between sub-tasks in the commercialization process; 4) Model fine-tuning is costly and difficult to quickly adapt to the decomposition needs of commercialization tasks in different technical fields. These problems are particularly prominent in commercialization scenarios involving multiple technical fields. Therefore, there is an urgent need for a fine-tuning method that is parameter-efficient, fast-training, and can accurately identify the temporal relationships of tasks to improve the application effect of large models in commercialization task decomposition. Summary of the Invention

[0005] To address at least one of the technical problems mentioned above, this invention provides a method and system for task decomposition of a large-scale model for technology transfer based on LoRA fine-tuning. Through parameter-efficient low-rank adaptation technology, it achieves characteristics such as low computational resource consumption, high training efficiency, accurate prediction of temporal relationships, and strong domain adaptability, making it more suitable for practical application scenarios such as technology transfer.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Construction of a dataset for technology transfer tasks. This invention first constructs a high-quality dataset for decomposing technology transfer tasks. The dataset construction process includes:

[0008] (1) Collect task samples. Collect representative multi-step task descriptions from typical achievement transformation scenarios such as technology assessment, market analysis, patent layout, and production planning as training samples.

[0009] (2) Step decomposition and annotation: each technology transfer task is broken down into detailed steps by domain experts. The decomposition process should follow the following principles: a) Each step should be specific and executable, such as patent search and technology route evaluation; b) The granularity of the steps should be appropriate, neither too general nor too fragmented; c) The step descriptions should be standardized and professional, using unified technical terminology; d) Ensure the completeness of the set of steps, covering the entire technology transfer process.

[0010] (3) Temporal relationship labeling: Label the execution dependencies of the decomposed set of steps. Dependencies are divided into the following categories: a) Strict sequential relationship: such as technology evaluation must be completed before business planning; b) Parallel relationship: such as market research and patent search can be carried out simultaneously; c) Optional sequential relationship: the order of some preparatory work can be flexibly adjusted. A complete technology transfer execution process is constructed through these relationships.

[0011] Task step decomposition module. Using a large-scale results transformation model as a foundation, and employing LoRA for efficient parameter fine-tuning, this module decomposes the task into a sequence of specific executable steps. Training configuration includes:

[0012] (1) Structure of the pre-trained model: The pre-trained large language model structure Llama-2-7B used in this invention is used as the large model for the transformation of results, but it can also be any other structure model.

[0013] (2) Fine-tuning configuration: LoRA fine-tuning is performed on the Query and Value matrices of the Attention layer, with the rank value set to 8 and the alpha value set to 16.

[0014] (3) Training strategy: The AdamW optimizer is used with a learning rate of 3e-4, and the learning rate is dynamically adjusted using a validation set.

[0015] (4) Optimization objective: Minimize the cross-entropy loss between the predicted step sequence and the actual step sequence.

[0016] Task step decomposition module. Using a large-scale results transformation model as a foundation, and employing LoRA for efficient parameter fine-tuning, this module decomposes the task into a sequence of specific executable steps. Training configuration includes: Input format:

[0017] (1) Structure of the pre-trained model: the same pre-trained large language model in the sampling and task step decomposition module.

[0018] (2) Input format: Take task steps as input and predict the type of dependency between them.

[0019] (3) Fine-tuning strategy: Use the same LoRA configuration as the step decomposer, but optimize the loss function specifically.

[0020] (4) Optimization objective: Minimize the binary cross-entropy loss between predicted dependencies and actual dependencies.

[0021] Joint optimization training. The loss functions for step decomposition and relation prediction are jointly optimized, and a unified optimization objective is designed through weighted combination. The same set of LoRA parameters is used to implement an end-to-end model training and inference process, and the overall performance is evaluated through a validation set and the parameters are dynamically adjusted.

[0022] A second aspect of the present invention provides a task decomposition system for a large-scale model of achievement transformation based on LoRA fine-tuning, comprising:

[0023] The technology transfer task dataset construction module is used to build a task decomposition training dataset for technology transfer scenarios. This dataset includes multi-step task descriptions such as technology evaluation, market analysis, and patent layout, along with corresponding step sequences and execution dependency annotations. The task step decomposition module decomposes the transfer task into specific executable steps using a LoRA-tuned large-scale technology transfer model and calculates the step decomposition loss. The relationship prediction module predicts the execution dependencies between steps based on the same LoRA-tuned large-scale technology transfer model and calculates the relationship prediction loss. The joint optimization module uniformly calculates the weighted combination of step decomposition loss and relationship prediction loss, efficiently fine-tuning the large-scale technology transfer model using a single set of low-rank parameters to improve the model's task decomposition capabilities in different technological fields.

[0024] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the LoRA-based large-scale model task decomposition method for achievement transformation as described above.

[0025] A fourth aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the LoRA-based fine-tuning large-scale model task decomposition method for achievement transformation as described above.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] (1) This invention uses LoRA technology to efficiently fine-tune the parameters of the large model for achievement transformation. Through low-rank matrix decomposition and adaptive update, it significantly reduces the consumption of computing resources and improves training efficiency.

[0028] (2) The present invention decomposes the task into a joint task of step generation and relation prediction, and improves the accuracy of decomposition and the prediction quality of temporal relations through unified LoRA parameter optimization.

[0029] (3) The present invention has good domain adaptability and can handle the task decomposition requirements of different domains, including the transformation of scientific and technological achievements, without the need to retrain the complete model for new domains.

[0030] (4) The method of the present invention is easy to deploy and expand, and can be quickly applied to practical scenarios such as technology transfer and project management, providing strong support for collaborative innovation between industry, academia and research.

[0031] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0032] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0033] Figure 1 This is a flowchart of the task decomposition method for a large-scale achievement transformation model based on LoRA fine-tuning according to an embodiment of the present invention;

[0034] Figure 2 This is a flowchart illustrating the construction of the dataset for the achievement transformation task in an embodiment of the present invention;

[0035] Figure 3 This is a flowchart of the task step decomposer training according to an embodiment of the present invention;

[0036] Figure 4 This is a flowchart of the training steps of the task step relationship predictor according to an embodiment of the present invention;

[0037] Figure 5This is a flowchart of the joint optimization training process based on low-rank parameters according to an embodiment of the present invention; Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0040] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] Reference Figure 1 This embodiment provides a task decomposition method for a large-scale achievement transformation model based on LoRA fine-tuning, which includes:

[0042] S101: Construction of the dataset for technology transfer tasks. Collect transfer task samples from typical scenarios such as technology evaluation, market analysis, patent layout, and production planning. Domain experts decompose the steps and annotate the execution dependencies to construct the training dataset.

[0043] S102: Task step decomposition training. Based on the large-scale achievement transformation model, LoRA fine-tuning is performed by designing a low-rank parameter matrix to decompose the complex transformation task into a specific executable sequence of steps, and the step decomposition loss is calculated.

[0044] S103: Step Relationship Prediction Training. Based on the same large-scale model for outcome transformation and shared LoRA parameters, the execution dependency relationship between step pairs is predicted, and the relationship prediction loss is calculated.

[0045] S104: Joint optimization training combines the step decomposition loss and relation prediction loss in a weighted manner, optimizes the objective function through a unified LoRA parameter, and outputs a sequence of task steps with a reasonable execution order.

[0046] Reference Figure 2 The flowchart illustrates the steps involved in constructing a dataset for technology transfer tasks, providing a detailed explanation of the dataset construction process.

[0047] Step 200: Collection of Samples for Technology Transfer Tasks. Collect descriptions of typical tasks from the field of technology transfer to ensure the diversity and representativeness of the sample. Step 200 can be further detailed as follows:

[0048] Step 200-1: Determine the scope of task collection, including typical scenarios such as technology assessment, market analysis, patent layout, and production planning;

[0049] Step 200-2: Establish task complexity standards to ensure that the task includes multiple key transformation steps;

[0050] Step 200-3: Set the target number of samples, N=1000, to ensure that the dataset size meets the training requirements;

[0051] Step 200-4: Standardize the task description and unify technical terminology and format standards.

[0052] Step 201: Decomposition and Annotation of Transformation Steps. Domain experts will decompose and annotate each achievement transformation task into its steps. Step 201 can be further detailed as follows:

[0053] Step 201-1: Develop professional step-by-step standards to ensure that each step is specific and executable;

[0054] Step 201-2: Break down each conversion task into stages;

[0055] Step 201-3: Ensure the professionalism and completeness of the step description;

[0056] Step 201-4: Conduct expert review and correction of the decomposition results;

[0057] Step 201-5: Mark the technical connections between steps.

[0058] Step 202: Perform dependency annotation. Determine the execution dependencies between transformation steps. Step 202 can be further refined as follows:

[0059] Step 202-1: Define the dependency type (e.g., technology assessment must precede business planning);

[0060] Step 202-2: Mark the execution order between the steps;

[0061] Step 202-3: Verify the feasibility of the dependencies;

[0062] Step 202-4: Establish a complete sequence of steps to be executed.

[0063] Reference Figure 3 The task steps decomposer training flowchart provides a detailed explanation of the decomposer training process:

[0064] Step 300: Load the task step decomposition training dataset D_train, set a random seed to randomly sample from the training set, and train in batches;

[0065] Step 301: Base Model Configuration. Select the large-scale technology transfer model as the base and configure the LoRA parameters. Step 301 can be further refined as follows:

[0066] Step 301-1: Select the base model M as Llama-2-7B, and input the dimension d = 1024;

[0067] Step 301-2: Sample the random seed set in step 300, initialize the parameters of the LoRA weight matrix W∈Rd×r according to the random seed, and freeze the remaining parameters of the pre-trained large language model, keeping them unchanged;

[0068] Step 301-3: Set the LoRA parameters (r = 8, α = 16);

[0069] Step 301-4: Determine the set of fine-tuning layers L = {Attn_Q, Attn_V};

[0070] Step 302: Training parameter configuration. Set the training-related hyperparameters and optimization strategies. Step 302 can be further refined as follows:

[0071] Step 302-1: Configure the AdamW optimizer with a learning rate η = 3e-4;

[0072] Step 302-2: Set up the cosine annealing scheduler with a period T = 3;

[0073] Step 302-3: Configure the training batch size B = 8;

[0074] Step 302-4: Set the gradient accumulation step number k = 4.

[0075] Step 303: Calculate the cross-entropy loss L(θ) of the step decomposition.

[0076] Reference Figure 4 The flowchart for training the relation predictor provides a detailed explanation of the training process.

[0077] Step 400: Load the task relationship prediction training set D_train corresponding to the task step decomposition training dataset, reuse the random seed from step 300 to randomly sample from the training set, and train in batches;

[0078] Step 401: Prediction Module Configuration. Using the same outcome transformation model and LoRA parameters as the step decomposition, configure the step for the input format.<s1,s2> Step 401 can be further refined as follows:

[0079] Step 401-2: Share the LoRA weight matrix W∈Rd×r of the step decomposition module, and keep the pre-trained model parameters frozen;

[0080] Step 401-3: Define the set of relationship types R = {seq, par, opt};

[0081] Step 401-4: Reuse the LoRA parameter configuration (r = 8, α = 16);

[0082] Step 401-5: Share the set of fine-tuning layers L = {Attn_Q, Attn_V};

[0083] Step 402: Training parameter configuration. Reuse the training strategy of the step decomposition module. The step 402 can be refined as follows:

[0084] Step 402-1: Reuse the AdamW optimizer, learning rate η = 3e-4;

[0085] Step 402-2: Reuse the cosine annealing scheduler, period T = 3;

[0086] Step 402-3: Keep the training batch size B = 8;

[0087] Step 402-4: Keep the gradient accumulation step k = 4;

[0088] Step 403: Calculate the multi-class cross-entropy loss L_rel(θ) for relationship prediction;

[0089] Refer to Figure 5 , and a detailed description of the joint optimization training process is given, which includes:

[0090] Step 500: Load the training dataset D_train, randomly sample from the training set using the previous random seed, and train in batches;

[0091] Step 501: Joint optimization configuration. Based on the same set of LoRA parameters, jointly optimize the loss functions of step decomposition and relationship prediction. The step 501 can be refined as follows:

[0092] Step 501-1: Define the joint loss function L_total = λ1*L(θ) + λ2*L_rel(θ), where λ1 and λ2 are weight coefficients, set to 0.6 and 0.4 respectively;

[0093] Step 501-1: Reuse the previous large model configuration for achievement transformation;

[0094] Step 501-2: Reuse the previous LoRA parameters and training configuration.

[0095] Step 502: Joint Training and Optimization. Perform joint optimization training based on a single LoRA parameter. Step 502 can be further refined as follows:

[0096] Step 502-1: Calculate the joint loss L_total of the weighted combination;

[0097] Step 502-2: Update the shared LoRA parameters via backpropagation;

[0098] Step 502-3: Periodically save model checkpoints to prevent training interruptions;

[0099] Step 502-4: Verify the overall performance of the model on both the step decomposition and relation prediction tasks.

[0100] Step 503: Determine if the number of training iterations has reached the threshold or if the joint loss has stabilized. If so, end the optimization training. Otherwise, continue optimizing the LoRA parameters.

[0101] Example 2

[0102] This embodiment provides a task decomposition system for a large-scale model of achievement transformation based on LoRA fine-tuning, which includes:

[0103] The technology transfer task dataset construction module is used to collect transfer task samples in typical scenarios such as technology evaluation, market analysis, patent layout, and production planning. Domain experts decompose the steps and annotate the execution dependencies to build a standardized training dataset.

[0104] Task step decomposition module: Based on the large-scale achievement transformation model, the LoRA technology is used to efficiently fine-tune the parameters and decompose the task into a specific sequence of steps; the optimization objective of the step decomposition is to minimize the cross-entropy loss L_task(θ).

[0105] Relationship prediction module: Based on the same large-scale model of achievement transformation and LoRA parameters, it predicts the temporal dependencies between steps; the optimization objective of relationship prediction is to minimize the cross-entropy loss L_rel(θ) of relationship classification.

[0106] Joint optimization training module: Through a single set of LoRA parameters, the module jointly optimizes step decomposition and relation prediction, and outputs task steps with a reasonable execution order; where the objective of joint optimization is to minimize the weighted loss function L_total=λ1*L_task(θ)+λ2*L_rel(θ).

[0107] Example 3

[0108] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the method for task decomposition of the achievement transformation large model based on LoRA fine-tuning as described above.

[0109] Embodiment 4

[0110] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for task decomposition of the achievement transformation large model based on LoRA fine-tuning as described above.

[0111] Practitioners in the technical field should recognize that the embodiments of the present invention can take the form of a method, a system, or a computer program product. Therefore, the present invention can be implemented by hardware, software, or a combination of hardware and software. In addition, the present invention can take the form of a computer program product on a computer-usable storage medium (including but not limited to disk storage and optical storage, etc.), which contains one or more computer-usable program codes.

[0112] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A task decomposition technique for a large-scale achievement transformation model based on LoRA fine-tuning, characterized in that, The system comprises dataset construction, task step decomposition, relation prediction, and joint optimization training. Dataset construction involves collecting samples of technology transfer tasks and performing step decomposition and execution dependency annotation. Task step decomposition uses LoRA technology to efficiently fine-tune the parameters of the large-scale technology transfer model, breaking down the transfer task into a sequence of specific steps. Relationship prediction uses shared LoRA parameters to predict execution dependencies between steps. Joint optimization training optimizes the same set of low-rank parameters, outputting a sequence of transfer steps with a reasonable execution order.

2. A process for constructing a dataset using a task decomposition technique for large-scale achievement transformation models based on LoRA fine-tuning, characterized in that... Task samples were collected from typical technology transfer scenarios such as technology assessment, market analysis, patent strategy, and production planning. Domain experts then decomposed the steps and labeled the execution dependencies. The dataset includes task descriptions, step sequences, and their execution relationships, with decomposition criteria and dependency types established based on the characteristics of technology transfer.

3. A process for training a decomposition module of a large-scale achievement transformation model task decomposition technology based on LoRA fine-tuning, characterized in that... Training batches are constructed through random sampling. The step decomposition module is based on a large-scale achievement transformation model and uses a low-rank parameter matrix for efficient fine-tuning. The input is a description of the transformation task, and the output is a sequence of specific execution steps. Configure parameters such as the rank and learning rate of LoRA according to the characteristics of the technology transfer task.

4. A process for training a relationship prediction module using a task decomposition technique for large-scale achievement transformation models based on LoRA fine-tuning, characterized in that... Training batches are constructed through random sampling. The relationship prediction module is based on the same large-scale outcome transformation model and LoRA parameters, taking step pairs as input and outputting execution dependencies. The low-rank parameter configuration of the step decomposition module is shared to ensure consistency of prediction results.

5. A process for jointly optimizing the training of a task decomposition technology for成果转化大模型 based on LoRA fine-tuning, characterized in that The step decomposition loss L(θ) and the relation prediction loss L_rel(θ) are jointly optimized. The joint loss function is L_total=λ1*L(θ)+λ2*L_rel(θ), where λ1 and λ2 are weight coefficients. Based on the characteristics of the technology transfer task, optimization strategies and iteration rounds are set. It should be noted that the term "成果转化大模型" in the original text seems to be an incomplete or incorrect expression. It might need to be further clarified or corrected for a more accurate translation.

6. A process for applying a task decomposition technique for a large-scale achievement transformation model based on LoRA fine-tuning, characterized in that, In practical applications, the model receives descriptive information about the technology transfer task and simultaneously performs step decomposition and dependency prediction through shared low-rank parameters, providing execution guidance for the transfer process.

7. A process for training a decomposition module of a large-scale achievement transformation model task decomposition technology based on LoRA fine-tuning, characterized in that, According to the training step decomposition module of claim 3, the learning rate and optimization parameters are dynamically adjusted until the model achieves the expected performance.

8. A process for training a relationship prediction module using a task decomposition technique for large-scale achievement transformation models based on LoRA fine-tuning, characterized in that... According to claim 4, the training relationship prediction module reuses the same optimization strategy and judgment criteria to ensure synergy with the step decomposition module.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the LoRA-based large-scale model task decomposition method for achievement transformation as described in any one of claims 1-8.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the LoRA-based large-scale model task decomposition method for achievement transformation as described in any one of claims 1-8.

Images