LMOps large model engineering platform and implementation method

Abstract

The invention discloses an LMOps large model engineering platform and an implementation method, the system comprises a data set management module, a sensitive word bank module, a data processing module, a model square module, a model fine tuning module, a model evaluation module, a module deployment module, a commercial large model module, a Prompt engineering module, an experience center module and a system management module, and the method comprises the steps of S1, managing data; s2, managing the model; s3, finely adjusting the model; s4, evaluating the model; s5, deploying the model; and S6, experiencing the model. According to the method, the capability is packaged into the platform through front and rear end programs, so that the technical threshold required by using a large model is reduced; the platform standardizes data standards and provides an efficient and visual data pipeline; a 1 + N low-cost deployment technology is supported, and the data and computing power cost is reduced; unified large model security, authority, interface and charging mechanisms are provided, and the engineering difficulty is reduced.

Description

Technical Field

[0001] The present invention relates to the field of large model technology, and in particular to a LMOps large model engineering platform and an implementation method thereof. Background Art

[0002] The development of large model technology is an important milestone in the field of artificial intelligence, marking the transition of machine learning from small-scale models based on specific tasks to general large-scale models that can handle multiple tasks. The models usually have hundreds of millions or even more parameters, and by training on large amounts of data, they can demonstrate outstanding capabilities in multiple fields such as natural language processing, computer vision, and speech recognition.

[0003] With the advancement of big model technology, various industries have begun to explore how to use this powerful tool to solve industry-specific problems, thus giving rise to the so-called industry big models. These models not only integrate the basic capabilities of general big models, but are also optimized or customized for application scenarios in specific industries, aiming to provide more accurate services and solutions.

[0004] However, while promoting the development of large industry models, it also faces several major problems: 1. Talent shortage: Building high-quality large models requires an interdisciplinary knowledge background, including but not limited to computer science, statistics, and expertise in specific fields. Currently, there is a relative shortage of professionals in the market who are proficient in AI technology and understand the business logic of a certain vertical field.

[0005] 2. Computing power and data shortage: Training large models is an extremely resource-intensive process. It not only requires massive amounts of data as support, but also requires powerful computing power to complete complex calculations. For many companies, obtaining enough high-quality data and affording the high computing costs have become a major obstacle.

[0006] 3. High engineering difficulty: It is not easy to transform the research results in the laboratory into practical products or services. It involves many aspects of work such as algorithm selection and tuning, system architecture design, and user experience optimization. At the same time, considering the particularities of different industries, how to ensure the security and stability of the solution is also one of the urgent problems to be solved. Summary of the invention

[0007] The purpose of the present invention is to propose a LMOps large model engineering platform and implementation method in response to the above-mentioned technical problems.

[0008] An LMOps large model engineering platform, including a data set management module, a sensitive word library module, a data processing module, a model square module, a model fine-tuning module, a model evaluation module, a module deployment module, a commercial large model module, a prompt engineering module, an experience center module, and a system management module; The data set management module defines data types, stores data in a versioned manner, imports data incrementally, exports data, and publishes data sets; The sensitive word library module defines nine categories of sensitive word libraries, each of which has built-in sensitive words and supports custom word libraries; The data processing module provides a visual and draggable data pipeline function to clean, enhance, filter and arrange the operation sequence of the data, and generate a new data set version after the processing is completed; The model square module centrally displays and manages selected preset general large models, realizes multi-dimensional screening, and supports fine-tuning, evaluation, and deployment of large models; The model fine-tuning module uses the data processed by the data processing module to select a model for parameter training according to business needs; The model evaluation module evaluates the model trained by the model fine-tuning module and evaluates the scene data set of the data set management module; The module deployment module deploys the model generated by the model fine-tuning module, and the model trained using loar supports the 1+N low-resource deployment mode; The commercial large model module uniformly manages token billing and introduces a variety of commercial large models; The Prompt engineering module provides Prompt templates, Prompt optimization, Prompt evaluation, and prompt words; The experience center module provides a model service experience center, which can input the models deployed in the module deployment module and the models connected in the commercial large model module according to the prompt engineering module or manually input them to experience the model generation effect; The system management module provides unified authority authentication, billing, and call statistics management.

[0009] Further, a LMOps large model engineering platform, the data types include Prompt+Response, Prompt+Multi-Response, Plain Text, Promp Set, Prompt-chosen-rejected; The versioned storage is implemented through MongoDB database and MinIO objects; The nine categories of sensitive word libraries include but are not limited to illegal words related to guns and explosives, pornographic words, and URLs; The large model uses Qwen1.5-14B-Chat; The module deployment module is based on the vllm and llama.cpp framework.

[0010] Further, a LMOps large model engineering platform, the data processing module includes a data cleaning submodule, a data enhancement submodule, and a vocabulary filtering submodule; The data cleaning module has preset cleaning operators, uses the BERT model to generate a language model, and uses the JIEBA model to perform Chinese word segmentation; The cleaning operators include but are not limited to normalizing spaces, converting traditional Chinese to simplified Chinese, and text repetition rate; The data enhancement submodule uses the large model prompt technology to enhance the data to achieve data generalization effect; The word library filtering submodule filters the sensitive words defined by the sensitive word library module.

[0011] Further, a LMOps large model engineering platform, the model fine-tuning module includes a training preparation submodule, a model training submodule, and a model management submodule; The training preparation submodule loads data and models from a shared file system according to the configuration; The model training submodule uses the downloaded data and the selected general model to train the model according to the filled-in hyperparameters and computing power configuration parameters; The model training module is implemented based on the LLaMA-Factory framework; The model management submodule uploads the trained model to the shared file system for use by the distributed system to publish the model.

[0012] Further, a LMOps large model engineering platform, the model evaluation module includes a manual evaluation submodule and an automatic evaluation submodule; The manual evaluation submodule integrates the subjective opinions and experience of human experts to score the model responses from different evaluation dimensions to evaluate the effectiveness of the model responses; The automatic evaluation submodule conducts an all-round evaluation of the output effect of the large model and provides a variety of scoring modes for factual or open-ended questions and answers.

[0013] Furthermore, a LMOps large model engineering platform, the BERT model is based on the Transformer architecture, which is a bidirectional deep neural network model, applies the bidirectional training of the transformer to language modeling, and adopts a masked language model for bidirectional training in the model; The BERT model only requires an encoder mechanism to generate a language model, without a decoder mechanism; The JIEBA model is a Chinese word segmentation tool, which consists of three word segmentation modes: precise mode, full mode and search engine mode; The precise mode cuts the sentence precisely, which is suitable for text analysis; The full mode scans all words in the sentence for irony; The search engine mode further segments long words based on the precise mode for search engine word segmentation.

[0014] A method for implementing a LMOps large model engineering platform includes the following sub-steps: S1: Administrative data; S11: Create training data set; S12: Create evaluation dataset; S13: data cleaning and filtering; S2: Management model; S21: Preset general large model; S22: Access to commercial large models; S3: fine-tuning the model; S4: Evaluation model; S41: Create a manual evaluation task; S42: Create automatic assessment tasks; S5: Deploy the model; S51: Select two deployment modes: normal mode and low resource mode; S52: If the low resource mode is selected, a weight model for deployment is selected; S53: Execute the deployment task and view the deployment result; S6: Experience model.

[0015] Furthermore, a method for implementing the LMOps large model engineering platform, wherein S3 includes the following sub-steps: S31: Select training type; S32: selecting a general model and version of a preset general large model; S33: Configure training hyper parameters; S34: Select a data set in the training data set; S35: Select the computing power configuration for training; S36: Execute the training task, monitor the training process, and view the training results.

[0016] Furthermore, a method for implementing a LMOps large model engineering platform, wherein S41 comprises the following sub-steps: S411: Select the model to be evaluated; S412: Select an evaluation data set; S413: Fill in model hyperparameters and computing power configuration; S414: Fill in the manual scoring rule configuration; S415: Execute the evaluation and view the evaluation results; The sub-steps of step S42 include: S421: Select the model to be evaluated; S422: Select an evaluation data set; S423: Fill in model hyper parameters and computing power configuration; S424: Fill in the configuration of the evaluation method and the automatic scoring rules; S425: Execute and check the evaluation results.

[0017] Furthermore, a method for implementing the LMOps large model engineering platform, wherein S6 comprises the following sub-steps: S61: Select the deployed model and the accessed commercial model service; S62: Adjust the temperature, top_n, and max_tokens model service call parameters; S63: Enter the question manually or select the Prompt template to ask the question; S64: Submit questions and experience model Q&A.

[0018] The beneficial effects of the present invention are as follows: through an LMOps large model engineering platform and implementation method, based on cutting-edge deep learning methods, a variety of open source technologies are cleverly integrated and optimized, and at the same time, the DevOps best practices of software engineering are introduced into the large model field, to realize the LMOps large model engineering platform and achieve low-code training deployment and application indicators; the capabilities are encapsulated into the platform through front-end and back-end programs, and large model training, evaluation, and deployment can be completed through the UI, reducing the technical threshold required to use large models; the platform uses cloud native technology to perform fine-grained management of computing power, supports 1+N low-cost deployment technology, provides data pipeline management, and can easily build high-quality private data sets; the platform encapsulates engineering capabilities and provides unified and consistent large model security, permissions, interfaces, and billing mechanisms. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a structural diagram of the LMOps large model engineering platform.

[0020] Figure 2 The present invention is a flowchart of an implementation method of a LMOps large model engineering platform. DETAILED DESCRIPTION

[0021] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.

[0022] As attached Figure 1 As shown, a LMOps large model engineering platform includes a data set management module, a sensitive word library module, a data processing module, a model square module, a model fine-tuning module, a model evaluation module, a module deployment module, a commercial large model module, a prompt engineering module, an experience center module, and a system management module; The data set management module defines data types, stores data in a versioned manner, imports data incrementally, exports data, and publishes data sets; The sensitive word library module defines nine categories of sensitive word libraries, each of which has built-in sensitive words and supports custom word libraries; The data processing module provides a visual and draggable data pipeline function to clean, enhance, filter and arrange the operation sequence of the data, and generate a new data set version after the processing is completed; The model square module centrally displays and manages selected preset general large models, realizes multi-dimensional screening, and supports fine-tuning, evaluation, and deployment of large models; The model fine-tuning module uses the data processed by the data processing module to select a model for parameter training according to business needs; The model evaluation module evaluates the model trained by the model fine-tuning module and evaluates the scene data set of the data set management module; The module deployment module deploys the model generated by the model fine-tuning module, and the model trained using loar supports the 1+N low-resource deployment mode; The commercial large model module uniformly manages token billing and introduces a variety of commercial large models; The Prompt engineering module provides Prompt templates, Prompt optimization, Prompt evaluation, and prompt words; The experience center module provides a model service experience center, which can input the models deployed in the module deployment module and the models connected in the commercial large model module according to the prompt engineering module or manually input them to experience the model generation effect; The system management module provides unified authority authentication, billing, and call statistics management.

[0023] Further, a LMOps large model engineering platform, the data types include Prompt+Response, Prompt+Multi-Response, Plain Text, Promp Set, Prompt-chosen-rejected; The versioned storage is implemented through MongoDB database and MinIO objects; The nine categories of sensitive word libraries include but are not limited to illegal words related to guns and explosives, pornographic words, and URLs; The large model uses Qwen1.5-14B-Chat; The module deployment module is based on the vllm and llama.cpp framework.

[0024] Further, a LMOps large model engineering platform, the data processing module includes a data cleaning submodule, a data enhancement submodule, and a vocabulary filtering submodule; The data cleaning module has preset cleaning operators, uses the BERT model to generate a language model, and uses the JIEBA model to perform Chinese word segmentation; The cleaning operators include but are not limited to normalizing spaces, converting traditional Chinese to simplified Chinese, and text repetition rate; The data enhancement submodule uses the large model prompt technology to enhance the data to achieve data generalization effect; The word library filtering submodule filters the sensitive words defined by the sensitive word library module.

[0025] Further, a LMOps large model engineering platform, the model fine-tuning module includes a training preparation submodule, a model training submodule, and a model management submodule; The training preparation submodule loads data and models from a shared file system according to the configuration; The model training submodule uses the downloaded data and the selected general model to train the model according to the filled-in hyperparameters and computing power configuration parameters; The model training module is implemented based on the LLaMA-Factory framework; The model management submodule uploads the trained model to the shared file system for use by the distributed system to publish the model.

[0026] Further, a LMOps large model engineering platform, the model evaluation module includes a manual evaluation submodule and an automatic evaluation submodule; The manual evaluation submodule integrates the subjective opinions and experience of human experts to score the model responses from different evaluation dimensions to evaluate the effectiveness of the model responses; The automatic evaluation submodule conducts an all-round evaluation of the output effect of the large model and provides a variety of scoring modes for factual or open-ended questions and answers.

[0027] Furthermore, a LMOps large model engineering platform, the BERT model is based on the Transformer architecture, which is a bidirectional deep neural network model, applies the bidirectional training of the transformer to language modeling, and adopts a masked language model for bidirectional training in the model; The BERT model only requires an encoder mechanism to generate a language model, without a decoder mechanism; The JIEBA model is a Chinese word segmentation tool, which consists of three word segmentation modes: precise mode, full mode and search engine mode; The precise mode cuts the sentence precisely, which is suitable for text analysis; The full mode scans all words in the sentence for irony; The search engine mode further segments long words based on the precise mode for search engine word segmentation.

[0028] As attached Figure 2 As shown, a method for implementing the LMOps large model engineering platform includes the following sub-steps: S1: Administrative data; S11: Create training data set; S12: Create evaluation dataset; S13: data cleaning and filtering; S2: Management model; S21: Preset general large model; S22: Access to commercial large models; S3: fine-tuning the model; S4: Evaluation model; S41: Create a manual evaluation task; S42: Create automatic assessment tasks; S5: Deploy the model; S51: Select two deployment modes: normal mode and low resource mode; S52: If the low resource mode is selected, a weight model for deployment is selected; S53: Execute the deployment task and view the deployment result; S6: Experience model.

[0029] Furthermore, a method for implementing the LMOps large model engineering platform, wherein S3 includes the following sub-steps: S31: Select training type; S32: selecting a general model and version of a preset general large model; S33: Configure training hyper parameters; S34: Select a data set in the training data set; S35: Select the computing power configuration for training; S36: Execute the training task, monitor the training process, and view the training results.

[0030] Furthermore, a method for implementing a LMOps large model engineering platform, wherein S41 comprises the following sub-steps: S411: Select the model to be evaluated; S412: Select an evaluation data set; S413: Fill in model hyperparameters and computing power configuration; S414: Fill in the manual scoring rule configuration; S415: Execute the evaluation and view the evaluation results; The sub-steps of step S42 include: S421: Select the model to be evaluated; S422: Select an evaluation data set; S423: Fill in model hyper parameters and computing power configuration; S424: Fill in the configuration of the evaluation method and the automatic scoring rules; S425: Execute and check the evaluation results.

[0031] Furthermore, a method for implementing the LMOps large model engineering platform, wherein S6 comprises the following sub-steps: S61: Select the deployed model and the accessed commercial model service; S62: Adjust the temperature, top_n, and max_tokens model service call parameters; S63: Enter the question manually or select the Prompt template to ask the question; S64: Submit questions and experience model Q&A.

[0032] Specific embodiment 1 JIEBA model The JIEBA model comes with a dictionary named dict.txt, which contains more than 20,000 words, including the number of occurrences and parts of speech of the words. JIEBA word segmentation belongs to the probabilistic language model word segmentation. The task is to find a segmentation scheme S among all the results obtained from the full segmentation so that P(S) is maximized. JIEBA also supports custom dictionaries, which can expand the vocabulary and improve the accuracy of word segmentation. At the same time, the JIEBA model also supports traditional Chinese word segmentation and English word segmentation to meet the needs of different scenarios.

[0033] Specific embodiment 2 MongoDB database: MongoDB is an open source, document-based NoSQL database that uses a JSON-like format called BSON, or Binary JSON, to store data. MongoDB is a non-relational NoSQL database that is particularly suitable for processing large amounts of unstructured or semi-structured data.

[0034] Following are some key features of MongoDB: Document storage: MongoDB stores data in the form of documents. Each document is a collection of key-value pairs, similar to a JSON object. This model is very suitable for storing hierarchical data and allows the flexibility to add new fields without pre-defining the entire structure. Dynamic mode: Unlike traditional relational databases, MongoDB does not require a fixed table structure to be defined in advance, which means that different documents in the same collection, equivalent to a table in a relational database, can have completely different structures; High scalability: MongoDB supports horizontal expansion and can distribute data across multiple servers through sharding technology, thereby improving read and write performance and storage capacity; Rich query language: MongoDB provides powerful query capabilities, including complex aggregation operations and full-text search; Built-in replication and high availability: Data redundancy is achieved through replica sets, ensuring data security and high availability of the system; Support for multiple storage engines: Users can choose different storage engines according to their needs, such as WiredTiger, MMAPv1, etc.

[0035] Index support: To speed up queries, MongoDB allows you to create indexes for any field in a document, and even composite indexes for arrays and nested documents; Transaction support: Starting from version 4.0, MongoDB introduced multi-document transaction support, allowing atomic operations to be performed across multiple documents, collections, and even databases; Community and support: As an open source project, MongoDB has an active community that provides a large number of resources and tools, and there are also commercial versions that provide professional technical support and services.

[0036] Specific Example 3 LlamaFactory model The LlamaFactory model is a large language model (LLM) fine-tuning framework that aims to achieve higher efficiency by optimizing the performance of large models, especially during inference and fine-tuning.

[0037] The core technology of the LlamaFactory model is divided into the following parts: Multimodal model integration LlamaFactory usually supports processing multimodal data, such as text, images, audio, etc., so that the fine-tuned model can perform well in multiple domain tasks. Through a modular architecture, users can easily expand the input source of the model and adapt to different data types; Efficient fine-tuning LlamaFactory uses lightweight adaptive fine-tuning, such as Low-Rank Adaptation of LLMs technology, to train large language models. LoRA allows the fine-tuning process to train and update only a small number of parameters without adjusting all parameters of the model, significantly reducing the computational cost and resource requirements of fine-tuning; Distributed training LlamaFactory model optimizes the training efficiency of the model through distributed architecture, and can perform large-scale parallel training on multiple GPUs or even multiple nodes. By optimizing communication strategies and memory management, the training process is accelerated and hardware bottlenecks are reduced. LlamaFactory integrates Prompt Tuning technology, which quickly guides the output of the model by inputting specific prompts, avoiding the adjustment of a large number of parameters in traditional fine-tuning, making the fine-tuning process of LlamaFactory more flexible and scalable.

[0038] Specific embodiment 4 MinIO object storage system: MinIO is a high-performance object storage system that is compatible with the Amazon S3 API, which means it can be seamlessly integrated with widely used cloud storage services. MinIO is designed for private cloud and edge computing environments, can handle large-scale data sets, and supports a variety of data protection mechanisms, such as Erasure Coding to ensure high availability and persistence of data.

[0039] The main features are: High performance: MinIO can achieve very high throughput and low latency on standard hardware, which is particularly suitable for large-scale data applications that require fast read and write speeds, such as machine learning and big data analysis; Compatible: Fully compatible with the Amazon S3 API, which means that all tools, applications, and services that use S3 can be directly connected to MinIO. This compatibility makes it easy to migrate existing S3-based workloads. Distributed architecture: MinIO can be deployed as a single node or a distributed multi-node cluster. In distributed mode, data redundancy is provided through erasure coding, ensuring that data is not lost even if multiple nodes fail. Security: Supports TLS / SSL encrypted transmission to ensure data security during transmission, and provides authentication and authorization mechanisms, including policy-based access control. Supports server-side encryption, which can encrypt data stored in MinIO; Easy to deploy and manage: MinIO can be easily deployed in containerized environments such as Kubernetes. It provides command line tools and graphical interfaces for management and monitoring; Cross-platform support: It supports multiple operating systems such as Linux, Windows, and macOS. It can also run on the ARM architecture, which is suitable for edge computing scenarios; Open Source: MinIO is open source software, released under the GNU Affero General Public License v3.0 (AGPLv3).

[0040] The open source community is active, with a large number of contributors and support resources; Rich plug-ins and tools: The community provides many tools and plug-ins for operations such as backup, synchronization, and migration. It supports monitoring tools such as Prometheus and Grafana to facilitate performance monitoring and alarm settings.

[0041] Specific embodiment 5 Qwen1.5-14B-Chat: Qwen-1.5-14B-Chat is a version in the Qwen series of large-scale language models launched by Alibaba Cloud. It is optimized for dialogue and chat scenarios. This model is based on the Transformer architecture and has 14 billion parameters. It aims to provide high-quality natural language understanding and generation capabilities and is suitable for a variety of application scenarios.

[0042] It has the following features: Large-scale parameters: With 14 billion parameters, Qwen-1.5-14B-Chat can capture complex language patterns and subtle semantic differences. A large number of parameters helps improve the model's ability to understand and generate long texts. Conversation optimization: The model is specially trained to handle multi-turn conversations and excels at maintaining contextual coherence, being able to understand and respond to user questions or statements while maintaining the fluency and relevance of the conversation; Breadth and depth of knowledge: The training data covers a wide range of subject areas, from general knowledge to professional knowledge. Through extensive pre-training and fine-tuning, the model can demonstrate a deep knowledge background on different topics. Multi-language support: Qwen series models usually support multiple languages. Although they are mainly optimized for Chinese, they can also handle input and output of English and some other languages, which makes them more flexible and practical in international applications; Security and Compliance: We adhere to strict data privacy protection principles, take security and compliance requirements into consideration during design, take measures to avoid generating harmful, false or biased content, and strive to ensure the security and reliability of output.

[0043] Specific embodiment 6 llama.cpp llama.cpp is an open source project that aims to run large-scale language models based on the Transformer architecture through an efficient inference engine implemented in C++, with a special focus on running these models in resource-constrained environments, such as personal computers or even embedded devices.

[0044] Here are the key points about llama.cpp: Lightweight and efficient: llama.cpp is designed to provide a way for even users with limited computing resources to run large language models; Portability: Supports multiple platforms, including but not limited to Windows, Linux, and macOS, as well as ARM architectures such as Raspberry Pi; Quantization technology: Quantization technology is used to convert floating-point parameters into lower-precision data types, such as int8 or float16, thereby reducing memory usage and accelerating the inference process; Multithreading: Improves computational efficiency through parallel processing, especially when executed on a CPU; Compatibility: In addition to the natively supported LLaMA model, other similar Transformer models can also be loaded with only a small amount of adaptation work; Community-driven: As an active open source project, llama.cpp is supported by a broad developer community, with new features being added and existing issues fixed.

[0045] This solution uses an LMOps large model engineering platform and implementation method, encapsulates capabilities into the platform through front-end and back-end programs, and uses the UI to complete large model training, evaluation, and deployment, reducing the technical threshold required to use large models; the platform standardizes data standards and provides efficient and visual data pipelines to facilitate the production of high-quality data sets; the platform uses cloud-native technology to perform fine-grained pooling management of computing power, supports 1+N low-cost deployment technology, and reduces data and computing power costs; encapsulates engineering capabilities, provides unified and consistent large model security, permissions, interfaces, and billing mechanisms, and reduces engineering difficulty.

[0046] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A LMOps large model engineering platform, characterized in that: It includes data set management module, sensitive word library module, data processing module, model square module, model fine-tuning module, model evaluation module, module deployment module, commercial large model module, prompt engineering module, experience center module, and system management module; The data set management module defines data types, stores data in a versioned manner, imports data incrementally, exports data, and publishes data sets; The sensitive word library module defines nine categories of sensitive word libraries, each of which has built-in sensitive words and supports custom word libraries; The data processing module provides a visual and draggable data pipeline function to clean, enhance, filter and arrange the operation sequence of the data, and generate a new data set version after the processing is completed; The model square module centrally displays and manages selected preset general large models, realizes multi-dimensional screening, and supports fine-tuning, evaluation, and deployment of large models; The model fine-tuning module uses the data processed by the data processing module to select a model for parameter training according to business needs; The model evaluation module evaluates the model trained by the model fine-tuning module and evaluates the scene data set of the data set management module; The module deployment module deploys the model generated by the model fine-tuning module, and the model trained using loar supports the 1+N low-resource deployment mode; The commercial large model module uniformly manages token billing and introduces a variety of commercial large models; The Prompt engineering module provides Prompt templates, Prompt optimization, Prompt evaluation, and prompt words; The experience center module provides a model service experience center, which can input the models deployed in the module deployment module and the models connected in the commercial large model module according to the prompt engineering module or manually input them to experience the model generation effect; The system management module provides unified authority authentication, billing, and call statistics management.

2. A LMOps large model engineering platform according to claim 1, characterized in that: The data types include Prompt+Response, Prompt+Multi-Response, Plain Text, Promp Set, Prompt-chosen-rejected; The versioned storage is implemented through MongoDB database and MinIO objects; The nine categories of sensitive word libraries include but are not limited to illegal words related to guns and explosives, pornographic words, and URLs; The large model uses Qwen1.5-14B-Chat; The module deployment module is based on the vllm and llama.cpp framework.

3. The LMOps large model engineering platform according to claim 1, characterized in that: The data processing module includes a data cleaning submodule, a data enhancement submodule, and a vocabulary filtering submodule; The data cleaning module has preset cleaning operators, uses the BERT model to generate a language model, and uses the JIEBA model to perform Chinese word segmentation; The cleaning operators include but are not limited to normalizing spaces, converting traditional Chinese to simplified Chinese, and text repetition rate; The data enhancement submodule uses the large model prompt technology to enhance the data to achieve data generalization effect; The word library filtering submodule filters the sensitive words defined by the sensitive word library module.

4. The LMOps large model engineering platform according to claim 1, characterized in that: The model fine-tuning module includes a training preparation submodule, a model training submodule, and a model management submodule; The training preparation submodule loads data and models from a shared file system according to the configuration; The model training submodule uses the downloaded data and the selected general model to train the model according to the filled-in hyperparameters and computing power configuration parameters; The model training module is implemented based on the LLaMA-Factory framework; The model management submodule uploads the trained model to the shared file system for use by the distributed system to publish the model.

5. The LMOps large model engineering platform according to claim 1, characterized in that: The model evaluation module includes a manual evaluation submodule and an automatic evaluation submodule; The manual evaluation submodule integrates the subjective opinions and experience of human experts to score the model responses from different evaluation dimensions to evaluate the effectiveness of the model responses; The automatic evaluation submodule conducts an all-round evaluation of the output effect of the large model and provides a variety of scoring modes for factual or open-ended questions and answers.

6. The LMOps large model engineering platform according to claim 3, characterized in that: The BERT model is based on the Transformer architecture and is a bidirectional deep neural network model that applies bidirectional training of transformers to language modeling and uses a masked language model for bidirectional training in the model. The BERT model only requires an encoder mechanism to generate a language model, without a decoder mechanism; The JIEBA model is a Chinese word segmentation tool, which consists of three word segmentation modes: precise mode, full mode and search engine mode; The precise mode cuts the sentence precisely, which is suitable for text analysis; The full mode scans all words in the sentence for irony; The search engine mode further segments long words based on the precise mode for search engine word segmentation.

7. A method for implementing a LMOps large model engineering platform, based on the LMOps large model engineering platform according to any one of claims 1 to 6, characterized in that: It includes the following sub-steps: S1: Administrative data; S11: Create training data set; S12: Create evaluation dataset; S13: data cleaning and filtering; S2: Management model; S21: Preset general large model; S22: Access to commercial large models; S3: fine-tuning the model; S4: Evaluation model; S41: Create a manual evaluation task; S42: Create automatic assessment tasks; S5: Deploy the model; S51: Select two deployment modes: normal mode and low resource mode; S52: If the low resource mode is selected, a weight model for deployment is selected; S53: Execute the deployment task and view the deployment result; S6: Experience model.

8. The method for implementing the LMOps large model engineering platform according to claim 7, characterized in that: The S3 comprises the following sub-steps: S31: Select training type; S32: selecting a general model and version of a preset general large model; S33: Configure training hyper parameters; S34: Select a data set in the training data set; S35: Select the computing power configuration for training; S36: Execute the training task, monitor the training process, and view the training results.

9. The method for implementing the LMOps large model engineering platform according to claim 7, characterized in that: The S41 comprises the following sub-steps: S411: Select the model to be evaluated; S412: Select an evaluation data set; S413: Fill in model hyperparameters and computing power configuration; S414: Fill in the manual scoring rule configuration; S415: Execute the evaluation and view the evaluation results; The sub-steps of step S42 include: S421: Select the model to be evaluated; S422: Select an evaluation data set; S423: Fill in model hyper parameters and computing power configuration; S424: Fill in the configuration of the evaluation method and the automatic scoring rules; S425: Execute and check the evaluation results.

10. The method for implementing the LMOps large model engineering platform according to claim 7, characterized in that: The S6 comprises the following sub-steps: S61: Select the deployed model and the accessed commercial model service; S62: Adjust the temperature, top_n, and max_tokens model service call parameters; S63: Enter the question manually or select the Prompt template to ask the question; S64: Submit questions and experience model Q&A.

Images