With the rapid development of AI technology, users have a growing demand for algorithm-driven AI search-related capabilities. The AI search open platform has provided 20+ atomic service capabilities, which can be flexibly combined to build AI search. This release will focus on solving the inconvenience of users in code development. By integrating DSW capabilities and adding notebook functions, it will provide users with a more convenient code writing, debugging and running environment, and further improve the service debugging experience.
01
1. Document parsing service
It supports minute-level parsing of documents and images. For PDF, DOC, HTML, TXT and other documents, it can distinguish multiple formats, extract logical hierarchical structures such as titles and paragraphs, as well as text, tables, images, codes and other information from unstructured documents, remove headers and footers, identify superscripts, subscripts and other information, and output them in a structured format.
2. Image analysis service
For image data such as architecture diagrams and analysis charts, we provide image content understanding services. We can analyze and understand image content and recognize text based on a multimodal large model. We can also recognize image text based on OCR capabilities and extract text information for use in scenarios such as image retrieval and question-answering.
3. Document Slicing Service
Provides general document slicing services, which can be segmented based on document semantics, paragraph structure, and specified rules to improve the efficiency of subsequent document processing and retrieval. The output slicing tree can be used for context completion during retrieval.
4. Multilingual Vector Model
Text vectorization provides services that convert text data into dense vector form. It supports multiple text vector models in different languages, input lengths, and output dimensions, and can be used in scenarios such as information retrieval, text classification, and similarity comparison.
Text sparse vectorization provides a service for converting text data into sparse vector form. Sparse vectors take up less space to store and are often used to express keywords and word frequency information. They can be combined with dense vectors for mixed retrieval to improve the final retrieval effect.
The vector fine-tuning service provides vector model tuning services. It can help reduce the dimension of high-dimensional vectors without causing too much loss in retrieval effect by customizing the training vector dimensionality reduction model, so as to improve cost-effectiveness.
5. Query and analysis services
It provides query content analysis services. Based on the large language model and NLP capabilities, it can perform intent recognition, similar question expansion, NL2SQL processing, etc. on the query content entered by the user, effectively improving the retrieval and question-answering effects in RAG scenarios.
6. Search Engines
It provides vector retrieval and text retrieval engines, which can store vector & text content, build indexes, and perform online vector & text retrieval. After the engine service is activated, it can be used in combination with the rich API services of the AI search open platform.
7. Sorting Service
Provides relevance sorting services for Query and DOC. In RAG and search scenarios, the sorting service can be used to find more relevant content and return it in sequence. The introduction of the sorting service can effectively improve the accuracy of retrieval and large model generation.
8. Large model content generation service
It provides a variety of large language model services, including the full range of DeepSeek models (including R1/V3 and 7B/14B distilled versions), Tongyi series Tongyi Qianwen-Turbo ( https://x.sm.cn/BYoxwgv) , Tongyi Qianwen-Plus ( https://x.sm.cn/6yuEbHj) , and Tongyi Qianwen-Max ( https://x.sm.cn/EGjIvp5) large models. At the same time, it has a built-in OpenSearch-Tongyi Qianwen-Turbo large model, which uses the qwen-turbo large-scale language model as the model base, and performs supervised model fine-tuning to strengthen the RAG retrieval enhancement capabilities and reduce the model hallucination rate.
02
1. Rich AI search capabilities:
Relying on the leading model base to train AI search-specific models, it provides built-in search and RAG scenario ( https://x.sm.cn/DsWx8pl) full-link componentized services.
2. Flexible calling methods:
Through API and SDK calling services ( https://x.sm.cn/2JjDMYF) , developers, corporate customers and ISV technicians can integrate partial or full-link AI search services into their own business links.
3. Ready to use out of the box:
After activation, you can flexibly call the full range of services ( https://x.sm.cn/9l0UAfM) .
4. Best Practices:
Based on OpenSearch's years of experience in intelligent search and RAG, it has built-in a variety of AI search best practices, which can quickly build a search link that is more adapted to business needs .
03
1. Create a service development instance
On the AI Search Open Platform ( https://x.sm.cn/1XYQx3E) , select Model Service > Service Development , and then click New Development Instance .
Enter the instance name and description , select the instance resource specifications , and submit. The system will start deploying the instance.
For more details, please visit: https://x.sm.cn/DBrVnDj
2. Service Development
When the created service development instance is in the running state, you can enter the Notebook development environment through the following entrances:
Enter from the service development list:
From the scene center, you can enter the scene development environment with one click:
From the Service Experience Center, you can enter the service development environment with one click:
04
Service development capabilities are charged based on the CPU/GPU model purchased and the computing resources consumed. The CU unit price is 1.07 yuan/CU/hour
If you have completed development or debugging, you can stop the development instance at any time. Billing will be suspended after stopping.
Models revealed to the public:
The price of the deployment service is: CU price * number of CUs consumed by the model * number of machines purchased
For example: The price of deploying development services for 1 ops.basic1.gi.large = 1.07 yuan/CU/hour*0.61*1=0.65 yuan/hour
05
The release of the service development capability of Alibaba Cloud AI Search Open Platform aims to further improve user orchestration efficiency by integrating DSW capabilities and adding notebook functions.
