By Aaron Handoko (Solution Architect) Rifandy Zulvan (Solution Architect), and Darlin Valentine (Channel Development Ecosystem) Alibaba Cloud Indonesia
Earlier this month, Alibaba Cloud Indonesia launched a new community group for partners, Alibaba Cloud Indonesia Digital Tech Ecosystem community, where the latest trends, trainings, products updates, and new initiatives being shared. During the first month, partners were able to engage with each other as well as getting introduced to the high level of Alibaba Cloud capabilities on providing Generative AI solutions.
Generative AI is seeing greater use across a range of industries. Applications and services are being rebuilt and reconfigured with generative AI technologies such as large language models (LLMs). Alibaba Cloud's comprehensive range of cloud products and solutions, in conjunction with leading AI capabilities, can support you to accelerate innovation, create new and intelligent customer experiences, and generate opportunities for business success.
Alibaba Cloud provides a full-stack solution for generative AI development, which includes a collection of compute-optimized ECS instances with high-performance GPUs, a powerful AI computing platform, a one-stop AI development platform, and acceleration features for AI model training and deployment.
Alibaba Cloud Services boasts a robust suite of capabilities on Generative AI. Alibaba Cloud provides IaaS, PaaS, and SaaS that can help customer to have Generative AI solutions, which includes a collection of compute-optimized ECS instances with high-performance GPUs, a one-stop AI development platform, and acceleration features for AI model training and deployment, and an easy-to-use LLM services. These are Alibaba Cloud services that can support Generative AI solutions.
1. ECS
ECS is Alibaba Cloud basic computing services, this service provides wide use computing use case including for AI development. ECS compute capabilities include ECS GPU-Accelerated instances with Nvidia T4, Nvidia P100, and Nvidia A10 GPUs that are available on Indonesia region. These GPUs can help customer to develop AI effectively on cloud.
2. Analytics DB for PostgreSQL (with VectorDB)
Analytics DB for PostgreSQL has a VectorDB features that can store data in vector, this implementation can be used in LLM solutions. Vector DB can make faster query with a high relevancy text search, and can be used to embed LLM with customer own data or knowledges.
3. Dataworks
Dataworks provides end-to-end solutions for AI and Big Data to facilitate your business in performing data ingestion, data integration, data visualization, and also can be integrated with Alibaba Cloud AI development platform, Platform for AI (PAI).
4. Platform for AI (PAI)
PAI is Alibaba Cloud main platform for AI development, this solution includes services for data modeling, training, and AI/ML model deployments. PAI also supported by ECS capability as the main computing resources, and can be integrated with other services like Analytics DB, RDS, OSS, NAS as the data sources.
5. OpenSearch
OpenSearch is Alibaba Cloud SaaS solutions for LLM and this can be easily use by customers with only API or SDK on customers’s application. OpenSearch is supported with various LLM models such as Alibaba’s model Tongyi Qianwen (for chat) and also Tongyi Wanxiang (Prompt to Image), Llama2, and Falcon.
Generative Artificial Intelligence (AI) has emerged as a groundbreaking innovation, revolutionizing various aspects of technology, creativity, and problem-solving. Tasks including creating pictures, making summary and suggestions, that were usually reliance on time, experience and creativity from human, can be quickly done just by inputting a prompt. No wonder modern industries are looking for a method to implement this technology in their business pipeline.
Generative AI models API can be easily accessible from the internet, but there are multiple concerns when using this technology. In this week Cloud Fighters series, attendees from multiple technical backgrounds interactively examined the pain-points of modern companies in using Gen-AI technology and also discussed the step by step how Generative AI works on Cloud.
1. The Difference of Traditional AI and Generative AI
According to Gartner 2021, “Generative AI is a disruptive technology that can generate artifacts that previously relied on humans, delivering innovative results without the biases of human experiences and thought processes”. Within the cloud fighter’s forum, it was emphasized that the clear difference of traditional AI and Generative AI (Gen-AI) is the creative output that Gen-AI generate. Compared to Gen-AI, traditional AI can produce prediction but solely based on the pattern of the training data provided.
2. The Pain Points of Generative AI
Pain Points of using Generative AI for large enterprise includes its Security, Integrity and Useability. Since the company data used to produce personalized Gen-AI models need to be uploaded via API, some company concerns about its data being breach. In the forum, we all agreed that this is a valid concern that should be addressed properly in order to avoid sensitive data leakage. Some ideas to solve this issue includes having the model in the same VPC or connection as the application or having the model and application completely on-premise.
Furthermore, integrity should also be the main concern when creating personalized Gen-AI models. Integrity means that the answer produced by the model must be solely based on the data provided by the company. This is very important to avoid misinformation for the company’s customer and also the decision makers.
Lastly, the model usability is another factor. Model should be easily accessible by the user and can be integrated effortlessly in an application. Multiple participants of cloud fighters, asked about the connections of LLM models shows their concern of usability. As we discussed in the forum, currently Alibaba Cloud LLM solution cannot seamlessly access an existing database and additional script must be made to make this possible. This is a great input for Alibaba Cloud team as well to improve our LLM solution.
3. Generative AI in Alibaba Cloud
Alibaba Cloud have launched their own LLM models called Tongyi Qianwen (for chat) and also Tongyi Wanxiang (Prompt to Image). Open Source Tongyi Qianwen and Tongyi Wanxiang are already publicly available here. In the forum we discussed two Alibaba Cloud Generative AI Solutions: IaaS Solution containing AnalyticDB (Vector Database) & PAI (Platform for AI), and also SaaS Solution (OpenSearch). Both solutions are using retrieval document technique to get the most relevant content for questions provided by users.
3.1. IaaS Solution
The detail of the above architecture can be found here. The solution above is flexible as it can use different LLM models or retrieval plugins. The model can be downloaded into the Alibaba Cloud’s Platform for AI (PAI) then connected to the AnalyticDB for Postgres (Vector DB) where the company’s data is saved. However, there are multiple configurations that needs to be done and the solution is difficult to maintain as it involves multiple instances.
3.2. SaaS Solution (OpenSearch)
OpenSearch is a more user-friendly solution of Alibaba Cloud’s Generative AI. The data needed for the model is saved in the OpenSearch console and only need SDK to integrate it to the existing application. OpenSearch has multiple parameters that can be used for your business needs. However, it is not flexible as it currently only provides Llama2 13b model and Falcon 6b model. The instruction on how to connect to OpenSearch can be found here.
In this new program, we can see the high enthusiasm of the audience in discussing the new technology to improve their existing applications. This program is a good place for tech minded people who would love to be exposed to new technology, step-by-step implementation, best practices, and learn it first hand in the cloud.
Exploring Top 4 Alibaba Cloud ISV Solutions in Diverse Industries
99 posts | 17 followers
FollowAlibaba Cloud Community - October 25, 2024
Alibaba Cloud Community - November 24, 2023
Farruh - November 23, 2023
Amuthan Nallathambi - July 12, 2024
Rupal_Click2Cloud - November 23, 2023
Alibaba Cloud Community - May 16, 2024
99 posts | 17 followers
FollowOffline SDKs for visual production, such as image segmentation, video segmentation, and character recognition, based on deep learning technologies developed by Alibaba Cloud.
Learn MoreA secure environment for offline data development, with powerful Open APIs, to create an ecosystem for redevelopment.
Learn MoreAlibaba Cloud is committed to safeguarding the cloud security for every business.
Learn MoreSimple, secure, and intelligent services.
Learn MoreMore Posts by Alibaba Cloud Indonesia