DEV Community: LisaLi

Join our ML Reading Group on May 14th ！

LisaLi — Tue, 07 May 2024 04:17:16 +0000

Hi Community, join us on May 14th for Jina AI’s ML Reading Group！

RSVP HERE
Presented Paper: QWEN technical report

Speakers bio:
Junyang Lin is a researcher at Alibaba Group, and he is now leading the research and open-source of the large language model and large multimodal model series QWEN. Additionally, he is a core member of OpenDevin. He has published papers and served as area chair and reviewer at top conferences including NeurIPS, ICML, ICLR, ACL, etc. He has gained over 4300 citations. Now he is devoted to developing open-source LLMs and LMMs.

Event Highlight:
During the talk, Junyang will share his experience in developing and training Qwen 1.5, one of the leading open-source LLMs!

Have you explored RAG already? – We Need Your Insights!

LisaLi — Mon, 22 Jan 2024 08:43:53 +0000

Hello Community! Recently, our team Jina AI launched the Jina Embeddings model which can be used to enhance Retrieval Augmented Generation (RAG).

Your Input Matters! Isn’t it exciting to tailor this technology to your needs?
Survey Link: https://forms.gle/xkyBpruTC7twZxSz7

Fill out our brief survey and get a 30-minute free consulting call to discuss how our model can support your projects. Let’s innovate together!

DocArray, an Inclusive and Standard Multimodal Data Model

LisaLi — Fri, 25 Nov 2022 16:00:35 +0000

November is a big time for Jina AI. From this month, DocArray will be hosted under the Linux Foundation AI & Data - a neutral home to build and support an open AI and data community. This is the start of a new day for DocArray.

In the ten months since DocArray's first release, we've seen more and more adoption and contributions from the open-source community. Today DocArray has over 150,000 downloads per month and powers hundreds of multimodal AI applications. At Jina AI, we're committed to delivering a powerful and easy-to-use tool for deep-learning engineers to represent, embed, search, store, and transfer multimodal data. But now we're sharing this commitment with you, our community and industry partners. Together with LF AI & Data, we're bringing together companies and individual contributors to build a neutral, inclusive and common standard multimodal data model. By donating DocArray to LF AI, we're letting it spread its wings and fly.

What does it mean to host a project at LF?
In this post we'll review the history of DocArray and unveil our future roadmap. In particular, we'll demonstrate some cool features that we're already developing and will roll out in an upcoming release.
**
A Brief History of DocArray**
We introduced the concept of "DocArray" in Jina 0.8 in late 2020. It was the jina.types module, intending to complete neural search design patterns by clarifying low-level data representation in Jina. Rather than working with Protobuf directly, the new Document class offered a simpler and safer high-level API to represent multimodal data.

If you are interested in knowing more, please check here.

Most Trending Open Source MLOps Tools of 2022

LisaLi — Thu, 10 Nov 2022 04:17:03 +0000

MLOps, or DevOps for machine learning, is a set of practices that aim to automate and improve the collaboration between data scientists and software engineers. It helps organizations better manage the complexities of developing and deploying machine learning models in production. In this article, we will review the top-5 most trending open source MLOps tools listed on OSSInsight.io in 2022.

Benefits of MLOps
One of the biggest benefits of MLOps is that it allows data scientists and engineers to work more closely together. Data scientists can focus on developing models, while engineers can focus on operationalizing them. This collaboration helps to ensure that models are deployed quickly and efficiently and that they meet the needs of the business.

Another benefit of MLOps is that it helps to automate the process of model development and deployment. This means that data scientists can spend less time on repetitive tasks, and more time on developing new models. Automation also helps to ensure that models are deployed consistently and with high quality.

Best 5 MLOps Tools in 2022
Implementing MLOps can be a challenge, but the benefits are clear. Organizations that adopt MLOps practices can improve the quality of their machine learning models and speed up the development and deployment process. In this article, we select the top-5 most trending open source MLOps tools in 2022 listed on OSSInsight.io, namely Jina (No.1), MLFlow (No.2), NNI (No.3), Kubefliow (No.4) and Label Studio (No.5). The rank is based on the stars, pull requests, pull request creators, and issues on the Github in 2022. OSSInsight is a powerful insight tool that can help one analyze in depth any single GitHub repository/developers, compare any two repositories using the same metrics, and provide comprehensive, valuable, and trending open source insights.

Jina from Jina AI is the top-1 most trending MLOps tool according to OSSInsight. Jina is an MLOps framework for multimodal AI. It eases the building of neural search and creative AI on the cloud. Jina uplifts a PoC into a production-ready service. Jina handles the infrastructure complexity, making advanced solution engineering and cloud-native technologies accessible to every developer. Despite the advanced cloud-native features that Jina offers, learning Jina is very straightforward. Document, Executor, and Flow are three fundamental concepts in Jina.

Document is the fundamental data structure. (This project is also an opensource project by the Linux Foundation)

Executor is a Python class with functions that use Documents as IO.

Flow ties Executors together into a pipeline and exposes it with an API gateway.
With these three concepts, one can easily build a semantic text search with sharding technology in just 45 lines of code!

With these three concepts, one can easily build a semantic text search with sharding technology in just 45 lines of code!

With Executor Hub, one can easily use LLMs or pretrained models on Hugging Face to embed Documents. However, in practice the performance is often suboptimal without proper domain adoption or knowledge transferring. Fine-tuning is an effective solution to improve the performance on neural search and embedding-related tasks. Jina AI also provides Finetuner tools makes fine-tuning easier, faster and performant by streamlining the workflow and handling all complexity and infrastructure on the cloud.

Finally, Jina AI also offers a hosting service for Jina projects, allowing one to deploy CPU/GPU-based Jina Flow with auto-provisioning in Kubernetes. It is in public beta and the hosting is for free at the moment.

Have you used Jina for multi-modal applications?

LisaLi — Tue, 25 Oct 2022 04:00:31 +0000

How will you build a multi-modal application? I just noticed the release ofJina which is a MLOps framework that empowers anyone to build cross-modal and multi-modal applications on the cloud. It uplifts a PoC into a production-ready service. Jina handles the infrastructure complexity, making advanced solution engineering and cloud-native technologies accessible to every developer.
If you tried before, please let me know how do you find about it? Thanks!