Rajeswaran V (PhD)’s Post

Deploying open-source models in real-world projects can be costly and complex; Het Trivedi shows how you can make the process more efficient, and compares the performance for Llama 2 using two different inference methods.

This is what happens behind the kernel trick which is a miraculous mathematical technique used in ML algorithms such as SVMs. The kernel trick transforms linearly inseparable data into a linearly separable form by adding extra dimension(s) to the original data, even without calculating the coordinates in the higher dimension. After transformation, data become linearly separable with a linear hyperplane in the higher dimension!

Inference Tables are here to simplify your monitoring and diagnostics for models! With continuous logging of serving request inputs and responses from Databricks Model Serving endpoints, Inference Tables save them into a Delta table in Unity Catalog. This means that you can easily track and analyze the performance of your models without the need for manual logging. Try Inference Tables today and streamline your model monitoring process! #InferenceTables #Databricks #ModelServing #UnityCatalog #AI #MachineLearning https://lnkd.in/gyjEdAqR

Graph of Thoughts Presents a prompting approach that models text generated by LLMs as an arbitrary graph. It enables combining arbitrary "thoughts" and enhancing them using feedback loops. The core idea is to enhance the LLM capabilities through "network reasoning" and without any model updates. This could be seen as a generalization of the now popular Chain-of-Thought and Tree-of-Thought. The interesting part is that this approach can enable rapid prototyping of even more novel prompting ideas. I've been thinking about something like this for some time. It makes sense to build prompts that operate more generally and that could work with all kinds of models regardless of formatting or style. I think the key with prompting models is how and when information is combined and graphs are excellent at solving these types of problems. I will be writing an extended summary of this and adding it to our prompting guide as well. The authors suggest that this work brings "LLM reasoning closer to human thinking or brain mechanisms such as recurrence, both of which form complex networks." (link in the comments)

💡 The Graph of Thoughts: A New Era for LLM Reasoning💡Thrilled to discover the cutting-edge graph-based approach to modeling text generated by LLMs that delves deep into the realms of "network reasoning"! 🌐🔍 This innovative technique, shared by Elvis S., opens doors for game-changing advancements in model prompts that go beyond formats and styles. 💥Imagine the potential: rapid prototyping, enhanced information combining, and the development of even more inspiring and diverse prompting ideas! 🚀✨ Such technological leaps bring us one step closer to our wildest dream—humans and complex networks engaging in reciprocal AI reasoning. 🧠🌱Stay tuned as we expand upon this revolution in our prompting guide, empowering all forward-thinking professionals to unlock the full potential of LLM wisdom. 📚✍️✨#ThoughtLeadership #AIAdvancement #GraphOfThoughts #LLMReasoning

This post is NOT about Gemini. Let's talk about Jellyfish instead. Jellyfish is an Open Source (OS) 13B parameter model designed to be a universal problem solver. Now that sounds... useful? More specifically, it has been fine-tuned to solve challenging tasks within the realm of Data Processing (DP). What sets Jellyfish apart is not just its ability to generalize across a variety of DP tasks but also its humble hardware demands, which could lower the threshold for companies to turn their raw data into a useful resource. Jellyfish is built upon the Llama 2 model, but unlike other heavyweight contenders, it has been designed to thrive on local, economical GPUs, providing users the twin benefits of data security and the freedom to further customize the model. On top of that, the model comes with an interpreter that sheds light on the 'why' behind its output, providing much-needed transparency in machine learning operations. All of this is quite a mouthful and only time will tell how useful this model will 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 proof to be in practice. With complex expert tasks and workflows, it's often not easy to achieve adoption among practitioners. [arXiv] https://lnkd.in/dbqk8ZY2 [Model] https://lnkd.in/dqqYzmJD ↓ Liked this post? Get weekly AI highlights and papers-of-the-week directly to your inbox 👉 llmwatch.com

Marqo : It is more than a vector database, it's an end-to-end vector search engine for both text and images. Vector generation, storage and retrieval are handled out of the box through a single API. No need to bring your own embeddings. Why Marqo? Vector similarity alone is not enough for vector search. Vector search requires more than a vector database - it also requires machine learning (ML) deployment and management, preprocessing and transformations of inputs as well as the ability to modify search behavior without retraining a model. Marqo contains all these pieces, enabling developers to build vector search into their application with minimal effort. https://www.marqo.ai/ #marqo #vectordatabase #machinelearning #datascience

This plus traditional encoder style models fine-tuned for the task will lead to cost effective ways to procure high quality datasets. This will unlock multiple downstream capabilities and benefit customers with delightful experience. #llm #dataquality #nlp #searchmeasurent

Interesting quick read (2 interesting charts) on the costs of using various LLMs. There's a 120x difference between the top and bottom options. https://lnkd.in/g3aftBxb

Labeling with Confidence: Confidence estimation is an effective tool to mitigate hallucinations when leveraging LLMs for data labeling and enrichment: If we are able to estimate the model’s inherent confidence in its response, we can automatically reject low confidence labels, chain and ensemble LLMs. Excited to share a bit more about what we've been exploring and building at Refuel in this direction: https://lnkd.in/gyg54vfZ. You can access all of these features in Autolabel (https://lnkd.in/g7dX8Awi) with a one line config change to your labeling task!