Chloe He

Unlock the potential of Meta’s Llama3 for biotech applications with torchtune, a PyTorch-native library that simplifies fine-tuning of large language models. Whether you're looking to enhance drug discovery, genetic analysis, or scientific writing, our latest blog provides a guide to leverage this powerful tool effectively. 🔗 Learn more here: https://lnkd.in/g9WnrNSN #Biotechnology #Llama3 #torchtune #AI #MachineLearning

6

🌟 Building Retrieval-Augmented Generation (RAG) Systems with LLMs and LangChain 🌟 I've been diving into how Large Language Models (LLMs) like GPT, Cohere, and Ollama, along with vector databases such as Pinecone and Qdrant, can create highly accurate and context-aware AI systems using LangChain. This innovative approach is known as Retrieval-Augmented Generation (RAG). 🔑 What is RAG? RAG is an AI architecture that enhances text generation by integrating information retrieval. It retrieves relevant data from external sources through a vector database, enriching the text generation process to ensure more informed and precise responses from LLMs. ⚙️ Steps to Build a RAG System: Create Embeddings: Use LLMs like GPT or Cohere to generate vector embeddings from unstructured data (e.g., text). Store Embeddings: Save these embeddings in a vector database like Pinecone or Qdrant for fast, similarity-based retrieval. Query Processing: Convert user queries into embeddings and retrieve similar documents from the vector database. Augment Generation: Feed the retrieved documents into an LLM via LangChain, enhancing the LLM's response generation with real-time data. LangChain Integration: Use LangChain to connect LLMs with vector databases, creating seamless workflows for retrieving and generating context-aware text. 🔍 Why Use RAG? RAG significantly boosts accuracy and contextual understanding by grounding LLM outputs in external, up-to-date information. This approach reduces hallucinations—instances where models generate incorrect facts—and provides contextually relevant responses. It’s particularly useful for applications like knowledge assistants, content generation, and recommendation systems. 🧠 Technologies Involved: LLMs: GPT, Cohere , Ollama are used for embedding generation and response generation. Vector Databases: Pinecone, Qdrant, and ChromaDB are utilized for storing and retrieving embeddings based on semantic similarity. LangChain: This framework connects LLMs with vector databases, enabling smooth RAG pipelines for enhanced AI responses. 🚀 Applications: Knowledge Retrieval: Systems that pull context-aware, real-time data for addressing complex queries. Personalized Content Generation: Generate accurate and fact-based content by retrieving the latest information. AI Assistants: Use retrieval-augmented LLMs to build intelligent assistants that provide relevant and timely information to users. Combining LLMs with vector databases and LangChain enables the creation of intelligent, scalable, and contextually aware systems for AI-driven applications. Exploring these emerging technologies has been an exciting part of my internship journey! Resources to dive deeper: LangChain Docs: https://docs.langchain.com Pinecone + LangChain Integration: https://lnkd.in/dZNbEZD9 OpenAI Embeddings Guide: https://lnkd.in/dxzCfzfV #RAG #LLMs #LangChain #VectorDatabases #AI #KnowledgeRetrieval #GPT #Cohere #Ollama #Qdrant

9

Extremely interesting research that dives deep into how best to scale the inference-time computation in LLMs for improving model performance. More specifically, given a fixed inference-time compute budget, what strategy for scaling LLMs given tradeoffs provides the best model performance results? Here, they found that leveraging a "compute-optimal" scaling strategy that adaptively allocate test-time compute per prompt can be very effective for improving efficiency and model performance compared to scaling model parameters, and this can have a lot of positive implications on how best to scale and deploy LLMs for impact. #generativeai #genai #llm #largelanguagemodels #large #language #models #efficiency #ai #ml #dl #artificialintelligence #deeplearning #machinelearning #compute #research #impact

43

3 Comments

🚀 Want to employ the power of #PyTorch for your scientific models without a steep learning curve? 👀 Check out our latest blog, where Travis Oliphant discusses how PyTorch's integration with the Array API makes it easy to use with familiar libraries like NumPy and scikit-learn.🔥 https://lnkd.in/dDRkEY3v #Innovation

10

✨ Exciting Side Project Discovery: LLM + MCTS Integration! ✨ Arushi Somani recently embarked on an intriguing project combining Large Language Models (LLMs) and Monte Carlo Tree Search (MCTS). Here's the setup: She utilized Humaneval and LLaMA. Implemented MCTS-assisted inference decoding with Humaneval pass/fail as the reward and performed k rollouts per question. Compared the results to sampling the model for pass@k scores. Her findings: For sufficiently large k, MCTS-assisted inference consistently outperforms random sampling. On average, while random sampling needed 10 tries to get the correct answer, MCTS only required 6.77 attempts. over enough attempts, MCTS searches solution space more efficiently Writeup : https://lnkd.in/dB82vYc4 Code : https://lnkd.in/ddUjmXzx References : https://lnkd.in/d3-QtmJ9 #AI #MachineLearning #LLM #MCTS #Innovation #DatamindAI #AIResearch

1

Combining the power of TensorFlow, NeuralForecast, and zero-shot LLMs, Jinhang Jiang shows how we can generate accurate multi-step quantile forecasts.

11

LLMs have become more capable with better training and data. But they haven’t figured out how to “think” through problems at test-time. The latest research from Scale finds that simply scaling inference compute–meaning, giving models more time or attempts to solve a problem–is not effective because the attempts are not diverse enough from each other. 👉 Enter PlanSearch, a novel method for code generation that searches over high-level "plans" in natural language to encourage response diversity. PlanSearch enables the model to “think” through various strategies before generating code, making it more likely to solve the problem correctly. The Scale team tested PlanSearch on major coding benchmarks (HumanEval+, MBPP+, and LiveCodeBench) and found it consistently outperforms baselines, particularly in extended search scenarios. Overall performance improves by over 16% on LiveCodeBench from 60.6% to 77%. Here’s how it works: ✅ PlanSearch first generates high-level strategies, or "plans," in natural language before proceeding to code generation. ✅ These plans are then further broken down into structured observations and solution sketches, allowing for a wider exploration of possible solutions. This increases diversity, reducing the chance of the model recycling similar ideas. ✅ These plans are then combined before settling on the final idea and implementing the solution in code. Enabling LLMs to reason more deeply at inference time via search is one of the most exciting directions in AI right now. When PlanSearch is paired with filtering techniques—such as submitting only solutions that pass initial tests—we can get better results overall and achieve the top score of 77% with only 10 submission attempts. Big thanks to all collaborators on this paper including: Evan Wang, Hugh Zhang, Federico Cassano, Catherine Wu, Yunfeng Bai, William Song, Vaskar Nath, Ziwen H., Sean Hendryx, Summer Yue 👉 Read the full paper here: arxiv.org/abs/2409.03733

178

2 Comments

Unlock Your AI Potential with $50 Free Credit on Anyscale Discover the power of Anyscale, the premier cloud platform that can supercharge your AI journey. In this introductory video, you'll learn about the incredible features that make Anyscale a game-changer for AI development and research: YouTube URL: https://lnkd.in/g_vb4PWw 🎁 $50 Free Credit: Get started with Anyscale without hassle - no mobile verification or credit card required. Unlock your AI potential with this generous offer. 🔧 Integrated Visual Studio Code: Experience seamless development with Anyscale's built-in Visual Studio Code, enabling you to create, test, and deploy your AI projects effortlessly. 🧪 Playground for Open-Source Models: Explore and experiment with a wide range of open-source models on Anyscale's user-friendly platform. Leverage this powerful playground to accelerate your research and fine-tuning. 🖥️ GPU/CPU Resources: Anyscale provides access to advanced GPU and CPU resources, empowering you to tackle complex AI tasks and push the boundaries of your research. Join the Anyscale community and unlock a world of AI possibilities. Start your journey today with the $50 free credit and discover how Anyscale can transform your AI dreams into reality. Hugging Face Profile: https://lnkd.in/gEBS-M8W GitHub Profile: https://lnkd.in/g4k8tNGg #Anyscale #AI #MachineLearning #CloudPlatform #OpenSourceModels #FreeCredit #AIDevelopment #AIResearch

6

Learn how to build a neural network in TensorFlow — just follow along Shreya Rao's step-by-step guide (with a bonus section for anyone interested in a PyTorch implementation as well).

27