CMND.ai’s Post

🌟 Exciting News! 🌟 I’m thrilled to share that I’ve published a new blog post about Large Language Models with real-time capabilities, featuring my #LLM application, #QuikNews. If you're passionate about GenAI or seeking insightful reading, this article is for you! Feel free to comment or share it with your network if you find it valuable. #MachineLearning #GenAI #LangChain #DataScience #Innovation #Linkedinpost

Ai2 recently unveiled #OLMoE, a cutting-edge Mixture-of-Experts model boasting 1.3 billion active parameters and a total of 6.9 billion parameters. Trained on a massive dataset of 5 trillion data-curated tokens, this model sets a new standard in performance within its parameter range, surpassing all existing open models. Noteworthy is OLMoE's exceptional adaptability to fine-tuning, showcasing substantial enhancements with advanced optimization techniques like KTO and DPO. The release of OLMoE also includes crucial features such as intermediate training checkpoints, enhanced post-training mix, code, and training logs, all available under the Apache 2.0 license. #AI #MachineLearning #Innovation

pub.towardsai.net: The article discusses the perception of large language models (LLM) as highly intelligent databases containing vast internet knowledge. It highlights their readiness to assist with various tasks and provides a deeper insight into their functionality.

Hello community! please let me share my latest article for LLM #Orchestration. LiteLLM 🚆: The Orchestrator You Need for Your LLM Application 🔑 Key Highlights: • Unified interface for all LLMs providers • Docker-based setup for isolated configurations • Step-by-step guide on registering and using models • OpenAI compatible API • Security and compliance features 🔗 Read the full article here: https://lnkd.in/dtJdbW33 Let's discuss! Have you used LiteLLM (YC W23)? What's your experience with integrating multiple LLMs in your projects? #AI #MachineLearning #LLM #TechInnovation #SoftwareDevelopment #LiteLLM Ishaan Jaffer

One more challenge in building scalable agents: as the number of tools and APIs available to an agent increases, it becomes more complex to plan the execution path accurately. This concept is like predicting the next API/Tool, similar to predicting the next token. Andrej Karpathy highlighted that an LLM can be considered a next token predictor system in one of his tweets. Language models serve as a type of text token predictor. Discover ToolGen, a paper implementing this idea to predict the next API/Tool token by embedding the APIs in the model's vocabulary. #Agents #ToolGen

Enrich your applications with LLM's function calling. Function calling is one of the capabilities of LLM's to build deep integrations with external applications. The spectrum of possibilities starts with simple real-time data access to more complex tasks like controlling physical devices via natural language. Here are some best practices to make function call more reliable. 1️⃣ Use strict structured outputs to make the model generate arguments that stick to the function schema. 2️⃣ Name your functions intuitively with detailed descriptions. This lets the model to select the right function when there are more functions in your tools. 3️⃣ Use clear & descriptive parameter names for your functions. 4️⃣ Use enums, bool for function arguments where possible. 5️⃣ Use system prompt to provide instructions on when and which function to call. 6️⃣ Use function definition to provide instructions on how to call and generate parameters. 7️⃣ Keep the number of functions as low as possible. More the number of functions the less the accuracy of picking the right function. #generativeai #genai #llm #openai

We’re excited to launch C4AI Command-R, a 35 billion parameter model weights release. C4AI Command-R offers researchers and developers access to a leading-edge frontier model to experiment, explore, and build on for non-commercial purposes, including fundamental research and safety auditing. C4AI Command-R is a generative large language model with released weights featuring leading performance on tasks including reasoning, summarization, and question-answering. It has been optimized for multilingual generation in 10 languages, tool-use, and has highly performant RAG capabilities. Many researchers and communities have been left unsupported due to a lack of access to high-performing models. We’re releasing C4AI Command-R as part of our commitment to bridging the growing gap in access to frontier technology for fundamental research and safety auditing. C4AI Command-R is now available to support the research community in solving complex machine learning problems through fundamental research that explores the unknown.  https://lnkd.in/gMHNujPQ

Yesterday, I had the pleasure of participating in the first MeetUp in Paris focusing on RAG. RAG and related concepts will make LLM even more useful going forward. Here are some key takeaways from this fast-moving topic. Joffrey THOMAS (Hugging Face) presented fascinating insights into open-source vs proprietary models and the advantages of using the former. However, in the Top 10 chat models leaderboard (https://lnkd.in/eVzNfmEi), there is just one open-source model Qwen1.5 (by Alibaba Group). My take: More open-source models will be highly ranked, especially with the release of Databricks' new open-source model DBRX (released on March 26th, 2024) "Introducing DBRX: A New State-of-the-Art Open LLM" by the Databricks Mosaic Research Research Team at Databricks. The need is there, the techno is out. Maxime Voisin (Cohere) hit the nail on the head by explaining that RAG is part of Tools and part of Agents. In this era of Compound AI Systems, it can't get any clearer. Pierre-Loic Doulcet (LlamaIndex) not only walked us through the intrinsic power of LlamaIndex but also presented the latest research topics such as RAFT (Retrieval Augmented Fine Tuning https://lnkd.in/eei9jCRN ) that will push the boundaries even further. This sounds soo energy-vore 🤔 . #RAG #LLM #AI #OpenSource #Databricks #Cohere #HuggingFace #LlamaIndex #CompoundAISystems #RAFT #RetrievalAugmentedFineTuning #DBRX #Qwen

🟢 The team at Lime developed a unique process called Refine/Retrieve/Rebuttal for their RAG flow that was designed to surmount practical challenges they encountered—like dealing with jokes, contextual integration, language detection, hallucinations, and more. ... Unfortunately, the process didn't work as well as expected. 😬 When mechanics typed in an error code, the chatbot would sometimes return information for a different code, and vehicle repair instructions were often inaccurate. 🤔 Learn how Log10 offered an end-to-end LLM Ops platform and, with just one line of integration, was able to get the Lime team up and running with debugging. 👉 https://bit.ly/4da8nCI #LLMOps #AI #RAG

GEN-AI post - 1/50 I’ve been experimenting using open source local LLM and Langchain to query PDFs efficiently using local LLM's (large language models). By leveraging the powerful language models from Ollama and Langchain’s I've been able to extract and query information from PDFs. Whether you’re dealing with large documents, complex data, or simply need to retrieve specific information from a PDF, this solution opens up new possibilities for working with text-based data. Looking forward to diving deeper into this space! 🤖✨ #AI #MachineLearning #LangChain #Ollama #PDFQuery #DataScience #Automation #Innovation github link :- https://lnkd.in/e-p9a7_i