Muhammad Khan Pathan

Thanks for sharing this detailed update on Hermes 3! 🚀 The advancements in Hermes 3 are truly impressive. With its three model sizes (8B, 70B, and 405B parameters), it brings significant improvements across various dimensions: Roleplaying & Agentic Tasks: Enhanced capabilities for complex interactions and task management. 🎭🤖 Function Calling: Streamlined integration for executing commands effectively. 🛠️ Multi-Turn Chats: Improved engagement and coherence in extended conversations. 🔄🗣️ Context Coherence: Large context window for maintaining relevant and coherent dialogues. 📚 Training Highlights: SFT Phase: Utilizes standard instruct fine-tuning with Llama 3.1 models for foundational improvements. 📈 DPO Phase: Employs direct preference optimization for additional performance boosts. ✨ Hermes 3 stands out for its: State-of-the-Art Performance: Among the leading open models available. 🌟 Neutral Alignment: Ensures unbiased and balanced responses. ⚖️ Advanced Reasoning & Creativity: Enhances problem-solving and creative tasks. 🧠 Diverse Roleplaying: Adapts to different personas effectively. 🎭 The training process, including efficient sample packing and special considerations for the 405B model, underscores the sophistication behind Hermes 3. 🛠️ Explore more about Hermes 3 and its capabilities on their website. 📖 #AI #GenerativeAI #MachineLearning #Innovation #TechNews #LanguageModels #NousResearch

1

Pack just released a new book on the practical application of graph theory. It provides valuable explanations and practical examples of how graph-based algorithms implemented using Python solve industry problems, inspiring and motivating you to apply these techniques in your own work. It is an excellent book to expand your machine learning programming skills around network science. If you are interested in this book, please follow this link: https://lnkd.in/g2mD_CFY

30

5 Comments

Exciting update! Google has just released Gemma 2, their latest open AI models. It's all about top performance and efficiency, making advanced AI more accessible. Built for developers and researchers, Gemma 2 integrates seamlessly into your workflows with broad compatibility across major AI frameworks like Hugging Face Transformers, JAX, PyTorch, and TensorFlow. Ready to see what Gemma 2 can do for you? #AI #Gemma2 #TechNews

16

👀 Discover how GPU acceleration is transforming causal inference. Using RAPIDS cuML with DoubleML, you can tackle large datasets up to 12x faster. ➡️ https://nvda.ws/3YRM4eQ

161

1 Comment

Plotting Pore Throat Distributions in Python We've made significant progress in visualizing array data as wiggle traces in Python, and Craig Lindsay had the brilliant idea to apply this to Pore Throat Distributions (PTD). Using the advanced visualization techniques from the Rougier GitHub repository: https://lnkd.in/d6AgRD25 we’ve taken our Python plotting capabilities to the next level. While we don’t have direct High-Pressure Mercury Injection (HPMI) capillary pressure data, we do have Thomeer parameters calibrated to HPMI data to generate capillary pressure curves and corresponding PTDs. Please view the last track in the attached example to see the Pore Throat Distributions in a depth plot. We’re happy to share the Python code for these plots, along with our complete carbonate reservoir characterization workflow. It’s available now on our Open Source Petrophysical GitHub repository at the following link: https://lnkd.in/ejYxRZn8

309

28 Comments

Are you struggling to minimise the number of classification errors in your machine learning model? 💻 Optimise Your Classification Model with these Loss Functions! Let's break it down: 0-1 Loss: A non-convex function that aims to minimise errors, but is hard to optimize. Convex Approximation: A smoother alternative that's easier to minimise, but still has degenerate solutions. Hinge Loss: A convex function that prevents degenerate solutions, but is not differentiable. Logistic Loss: The smooth, differentiable solution that combines the best of both worlds. It's a game-changer for optimising your classification model! Note however, this may not be the best optimiser and may not give the widest margin between the opposing classes even in binary classification so be careful and be sure to try different available options. Dive deeper into the world of loss functions and learn how to optimise your model for better performance. Share your thoughts and experiences in the comments! #machinelearning #lossfunctions #classification

5

This is an important point. We as data scientists often take code efficiency for granted, assuming our "to(cuda)" or parallel processors will handle it for us. That code might be fast, but is it efficient? No. This slide from a course I am currently pursuing (CS149 by Stanford Online) opens up a clear perspective about what it means to have an efficient code. #HPC #AI

17

2 Comments

How to Build a Custom Optimizer in PyTorch: 5 Simple Steps via #TowardsAI → https://bit.ly/3Y2UXTB

1

How to Build a Custom Optimizer in PyTorch: 5 Simple Steps via #TowardsAI → https://bit.ly/3Y2UXTB

2

If you use Seaborn for plotting with Python, check out my popular article on the new Seaborn Objects System. Based on the "Grammar of Graphics" paradigm, it makes Seaborn plotting more flexible, modular, and intuitive. #pythonprogramming #seaborn #datavisualization

3

🔺 A little gotcha in #python Or maybe an artifact of #pytorch (because the logic is same, outputs are differently shaped) Suppose you have a tensor of the shape T =: (Batch, X, Y, Channels) If you slice the last channel-layer, Approach 1 : T[ : , : , : , -1] Output Shape : (Batch, X, Y) Approach 2 : T[: , : , : , -1:] Output Shape : (Batch, X, Y, 1) While it is a common norm in python that -1 is the last index, it makes no logical sense to interpret -1: as anything but -1 itself. That additional dimension might break codes downstream. It might go down as a difficult bug to find in your deep-neural-architectures! Happy weekend. Happy Learning! (Edit, second print statement: Approach 2*)

18

2 Comments

Learn how you can convert a PyTorch model, trained on the web, to a mobile-optimized format. Madhur Zanwar describes the process using a few lines of code. He also discusses his journey and how one approach didn't yield satisfactory results while trying to run it in a React Native app. #machinelearning #pytorch #computervision #edgeml #reactnative #objectdetection #yolov5

14

📝 The release of Meta's Llama 2 model last year prompted a wave of excitement over 'open' AI models, with a host of major companies having launched their own since, including Mistral and Databricks. 🔎 But when it comes to AI models, there's a stark difference between 'open' and 'open source', writes ITPro's George Fitzmaurice. 🧠 George caught up with Amanda Brock, CEO at OpenUK, who said that while open AI development should be welcomed, it's about time firms started making things clear for users. ❌ Many 'open' LLMs, including Meta's new Llama 3 model, come with restrictions that are shrouded amidst the hype of a new launch. 👀 Read more here #AI | #OpenSource | #LLMs

1

🛠️ LangChain State of AI 2024: Trends in LLM Application Development 🤖📊 The 2024 LangChain State of AI Report offers valuable insights into how LLM development practices are evolving, reflecting both technological advancements and shifting developer priorities. Two standout trends dominate this year: the rise of multi-step agentic workflows and the growing reliance on open-source models for enhanced customization and scalability. 🔄 1. Transition from Retrieval-Only to Multi-Agent Workflows: The shift from retrieval-augmented generation (RAG)-centric designs to multi-step, agentic applications signals a maturity in the AI ecosystem. Developers are building modular agent architectures where agents perform specialized tasks (e.g., data validation, iterative refinement, and decision analysis). This enables workflow resilience, scalability, and adaptability across complex use cases. 🧠 2. Agentic Reasoning and Orchestration: Modern agent frameworks increasingly incorporate state management, memory architectures, and self-correcting loops. Tools like LangGraph are becoming central to designing cyclical workflows where agents dynamically adapt based on intermediate outputs, significantly improving reasoning and problem-solving capabilities. ⚙️ 3. Open-Source Model Adoption: The report highlights a clear trend towards open-source LLMs (e.g., LLaMA, Falcon, Mistral) as organizations prioritize customization, privacy, and deployment flexibility. Open-source models allow fine-tuning on domain-specific datasets and reduce vendor lock-in, enabling cost-effective scalability in production-grade applications. 📊 4. Observability and Monitoring with LangSmith: With the growing complexity of multi-agent workflows, observability tools like LangSmith have become critical for monitoring agent behavior, tracing decisions, and identifying optimization opportunities. Real-time telemetry ensures that workflows remain transparent, auditable, and efficient at scale. 🚀 5. Modular Design Patterns: Modern LLM applications favor modular designs with distinct components for retrieval, reasoning, validation, and orchestration. This modularity not only improves developer productivity but also facilitates rapid experimentation and deployment across different use cases without overhauling core architectures. 🔍 Key Takeaway: The 2024 report underscores the importance of building adaptable, modular, and transparent AI systems. The convergence of agentic workflows, open-source adoption, and robust observability practices will continue to drive innovation and scalability in LLM applications. Thanks for sharing this report! It’s exciting to see how the ecosystem is evolving towards intelligent, agent-driven architectures while maintaining a strong emphasis on developer tooling and infrastructure optimization. 💡 #LangChain #AgenticAI #LLM #AIWorkflows #MachineLearning #AIEngineering #OpenSourceAI #AIObservability

2

I wrote a comment piece for The Economic Times discussing OpenAI's new o1 series of foundation models and the implications the progression of this technology has for reasoning based tasks like software development, along with a discussion of the limitations of these models. 'Hello o1, need some human help?' 'OpenAI's new o1 series of foundation models enhance ChatGPT's problem-solving skills with advanced chain-of-thought reasoning, excelling in complex tasks like coding and portfolio optimization. This progress suggests a shift towards fewer human developers, with GenAI handling simpler tasks, and emphasizes the need for ethical oversight. o1 models are progressive, using more advanced chain-of-thought reasoning than previous foundation models like OpenAI's GPT-4, Anthropic's Claude and Google's Gemini. Their step-by-step problem-solving, learning from mistakes, takes longer. But it allows them to tackle more complex tasks than earlier benchmarks.' Read more: https://lnkd.in/ekr5HsUc

61

7 Comments

GPTs are good for wrangling and summarizing data, giving you the power of a programming language like Python with a VERY beginner friendly interface. However, the element that's hamstringing this approach at the moment is the limited amount of RAM that GPT Python sessions have to work with (around 1-1.8 GB). OpenAI please give us the option to scale up the allowable RAM! That will unlock a lot of new use cases within your existing tool.

6

I’ve never thought of Data Science as just training ML models - to me it’s always been so much more. For me, a broad description would be: “Using data to help businesses make better strategic decisions, to understand their customers, and to predict what is likely to happen next” There is so much value in this, and it will always be in high demand - regardless of what tools help you do it.

90

8 Comments