Gopi Subramanian’s Post

Excited to share an extensive benchmark of C and C hash tables. This insightful post evaluates the performance of various hash tables and provides valuable insights for developers. Check out the full analysis here: https://ift.tt/CtXokhW #coding #algorithm #development

LightLLM is a Python-based LLM (Large Language Model) inference and serving framework, notable for its lightweight design, easy scalability, and high-speed performance. LightLLM harnesses the strengths of numerous well-regarded open-source implementations, including but not limited to FasterTransformer, TGI, vLLM, and FlashAttention. #LightLLM #LargeLanguageModels #LLMInference #PythonFramework #NaturalLanguageProcessing #AIEngineering #ScalableAI #LanguageModelServing #OpenSourceAI #MachineLearningOptimization https://lnkd.in/gCyzuYAn

DeepSeek V3 - new open model that is better than Claude Sonnet at the Aider Polyglot coding benchmark and - perhaps most importantly - has 10x cheaper inference costs. The massive price drop means you can experiment with having a Aider build you a feature for $2 instead of $20. $20 is a little too pricey for rapid experimentation in your development workflow. Unsurprisingly, Deep Seek's site is down. Other Model Inference providers like Hyperbolic seem to have also taken Deep Seek V3 down for maintenance https://lnkd.in/gtn8Hb36

OpenAI o1 Sucks at Coding! While its slow responses continue to irk developers, the issue goes beyond response time. A developer wrote on Hacker News that the o1-preview model was hallucinating to the point where it started responding in the context of non-existing libraries and functions. When AIM compared multiple LLMs for coding completion tests on LiveBench, the results were shocking as o1-mini was ranked below the open-source model Qwen2-72B and GPT-4. “The model sometimes gets stuck in thinking mode and never returns a response—happening about 40% of the time. It acts like it’s done processing, but the answer never comes - it’s often just a blank reply or just a few characters,” said Mike Young while reviewing the o1 model for programming. Even if we ignore the use of more tokens and delay in response, the reasoning which is the pro feature of o1 models, still generates buggy code. Link to full story in comments.

💡 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐭𝐮𝐧𝐢𝐧𝐠 𝐢𝐧 .𝐍𝐄𝐓 & 𝐂#: 𝐀𝐠𝐠𝐫𝐞𝐬𝐬𝐢𝐯𝐞 𝐈𝐧𝐥𝐢𝐧𝐢𝐧𝐠! 🚀 Have you often stumbled across the [Aggressive Inlining] attribute in C#? What does it actually do - and why is the performance potential so great? Find out how you can: ✅ Make your applications even faster ✅ Make optimum use of the JIT compiler ✅ And avoid typical performance pitfalls https://lnkd.in/en3fxQPx #csharp #dotnet #mvpbuzz #sustainablecode #medialesson

I have used llama 3 model on groq, using groq api key. it was little tedious to work, since I am new to python . getting api key to the right variable depends on ide environment, and it's different for all of them. my personal preference is jupyter notebook python development environment. next to that is vs code, and distance 3rd is pycharm this article uses codegpt plugin to vs code ide. it seems it makes thing real simple. https://lnkd.in/gr2BS9Ks

🌟 Day 70 of the 75-day LeetCoding Challenge! 🎉 🚀 Problem: 945. Minimum Increment to Make Array Unique 🔍 Level Description : Medium ⚡Link to Problem : https://lnkd.in/gRCqzh_q 💡Approach || 🕒 O(nlogn) || 💾O(1) --Algorithm: 1.Sort the Array: Sort the input array nums in non-decreasing order. 2.Initialize Moves Counter: Set a counter moves to 0. This will keep track of the total number of increments needed. 3.Iterate Through the Array: Starting from the second element (index 1) to the end of the array, do the following: Check for Duplicates or Lower Values: If the current element nums[i] is less than or equal to the previous element nums[i-1]: Calculate Required Increments: Compute the difference needed to make nums[i] greater than nums[i-1] by at least 1, which is (nums[i-1] - nums[i] + 1). Update Moves Counter: Add this difference to moves. Update Current Element: Set nums[i] to nums[i-1] + 1 to make it unique. 4.Return Result: Return the value of moves which represents the total number of increments needed. __________________________________________________________________ #75daysCodeChalleng #leetcode #algorithm #datastructures #codingjourney #linkedincodingcommunity #coding #codingchallenge

🦜 𝗟𝗟𝗠-𝗣𝗼𝘄𝗲𝗿𝗲𝗱 𝗔𝗽𝗽 𝘄𝗶𝘁𝗵 𝗟𝗮𝗻𝗴𝗖𝗵𝗮𝗶𝗻 𝗮𝗻𝗱 𝗦𝘁𝗿𝗲𝗮𝗺𝗹𝗶𝘁 𝗶𝗻 𝗝𝘂𝘀𝘁 𝟭𝟴 𝗟𝗶𝗻𝗲𝘀 𝗼𝗳 𝗖𝗼𝗱𝗲 LangChain is a powerful framework designed to harness the capabilities of Large Language Models (LLMs) for building versatile applications. LangChain’s strength lies in its six core modules: • 𝗠𝗼𝗱𝗲𝗹 𝗜/𝗢: Seamlessly integrates model inputs (prompts) with the LLM, producing outputs through advanced parsing techniques. • 𝗗𝗮𝘁𝗮 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗼𝗻: Handles loading, transforming, storing, and querying of user data with tools like document loaders, transformers, embedding models, and vector stores. • 𝗠𝗲𝗺𝗼𝗿𝘆: Equips chains or agents with both short-term and long-term memory, enabling them to recall previous interactions. • 𝗖𝗵𝗮𝗶𝗻𝘀: Allows you to combine multiple components or chains into a single, cohesive pipeline. • 𝗔𝗴𝗲𝗻𝘁𝘀: Makes intelligent decisions on what actions to take based on the input, utilizing available tools and data. • 𝗖𝗮𝗹𝗹𝗯𝗮𝗰𝗸𝘀: Executes specific functions at key points during an LLM run. Streamlit is a leading Python library for building web apps in machine learning and data science. Now, you can create a basic LLM-powered app using LangChain and Streamlit in just 18 lines of code. Check out the code link in the comments! #LLM #LangChain #Streamlit #Python #MachineLearning #DataScience #AI #WebApp #Coding #TechInnovation #OpenSource #ArtificialIntelligence #Programming #ML #DataEngineering

For the last 4 days, I've noticed that most of my hard drives are bloated with Python environments; taking most of the space to the #transformers and #pytorch related libraries; usually due to the fact that each repo uses the different libraries with different versions. I also felt like, we are at a point in which the software stack for #AI is so well established that #ai is becoming a #systems #engineering problem; rather than a problem to be solved with #maths (there is still a long way to go in maths, but the sys engineer need appeared recently with the rise of bigger models taking around hundreds of gigabytes to terabytes of size). The need for solutions that run LLMs fast in a tiny manner without tons of dependencies and bloated software; has been acknowledged since the beginning of #chatgpt by the open source community: the great #ggml and #llamacpp libraries have given rise to the possibility of running LLMs at top speed in consumer hardware; supporting pretty much any kind of #gpu and device. Somehow, if you try to run #llamacpp, #ollama or similar frameworks in tiny hardware like #orangepi zero or #android devices, you'll hit a wall: errors and difficulties during the compilation process, a kind of big codebase that has been integrated with multiple additions by the community... making it barely impossible to run easily. So that, I've felt the need to go a bit lower, and take the work from Andrej Karpathy and others; be influenced by them, obsess myself with understanding every bit of it, and trying to expand onto other models. By tinkering with different repos, soon it became hard to keep track of the different versions that I was testing; thus, I created a new repo called "neurons.c"; meant to be a place where all repos that implement #llms in a tiny manner in #C are collected. The plans are to have a support for all the model architectures, and mimic the path of #llamacpp, trying to trade off performance and code reuse for simplicity. Looking for that: model inference implementation in a single ".c file" that we can extend on. Check it out: https://lnkd.in/dGGx4_rK

AI agent just cracked the code for internal APIs. Integuru, a powerful AI agent reverse-engineers internal APIs, builds a dependency graph for requests, figures out dynamic parameters, and generates runnable Python code. And it's 100% opensource. https://lnkd.in/gKctuK7w Great share Eric Vyacheslav