Show HN: Define and implement any function on the fly with LLMs It's not a serious library, but I figured it could be something fun to make. https://ift.tt/IFn5fpO April 30, 2024 at 04:29AM
Show HN: Attorch – PyTorch's nn module written in Python using OpenAI's Triton attorch is a subset of PyTorch's nn module, written purely in Python using OpenAI's Triton. Its goal is to be an easily hackable, self-contained, and readable collection of neural network modules whilst maintaining or improving upon the efficiency of PyTorch. In other words, it intends to be a forkable project endowed with a simple, intuitive design that can serve as an accessible starting point for those who are seeking to develop custom deep learning operations but are not satisfied with the speed of a pure PyTorch implementation and do not have the technical expertise or resources to write CUDA kernels. There already exist a number of wonderful PyTorch-like frameworks powered by Triton, but most concentrate solely on Transformers and NLP applications, whereas attorch aims to be more inclusive by also presenting a variety of layers pertaining to areas besides NLP such as computer vision. Moreover, attorch is not an inference-only package and fully supports both forward and backward passes, meaning it can be used during training as well as inference, though its performance for the latter is generally not on par with dedicated inference engines. Questions and feedback are welcome in the comments sections. https://ift.tt/tJ6OyN4 April 30, 2024 at 01:07AM
Show HN: Kaytu – Optimizing cloud costs using actual usage data Reduce your cloud costs - SREs/DevOps/Cloud Engineers Hi community! We are Kaytu (“Kay-two,” named after the K2 mountain), and we've developed an open-source tool for engineering, DevOps, and SRE teams to reduce cloud costs. Cloud inflation (“cloud-flation”) is real—AWS EC2 costs are up 23% (4-5x global inflation average [1]), and 30% of the capacity that is paid for is simply wasted ([2]). The best way to improve cloud utilization is by simplifying the process so engineers can spot inefficiencies and suggest changes. We built a simple open-source CLI tool that recommends a cost-optimal workload based on actual usage data from observability tools. Check it out at https://ift.tt/vdRLJYy Currently, we support AWS EC2 On-Demand Servers & EBS Storage using observability data from CloudWatch to determine utilization. You can optimize EC2 Servers based on CPU, Network, Memory, and Storage. We're expanding support to include OS License, GPU metrics, RDS, and Prometheus integration, and we plan to add more AWS services like EKS and OpenSearch, as well as Azure. This is more than just a utility—we want to provide a no-nonsense platform that makes it ridiculously easy for engineers to build cost-effective apps on the cloud by optimizing workload configurations and customizing to scenarios. Open Core: Inspired by Sid Sijbrandij and GitLab, we've open-sourced our CLI and are actively working on the server side. Our tooling will always remain straightforward and support open-source tools for free. We made it as simple as possible to try out - it’s one command, no sign-up needed, no SaaS platform to share your credentials. We would love you to try it out and give us your feedback! If there are bugs, we would greatly appreciate it if you reported them on GitHub. Cheers, The Kaytu Team (Anil, Arta, Mahan, and Saleh) References: [1]Tangoe IT Trends Savings Recommendations and Liftr Insights data Cloud Pricing [2] Flexera State of Cloud Report - Multiple reports spanning 2017-2023 https://ift.tt/vdRLJYy April 29, 2024 at 09:27PM
Show HN: Dotenv, if it is a Unix utility I like the idea of using dotenv files, but I dislike having to use different language-specific libraries to read them. To solve this, I created a small utility that lets you prefix any command with "dotenv" to load the ".env" file. This is how I imagine dotenv would work if it had started as a UNIX utility rather than a Node.js library. https://ift.tt/AoKapgn April 29, 2024 at 01:55AM
Show HN: OpenLIT – Open-Source LLM Observability with OpenTelemetry Hey HN, we're super excited to share something we've been working on: OpenLIT. After an engaging preview that some of you might recall, we are now proudly announcing our first stable release! *What's OpenLIT?* Simply put, OpenLIT is an open-source tool designed to make monitoring your Large Language Model (LLM) applications straightforward. It’s built on OpenTelemetry, aiming to reduce the complexities that come with observing the behavior and usage of your LLM stack. *Beyond Basic Text Generation:* OpenLIT isn’t restricted to just text and chatbot outputs. It now includes automatic monitoring capabilities for GPT-4 Vision, DALL·E, and OpenAI Audio. Essentially, we're prepared to assist you with your multi-modal LLM projects all through a single platform and we're not stopping here; more updates and model support are on their way! *Key Features:* - *Instant Alerts:* Offers immediate insights on cost & token usage, in-depth usage analysis, and latency metrics. - *Comprehensive Coverage:* Supports a range of LLM Providers, Vector DBs, and Frameworks - everything from OpenAI and AnthropicAI to ChromaDB, Pinecone, and LangChain. - *Aligned with Standards:* OpenLIT follows the OpenTelemetry Semantic Conventions for GenAI, ensuring your monitoring efforts meet the community's best practices. *Wide Integration Compatibility:* For those already utilizing observability tools, OpenLIT integrates with various telemetry destinations, including OpenTelemetry Collector, Jaeger, Grafana Cloud, and more, expanding your data’s reach and utility. *Getting Started:* Check our quickstart guide and explore how OpenLIT can enhance your LLM project monitoring: https://ift.tt/SFQETeJ We genuinely believe OpenLIT can change the game in how LLM projects are monitored and managed. Feedback from this community could be invaluable as we continue to improve and expand. So, if you have thoughts, suggestions, or questions, we’re all ears. Let’s push the boundaries of LLM observability together. Check out OpenLIT here: https://ift.tt/SyAajCn Thanks for checking it out! https://ift.tt/SyAajCn April 26, 2024 at 03:15PM
Show HN: Scenestamps – A website for sharing movie scenes with timestamps Hello hackers, I've launched a website specifically for sharing scenes, complete with descriptions and timestamps from various films and TV shows. I'm reaching out to gather your perspectives and recommendations in these domains to improve the site and extend my outreach. Link : https://scenestamps.com You are not required to register/login to browse the site. Scenestamps Features: 1. Search : You can directly search for a scene or a source. 2. Upload Posts : You can register with your google account, login and start posting right away. Unlike other sites in this specific domain, users are allowed to upload posts. There are two types of posts - scene - source Source is a movie,tv show, documentary, etc... One source can have multiple scenes While creating a scene post, source can be selected there. Scene post will have the timestamp fields. There are two types of it: - single - one input field of timestamp in which the scene happens. - from-to - two input fields, from and to within which the scene takes place. 3. Share posts : Share feature is available on both source and scene posts, with which you can share the post to your favorite social media platforms 4. Tagging system : You can add tag to the scene posts. You can also get all the scenes that has that tag name by clicking on it. I think people wanting to create scenes is quite a small audience, but I want to make this the best it can possibly be so please post any problems or suggestions in the replies or at reddit.com/scenestamps.com or message me x.com/gjpx_ if you prefer. April 26, 2024 at 12:10PM
Show HN: Htpy – generate HTML from Python without templates I built a library that to generate HTML from Python. We have been using this library with Django the last couple of months instead of classic templates and find it to be productive. It is easy to debug, works great with static type checkers and it is easy to build reusable components/partials. Give it a try! https://htpy.dev April 28, 2024 at 01:04AM
Show HN: I want my family to listen to more music(less movies) I decided to use the home entertainment system to take over the TV. Limited visual stimulation. Front Row is a collection of live musical performances gathered by music fans of all kinds. https://thefrontrow.app Download for iOS or Apple TV https://ift.tt/RUvhoQT... And fuck it. I'm letting anyone add videos directly to the database! https://ift.tt/v5Mf3p9 April 27, 2024 at 11:02PM
Show HN: Cognita – open-source RAG framework for modular applications Hey HN, exciting news! Our RAG framework, Cognita ( https://ift.tt/6iKL9EH ), born from collaborations with diverse enterprises, is now open-source. Currently, it offers seamless integrations with Qdrant and SingleStore. In recent weeks, numerous engineers have explored Cognita, providing invaluable insights and feedback. We deeply appreciate your input and encourage ongoing dialogue (share your thoughts in the comments – let's keep this ‘open source’). While RAG is undoubtedly powerful, the process of building a functional application with it can feel overwhelming. From selecting the right AI models to organizing data effectively, there's a lot to navigate. While tools like LangChain and LlamaIndex simplify prototyping, an accessible, ready-to-use open-source RAG template with modular support is still missing. That's where Cognita comes in. Key benefits of Cognita: 1. Central repository for parsers, loaders, embedders, and retrievers. 2. User-friendly UI empowers non-technical users to upload documents and engage in Q&A. 3. Fully API-driven for seamless integration with other systems. We invite you to explore Cognita and share your feedback as we refine and expand its capabilities. Interested in contributing? Join the journey at https://ift.tt/Qz67g2N . https://ift.tt/6iKL9EH April 27, 2024 at 10:10PM
Show HN: I made a spaced repetition tool to master coding problems As you solve LeetCode questions, you can mark them as hard, medium, or easy. The tool will then recommend questions you should review based on (1) how hard the question was for you and (2) how much time has passed since you last reviewed it. I'd recommend normally attempting LeetCode problems and just marking them as hard, medium, or easy for you at first so the tool knows which problems to recommend you review! Here's the theory behind spaced repetition and learning if interested: https://ift.tt/RH8hCtz https://www.lanki.xyz/ April 27, 2024 at 01:08AM
Show HN: I made a site for practicing front end debugging with real-world bugs I am excited to introduce my solo project, a platform built with the frontend development community in mind. It's an interactive environment aimed at refining debugging skills through exposure to real-world bugs. This project stems from the need for a hands-on, practical method of learning to debug. The site features a variety of intentional bugs for users to solve, mirroring the types of challenges faced in professional settings. Over the coming weeks, I will be adding many more exercises and new features to enhance your learning experience further. Check out CodeMender at https://ift.tt/Lu3TEay https://ift.tt/Lu3TEay April 26, 2024 at 11:29PM
Show HN: I put PubMed in a vector DB Hi HN, As a researcher, I often found myself struggling with the limitations of keyword-based search when exploring PubMed papers. To address this, I created PubMed Search ( https://ift.tt/VIZKYLw ), a tool that leverages a vector database to enable semantic search across medical research literature. Some key features: * Daily updates to ensure access to the latest articles * Semantic search using latest & greatest embedding models * Some additional useful info about the papers (tldr, journal, publication date, etc.) Hope you find it useful! https://ift.tt/VIZKYLw April 26, 2024 at 01:23AM
Show HN: I made a programmable computer from NAND gates I am proud to present my solo hobby project NAND. This year-long undertaking follows the completed Nand to Tetris course, but ported to the web with its own runtime, user interface, and IDE. Using the "Load example program" selector, you can try out some programs I wrote on NAND's emulated hardware such as 2048, a genetic algorithm, and a manual stack overflow to corrupt the screen. Check out NAND at https://nand.arhan.sh Additionally, I've authored an extensive writeup about the project. Read about it on the GitHub repository's readme. https://ift.tt/cufyWtj April 25, 2024 at 09:38PM
Show HN: Sound AiSleep – create kids audiobooks in their parents voice using AI You can use the promo code: FREEBOOK2024 to test an audiobook for free. We created an app that uses AI voice clones, to clone parents voice. They then choose any book from our kids library to create their audiobook spoken in their voice. Would love to hear any feedback! https://ift.tt/khsWGMC April 24, 2024 at 12:54PM
Show HN: Storing Private Keys in the Browser Securely So the main purpose here is to show _a_ way that session-token theft can be mitigated. Clearly, this isn't NSA proof or something you'd use to secure a BL5 containment facility, but to prevent session-jacking; if feels like it could help a lot, and would be pretty quick and easy to roll out if an IDP wanted to implement it. https://ift.tt/PqGF5Md April 24, 2024 at 12:32AM
Show HN: Metashade – a Pythonic GPU shading/compute EDSL Superficially, it may look similar to Nvidia's Warp or OpenAI's Triton, but instead of transpiling a subset of Python to the target language, it implements dynamic codegen in pure Python. These slides discuss the existing GPU programming solutions and make a case for Metashade's approach: https://ift.tt/wQyT91z... https://ift.tt/7Vk3xsL April 22, 2024 at 04:51AM
Show HN: Auto-optimizing deterministic LLM outputs using knowledge graphs Hi, We are building an open-source framework for loading and structuring LLM context to create accurate and explainable LLM answers using knowledge graphs and vector stores. We built the tool with four main concepts in mind: 1. Loader -> uses dlt in the backend to load and structure the data 2. Cognify step -> creates a graph with summaries, labels and factoids that are interconnected across the documents and stored as a representation in the vector store 3. Optimizer -> Uses DSPy to optimize LLM queries, and we plan to extend it to most of the knobs we can turn, like chunking etc. 4. Search -> allows for searching using search types supported in graph stores (ex. Neo4j) or hybrid, BM25, or other search types available in vector stores. We are quite early with the product but we would love to hear feedback on what we can improve. https://ift.tt/XaUI0nm April 22, 2024 at 11:55PM
Show HN: AppView 1.0.0 is released Instrument, Observe, Secure your deployments AppView is an open source, runtime-agnostic instrumentation utility for any Linux command or application. It helps users to explore, understand, and gain visibility with no code modification. With one instrumentation approach for all runtimes, AppView offers ubiquitous, unified instrumentation of any unmodified Linux executable. With AppView 1.0.0 comes the new threat detection logic that allows users to capture security-related events. Is your application accessing secure files? Is it making connections it shouldn't be? Is it exfiltrating data over DNS or is GOT poisoning in effect? Other features of the 1.0.0 release include the ability to: - Generate metrics on process and application performance. - Generate events, reporting on network, file, logs, console messages and http/s activity. - Capture (decrypted) payload data without the need for keys. - Generate a stack trace, and a core dump when an application crashes. - Generate network flow information. - Create a report on unique file and network activity. - Install AppView in a Kubernetes cluster. - Secure file and network access in an application. - Instrument both static and dynamic executables. - Attach to processes while they are running or start when the process does. - Normalize and forward metrics and events, in real time, to remote systems. - Summarize metrics and detect protocols. We are looking for users and contributors alike. https://appview.org/ April 21, 2024 at 10:52PM
Show HN: What Are You Working On? Hey HN, I'm sure you've seen the monthly "Ask HN: What Are You Working On?" headlines on [Hacker News]( https://ift.tt/W1ywtGR... ). Honestly, it's my favorite topic because it's packed with insights about what other hackers are up to. I wondered what it would be like if instead of just a headline, there was a whole website where hackers could post daily updates, and where we could follow the hackers we're interested in for their latest updates. And so, this web site was born. I hope it gets used frequently so we can all benefit from it together. I look forward to hearing your thoughts. Let me know what you think! https://ift.tt/3mB7r5R April 21, 2024 at 05:13PM
Show HN: I built a small utility that handles multiple browser instances for you I'm a consultant and I normally deal with multiple clients. I need an isolated browser for each to maintain my sanity. I could create separate profiles in Chrome and Firefox but the main issue is when you click on a link outside of the browser (e.g. in your IDE), in which profile should it be opened? I wrote a Python script that redirects URLs to the correct browser instance and set it as my default browser. Over time, the script has grown and become essential to my work. I cleaned it up and open-sourced it. Hope that someone else may find it useful too! https://ift.tt/WiNy0HE April 20, 2024 at 06:33AM
Show HN: Talk to Me Human – my game about social persuasion Hey all, I recently graduated from a good PhD program studying NLP. Unlike any sane person who would go become a professor or make a gazillion dollars in industry, I decided to try bootstrapping my own software business. This is my first product. The inspiration was from my research on computers understanding social norms. When ChatGPT came out, I was amazed how well it could understand social etiquette. I thought it'd be fun to make a game where you have to talk your way out of sticky situations - like you miss your friend's birthday party, or your boss catches you trying to leave work at 2pm. I made a prototype in a couple days, and it was super fun to play with. I thought I'd spend a "couple months" making a game for others to play online. Now, only 10 months and 923.3 hours of work later, it's playable in early access. In the game, you talk out loud (ASR), and the NPCs (LLM + TTS) talk back at you. It is fun to play with a friend! And because it's just talking, non-gamers do great, often better than gamers. I really want to have a free demo, but no time yet to implement. For now, it's purchase only ($4.99). If anyone decides to try it, I'd really love to get more feedback. It was an enormous learning experience, especially targeting the web - so many partially supported web APIs and browser inconsistencies! Still feels like 2008 in some ways. Also happy to answer questions of course. Thanks, and enjoy the weekend! https://ift.tt/Qdf32qE April 20, 2024 at 01:33AM
Show HN: Composable (as in iGoogle, but modern) privacy-friendly new tab I spent quite a lot of time working on this one over the last 1.5 years. It started as a small project for my personal use because I wanted to keep all my self-hosted services visible so I wouldn't forget they existed lol. Using a web page wasn't ideal because of the white flicker every time I opened a new tab, so I decided to make this into a browser extension. From that time on, it became a lot bigger and got some traction (which I'm very happy about). It's made with React, but I tried to squeeze maximum performance (limited by my skills and desire to keep it somewhat readable, though) out of it. UI/UX was a big priority for me in this project, so I also tried to streamline it as much as possible and make Anori a joy to use. If you decide to try it, let me know how good I did! Oh, and it's open source [1] and the process of adding new widgets is documented [2], so you can make your own! [1]: https://ift.tt/BFUcw1K [2]: https://ift.tt/6VoP9rj... https://anori.app/ April 19, 2024 at 02:04PM
Show HN: MonsterWriter – Write a thesis, post, or organize notes Hello HN, in 2017 I started a project that would become MonsterWriter. First envisioned as a semi-structured wiki it became a writing application specialized for scientific content. It is a perfect tool if you write your thesis. While it is focused on technical content, you can still see the knowledge management spirit in it. One or two years ago, my wife joined me in my efforts and redesigned the whole project and we recently released it as MonsterWriter2. To celebrate this milestone we are giving away free lifetime licenses for the Desktop version (till Apr 28). Just use the promo code "gu0ho4q" for a 100% discount. You can find detailed instructions here: https://ift.tt/AzZcs6h You can also find a short introduction video to the app on YouTube: https://www.youtube.com/watch?v=vR8i-EY_UBk We are happy to receive any kind of feedback! April 19, 2024 at 01:12AM
Show HN: Minard – Generate beautiful charts with natural language Hi HN – Excited to share a beta for Minard, a new data visualization toolkit we've been working on that lets you generate publication-quality charts with simple natural language (throw away your matplotlib docs and rejoice!). Upload or import CSVs, Excel, and JSON, give it a spin, and please let us know what you think! (Long format data works best for now) For those curious, the stack is a simple Django app with HTMX/Alpine and all of the charts are specified and rendered as Vega ( https://ift.tt/gf508kq ). Lots of LLM function calling under the hood as well. https://minard.ai April 18, 2024 at 11:27PM
Show HN: Trained Tiny Tales GPT(30M model)from scratch and deployed in $15 For the last few weeks, I have been working on training an LLM from scratch and deploying it in production on Google Cloud Platform. Finally, I trained a 30 million parameter model on 1 billion tokens and deployed it as a web service. You can access the LLM using this site - https://ift.tt/WgQxNv1 The following steps were taken to build Tiny Tales GPT 1. Downloaded and preprocessed 8GB of dataset using multiprocessing library. 2. Tokenized the data using byte pair encoding to create 1 billion tokens sharded in different bin files. 3. Defined a training setup and trained the model on a small version of the LLaMA model architecture with 30 million parameters. 4. The training was done using Distributed Data-Parallel on two A-100 GPUs provided by JarvisLabs.ai (they are most cost-optimized) 5. After the training is done, an inference script is created to predict the tokens from the trained model given the input context vector. 6. Developed REST-based API service using Flask framework to interact with the inference service to the end user. 7. Finally used GCP's virtual machines, instance groups, load balancers, and DNS services to deploy the service on the internet. https://ift.tt/Rm1GCvH April 18, 2024 at 02:25AM
Show HN: Easy Folders – A folders based organization tool for ChatGPT and Claude Download for free: https://ift.tt/Vq8EitN A simple and powerful tool to organize chats in ChatGPT and Claude. ** NEW! Claude support added - Create folders and organize your chats for Claude ** ** NEW! Multi-account support added - Create folders and organize your chats for multiple ChatGPT & Claude accounts, and switch between them seamlessly. ** - Privacy focused, your data is all stored on your local machine - Create folders & sub-folders for your chats - Easy-to-use drag & drop - Colored folders - Automatic folders for Custom GPTs - Sync folders across multiple devices - Full chat history search - Bookmark important chats - Bulk delete chats - Backup & Import data - Dark & light mode support April 18, 2024 at 01:11AM
Show HN: Brewer X, a native macOS client for Homebrew Hi HN! Like many of you, for my entire career I have relied on Homebrew to install all kinds of software on my Mac. That's why today I'm really excited to share a new app that my partner and I are building: Brewer X, a refreshing user interface for Homebrew. Brewer X is graphical interface that lives on top of Homebrew. Leveraging the power of native APIs improves the classic experience and unlocks new features. For example: • bulk actions are performed in parallel • syncing the entire library locally provides incredible search performances and the ability to query descriptions and other previously unaccessible fields • maintenance scripts run automatically for you • last but not least... app icons (or favicons when not available) let you quickly identify what you're looking for The app is written in Swift and uses only AppKit with Nib files for top performance, pixel perfect design, and maximum flexibility. We also designed the app icon and all the others in the UI ourselves. Following the great insights from Sketch[^1] we managed to make them super crispy. We're also very proud to have been featured in the macOS App Icon Book[^2]. The app has only been out for a couple of weeks, but we've already seen an amazing response from the community. We can't tell you all of our future plans yet, but here's a list of things we'd like to see in the app in the near future: • Import/Export • Automatic replacement of apps installed without Homebrew • Notifications about available updates • Finder actions • Spotlight integration I hope you find Brewer X interesting. We're happy to answer any question! [^1]: https://ift.tt/Pe5bGkw... [^2]: https://ift.tt/xZ809H2 https://ift.tt/6DJCucx April 17, 2024 at 09:27PM
Show HN: Device-Bound Session Tokens in JavaScript Google’s recent announcement of a proposed ‘Device Bound Session Credentials’ feature[1] for Chrome reminded me of a project we worked on last year at my company. We focus on fraud prevention at signup and login (preventing multi-accounting and account theft), but some customers were concerned about post-login security and asked us to add a session hijacking prevention feature to our fraud prevention API. In the end, we decided to just implement a solution in Javascript. We call it session-lock, and it can be used today across all browsers[2] and, theoretically, native mobile apps. For a more comprehensive writeup and quick demo, you can visit the link. In short, the library adds a device-bound signature to the session token each time it’s used. At login, it creates a signing key pair on the browser using SubtleCrypto, with the private key set as “unextractable” and stored in IndexedDB (this forms the basis of its device-binding). Also at login, the public key is sent to the server along with the user’s credentials. If the credentials are valid, the server adds the public key to the payload of the JWT it returns to the client. When the client uses the JWT to access a protected resource, session-lock adds a signature to the end of it, along with a timestamp to mitigate replay attack risk. The server then validates the signature using the public key embedded in the JWT. The private key must be removed from IndexedDB upon logout. While Chrome’s DBSC would be a nice solution when it gets around to being deployed in Chrome and Edge, I think session-lock would help a lot today across all browsers in countering common attacks like malicious browser extensions that rip LocalStorage and cookies. Also, implementing the general flow in native mobile apps[3] would have the same key advantages as DBSC - compiled code already on the device and utilization of hardware TPMs. Aside from JS tampering and extracting “unextractable” CryptoKeys from IndexedDB, please let me know if you can think of any other potential attacks. Happy to answer any questions. [1] https://ift.tt/gdBb1LQ... [2] Other than Firefox private browsing mode due to its blocking of IndexedDB [3] Using CryptoKit / KeyStore for ECDSA https://ift.tt/NuU3TcI April 16, 2024 at 08:18PM
Show HN: YouTube Shorts Redirector I am neurodivergent and noticed the Youtube Shorts format was hacking my brain to engage longer than I wanted. I wrote this quick extension to gain my time back. If you have suggestions for improvement, I'm all ears. Thank you :) https://ift.tt/Jhsyuk5 April 16, 2024 at 06:29AM
Show HN: Building a GPS receiver Hi everyone! Shortly after publishing my iOS 4 jailbreak last October[1], I got to work on my next hobby project: a from-scratch homebrew GPS receiver, which can solve the user’s location solely from billions of radio antenna samples. I took a commodity SDR (alongside the Python standard library and numpy) and built a signal processing pipeline that can detect and track GPS satellites over many minutes, drop and pick up satellites as they come in and out of view, and precisely determine the user’s position and clock inaccuracy. All told, gypsum can go from a cold start to a fix on the user’s position, and the precise time, in less than a minute of listening to the antenna. I went on a journey of learning how to detect and track satellite signals that are literally too quiet to hear, and I hope that some of the magic comes through in the posts! After implementing this myself and walking the long road of getting it working, I’m left completely stunned by the brilliance of GPS, across so many axes. I hope you enjoy the read! On a more personal note, I’ll be starting a new job next week which isn’t as amenable to publishing side projects, and therefore this will be my last publicly-published project for some time. I’ve had great experiences making and sharing projects on here, and I’m really grateful for the positive feedback that’s been shared! [1]: https://ift.tt/gSVksFB https://ift.tt/7sM6EKR April 15, 2024 at 08:12PM
Show HN: Docker-boot – Run a system from RAM without LiveCD How often do you screw up the system so much you have to reformat the disk (without losing data) to fix it? Well, sometimes I do, and sometimes I can't be bothered to burn a live ISO onto a USB stick. There's initramfs, but it's hardly a pleasant environment, with network configuration and all. My go-to solution has typically been to create a chroot with busybox and a few utilities in /tmp, chroot into it, and then kill services that use the solid drive so that I can unmount it. That's an error-prone process, and sometimes systemd itself uses disk, so you can't unmount the drive despite killing all the userland but PID 1. This script improves the UX. It uses a Docker image as the chroot base, which is much easier to tailor to your needs, and automagically commits all the atrocities, such as tearing down all the userland processes, including PID 1, and re-spawning the host system from the container filesystem. It also drives libostree and Nix users mad, because it can be used to try out a new DE or even a whole OS without polluting the host filesystem or spawning a virtual machine. The video in the README shows me trying out KDE + SDDM from a host running GNOME + GDM3. https://ift.tt/EfLr1J2 April 14, 2024 at 11:08PM
Show HN: GPU price-per-hour tracker for A100/H100s Out of curiosity, I put together a simple website which tracks the prices for a few variations of A100/H100 GPUs by hour broken out between spot/ondemand, form factor and provider. Specifically I was tailoring the tool towards the smaller, emerging providers like runpod, gpulist.ai, lambda labs etc. Anyone have any ideas to expand/refine it? https://ift.tt/QxguWLd April 14, 2024 at 12:38AM
Show HN: Using Google Sheets as the back end/APIs of your app Hello everyone! At a company I worked for, we needed to develop an MVP (basically a web page) and apply certain business logic to a Google Drive spreadsheet that was frequently updated by the Sales team. In this case, we had two options: Develop a backend to replace the current spreadsheet and have the sales team use it as a new "backoffice" - This would take a very long time, and if the hypothesis we were testing was wrong, it would be time wasted. Create the web page and use Google's SDK to extract data from the spreadsheet. We chose to go with the second option because it was quicker. Indeed, it was much faster than creating a new backoffice. But not as quick as we imagined. Integrating with Google's SDK requires some effort, especially to handle the OAuth logic, configure it in the console, and understand the documentation (which is quite shallow, by the way). Anyway! We did the project and I realized that maybe other devs might have encountered similar issues. Therefore, I developed a tool that transforms Google spreadsheets into "realtime APIs" with PATCH, GET, POST, and DELETE methods. Since it's a product for devs, I think it would be cool to hear your opinions. It's still quite primitive, but the basic features already work. https://zerosheets.com https://ift.tt/5DWm6ZJ April 12, 2024 at 09:59PM
Show HN: Chapter-level summaries of bestselling books using LLMs Most book summaries on the internet are not detailed enough. In this project, I've tried to create detailed chapter-level summaries that retain a greater level of detail and vocabulary of the original. The aspiration is for these summaries to be a credible alternative to reading the books themselves. Website hosted on Vercel: https://ift.tt/3WRV4bx The code behind the project on GitHub: https://ift.tt/cHhy9s1 Request additional books: https://ift.tt/liqtHA2... https://ift.tt/3WRV4bx April 13, 2024 at 12:16AM
Show HN: CodeFusion, use LLMs with large codebases (with token context in mind) Manually copying and pasting code snippets is tedious. CodeFusion allows you to easily select and merge specific code files across your codebase into a single, cohesive prompt block so that you can stay within LLMs token context budget. https://ift.tt/WBUfmFo April 11, 2024 at 10:21PM
Show HN: Aarde – Self Replicating Programs Last weekend I created a new visualization/demo. It's a sort of crossover between artificial life and defragmentation on MSDOS. The organisms "live" in the memory and continuously copy themselves to a free piece of memory. The programming language consists of only 5 instructions to keep it simple: S - Program start E - Program end F - Find location to copy to C - Copy to location J - Jump to start of program Every time a code is copied, a mutation can occur. We start with the following simple life form: SFCJE The rest is evolution. https://ift.tt/vEoSY2M April 12, 2024 at 01:29AM
Show HN: Wunderbar, Learn Language While Working on Your Mac Hey everyone, I've been living in Germany for a while now and struggling to pick up German. Even though I finished a B1 level language course, remembering all the common German words has been tough. I've tried a bunch of apps, but couldn't stick to any of them. Last month, I had an idea. I thought, what if I could see a German word and its meaning on the Menu Bar of my Mac? That way, I could learn new words while working without needing to set aside dedicated time. The app uses a spaced-repetition algorithm, so you'll see the same word multiple times until you've got it down. And even after you've learned it, you'll still see it again to make sure you don't forget. After making the app, lots of people asked me to include other languages, therefore I’ve added languages like Spanish, French, Dutch, Swedish, Japanese, and Italian. You can download the app here ($3.99): https://ift.tt/vodgA5p There's no subscription or anything like that. You pay once and it's yours forever. Let me know what you think! April 11, 2024 at 12:35PM
Show HN: Next-token prediction in JavaScript – build fast LLMs from scratch What inspired this project today was watching this amazing video by 3Blue1Brown called "But what is a GPT?" on Youtube ( https://www.youtube.com/watch?v=wjZofJX0v4M - I highly recommend watching it). I added it to the repo for reference. When it clicked in my head that "knowing a fact" is nearly synonymous with predicting a word (or series of words), I wanted to put it to the test, because it seemed so simple. I chose JavaScript because I can exploit the way it structures objects to aid in the modeling of language. For example: "I want to be at the beach", "I will do it later", "I want to know the answer", ... becomes: { I: { want: { to: { be: { ... }, know: { ... } } }, will: { ... } }, ... } in JavaScript. You can exploit the language's fast object lookup speed to find known sentences this way, rather than recursively searching text - which is the convention and would take forever or not work at all considering there are several full books loaded in by default (and it could support many more). Accompanying research yielded learnings about what "tokens" and "embeddings" are, what is meant by "training", and most of the rest - though I'm still learning jargon. I wrote a script to iterate over every single word of every single book to rank how likely it is that word will appear next, if given a cursor, and extended that to rank entire phrases. The base decoder started out what I'll call "token-agnostic" - didn't care if you were looking for the next letter... word... pixel... it's the same logic. But actually it's not, and it soon evolved into a text (language) model. But I have plans to get into image generation next (next-pixel prediction), using this. Overall the concepts are similar, but there are differences primarily around extraction and formatting. Goals of the project: - Demystify LLMs for people, show that it's just regular code that does normal stuff - Actually make a pretty good LLM in JavaScript, with a version at least capable of running in a browser tab https://ift.tt/NbrXJOt April 11, 2024 at 02:57AM
Show HN: We built a no-code tool on top of a spreadsheet (Game of Life example) Hi everyone! I want to share a spreadsheet tool we are building. The main idea is to be able to specify logic for a no code tool inside a spreadsheet. We have automations, data integration, action formulas, dashboards. Linked example: We use our UPDATECELL formula to make a step in a board of game of life. We can also use CHAIN to make it run indefinitely. The actions are exposed as buttons in a dashboard. Our automations work by loading your spreadsheet on the server and running an action specified in the spreadsheet. So the logic could be, pull in data from google sheet, send a personalized email to every row which we haven’t emailed yet, mark those rows as complete. Run this automation on 5 minute schedule, add any other custom logic you need (e.g. test email for deliverability) and you have yourself an automatic email tool. We are leaning into automations to differentiate ourselves from other products in the field at the moment. Technical kinks: - Since our automations make changes to the spreadsheet, we needed a robust multiplayer system. We implemented a CRDT system. This is a topic for a longer blog post, but one fun challenge was to implement CRDT for rows/column which allowed for adding/deleting/reordering performantly while preserving most user intent in the case of offline editing. - Some of our formulas are inherently asynchronous; for example those which fetch data from an external data source. That added an extra kink when writing the spreadsheet engine; e.g. topologically sorting and then evaluating in order can cause you to get stuck on a long running evaluation. - One of our premises is that we do not store your data; it gets pulled when the spreadsheet evaluates. That means everything reevaluates in the browser when you open a spreadsheet. Get this to be performant (we still have a long way to go) was a challenge. Currently, everything runs in a single worker. Let me know what you all think. https://ift.tt/nZBsXzT April 11, 2024 at 02:29AM
Show HN: Hacker News Blogroll This was submitted about 9 months ago as a Show HN ( https://ift.tt/57dpaCT ), people was generally favorable to it, but I never got around to do anything else with it past the first few days. I recently rescued a Github account I had, so I'm putting the source of the Rails app over there in case anyone wanted to do anything with it. The site still runs on https://dm.hn Generates about 2 to 3 GB bandwidth usage every month, and the blogs are still checked every day for new content. I must say, I still visit every week and find interesting entries to read. https://ift.tt/QPz9mOJ April 11, 2024 at 01:04AM
Show HN: Visualize eBay laptops in bulk with laptopscout A bit late in the new diskprices-like website trend, but here's my contribution to the ecosystem. The goal here is to get an instant comparison of the laptops on ebay, using mostly 3 metrics of now: CPU benchmark rank, GPU benchmark rank, and price, without having to scroll through the item description or google the PC model. At the moment there are some parsing bugs, and some components that aren't parsed yet. I also quickly run into the ebay rate limit, making it hard to maintain an hour-by-hour listing. Curious to know what you guys think! https://laptopscout.xyz/EBAY_US April 10, 2024 at 02:04AM
Show HN: The fastest way to run Mixtral 8x7B on Apple Silicon Macs I’d originally launched my app: Private LLM[1][2] on HN around 10 months ago, with a single RedPajama Chat 3B model. The app has come a long way since then. About a month ago, I added support for 4-bit OmniQuant quantized Mixtral 8x7B Instruct model, and it seems to outperform Q4 models at inference speed and Q8 models at text generation quality, while consuming only about 24GB of RAM[3] at 8k context length. The trick is: a) to use a better quantization algorithm and b) to use unquantized embeddings and the MoE gates (the overhead is quite small). Other notable features include many more downloadable models, support for App Intents (Siri, Apple Shortcuts), on-device grammar correction, summarization etc with macOS services and an iOS version (universal app), also with many smaller downloadable models and support for App Intents. There's a small community of users building and sharing LLM based shortcuts on the App's discord. Last week, I also shipped support for the bilingual Yi-34B Chat model, which consumes ~18GB of RAM. iOS users and users with low memory Macs can download the related Yi-6B Chat model. Unlike most popular offline LLM apps out there, this app uses mlc-llm for inference and not llama.cpp. Also, all models in the app are quantized with OmniQuant[4] quantization and not RTN quantization. [1]: https://privatellm.app/ [2]: https://ift.tt/yEh5Pbr [3]: https://www.youtube.com/watch?v=4AE8yXIWSAA [4]: https://ift.tt/3jrCQPy April 8, 2024 at 09:37PM
Show HN: Dimity Jones in Puzzle Castle: An Electronic Escape Novel _Dimity Jones In Puzzle Castle: An Electronic Escape Novel in Eighty-Nine Ciphertexts_ is a (mostly) fictional story, contained in a single text file, that requires the reader to solve puzzles as they go along, and to use each chapter's solution as a key to decipher the next. Think: escape room in the form of a novel. A computer, and rudimentary coding skills in a language of your choice, will be indispensable for performing the transformations -- and might help with the solving too! My wife, the author, passed away five years ago. This is not the last thing she wrote, but it is the most unusual, unapproachable, and personal of her major works. It is also, as the only novel of hers that I cannot breeze through in an afternoon (and despite my unflattering appearance in it), my favorite. Though _Dimity Jones_ was left unfinished, and perhaps abandoned, at the time of my wife's death, its elements were all there, on her hard disk, awaiting only a final compiling. My contribution to this text has therefore been little more than that of an occasional copyeditor (my wife was a meticulous speller and self-proofreader) and playtester. Before releasing this work more widely, I would love to have it test-driven by better coders and puzzlers than I. Any and all feedback is welcome, from positive to negative, from the sweeping to the picayune. Let me know what confuses or frustrates you -- and especially let me know if (where) you get stuck. Otherwise, there are no special instructions; it's all in the book. While _Dimity Jones_ is still in its debugging/proofreading phase, please refrain from sharing it or putting it any (other) public place. (Christine would have been horrified to see her work thus published before it was letter-perfect; but I have exhausted my pool of friends and colleagues both able and willing to tinker with it. This seems like the ideal community of potential testers.) Thanking you in advance. I hope you enjoy! https://ift.tt/EfIWtOL April 9, 2024 at 12:17AM
Show HN: Toolkit for LLM Fine-Tuning, Ablating and Testing Hello all! Very happy to share this toolkit that allows you to fine-tune your choice of open-source LLMs on your data! The toolkit also allows you to run ablation studies across LLMs, prompt designs, training configurations, and can ingest different data files -- all through just ONE YAML file! After fine-tuning, you can also run a bunch of tests to ensure that the fine-tuned LLM behaves as expected, enabling faster time-to-production! Why this toolkit? Why now? While closed-source LLMs have become popular for chat-based applications, enterprises are considering a shift to self-hosted SLMs (smaller language models) since there is evidence that you don't need a gigantic model to solve narrow edge-cases. Plus, enterprises want to own the data pipeline from start to end, i.e., data ingestion, training, deployment, feedback collection and testing! Their customers' valuable data stays within their ecosystem, allowing enterprises to not worry about compliance or data leakage issues that come up using third-party APIs. While there are a few repositories out there that do vanilla fine-tuning, it is well known that it takes more than a one run to find the desirable setting of weights / parameters for your specific data. Bearing this pain-point in mind, we designed the toolkit to allow running multiple experiments through one config file! Around 5 months ago, I had shared a repository that contained individual fine-tuning scripts for the most popular LLMs. While the repository received great reception from this community, there was one unanimous feedback -- the community wants to build on top of our scripts! This prompted us to design the toolkit, bearing in mind the pain-points that data scientists / researchers / engineers like myself face! Please feel free to give it a try! Looking forward to your feedback! https://ift.tt/YlxmsGt April 7, 2024 at 10:03PM
Show HN: Basecode – The fastest way to create a web app Hi all! Hereby I proudly present my open source project to help you quickly create web applications called Basecode. This tool allows you to quickly generate the overall structure of your application - including scaffolds and tests - using Kotlin, Spring Boot, Typescript, React (Next.js), Postgres and GraphQL. This means you'll get a modern, statically typed programming language with a solid web framework on the backend, and a highly flexible frontend that is also statically typed and has a huge ecosystem. What sets Basecode apart from other code generation platforms is its focus on: - Loose coupling: by making your application as loosely coupled as possible, it remains maintainable, even at larger scale. For example, check out the video on the Github page to see just how easily it deals with relationships between different aggregates. - Giving you everything you need, but nothing more: it generates well-structured code that you need for your project, but it does not add any bells or whistles. So... Who would benefit from using Basecode? You'll benefit from Basecode if: 1. You just want to focus on providing excellent business value, not on reinventing the wheel. 2. You want to deliver - and be able to pivot - your application as quickly and as often as possible. 3. You're either a solopreneur, prototyping your next app - or you're creating a bigger web application that is built on a platform designed to scale (or anything in between). (4.) Extra bonus points: if you like working with Kotlin, Spring Boot, React (Next.js), Postgres and GraphQL. I hope you give Basecode a try and let me know what you think! Kind regards, Wouter https://ift.tt/fzLmOnb April 6, 2024 at 06:27PM
Show HN: Bluetuith – A console TUI-based Bluetooth manager Hello HN, bluetuith is a Bluetooth manager for the terminal, which aims to replace most Bluetooth managers, and can perform any Bluetooth based operations and interact with devices. Please have a look at the repository and the documentation for more information. I have been working on this for some time now, and I would like constructive suggestions to further improve the application, especially with regards to cross-platform functionality, so that the user experience can be enhanced. Any suggestions will be highly appreciated. Contributors are welcome as well. https://ift.tt/PVhJla4 April 6, 2024 at 12:17PM
Show HN: Built a premium directory dedicated to high-quality tools This is a previous project I decided to relaunch. There's tons of directories that accept all submissions, but I feel like that just waters down the platform. It also isn't fair for the amazingly built tools and services. I pivoted to a premium-only directory and feel like charging a small fee, while it might take longer to build the portfolio of listed tools, they'll be of much higher quality than other platforms. https://findcool.tools April 6, 2024 at 11:28AM
Show HN: My Free SEO Scoring Tool I hope that this tool proves to be of immense help to you. However, I understand that it may have its limitations and there may be room for improvement. To make this tool more effective and user-friendly, I welcome your feedback and suggestions. Your insights will be instrumental in improving the tool and making it more beneficial for all users. Lastly, your support and guidance mean the world to me. Together, we can make the SEO journey smoother for everyone. So, dive in, explore the Free SEO Scoring Tool, and let's take your website's performance to new heights! https://seonaai.com April 6, 2024 at 08:20AM
Show HN: Diego – A CLI tool for importing into Hugo exported data from services Hey there! Initially, I had created a script to automate importing some exported files into my Hugo website. As I implemented support for the second service though, I realized that it would be better to convert the script into a CLI tool. (Also, a good opportunity to learn/practice Go). That's 'diego.' I released it about a month ago, but I'm only announcing it here now. Basically, it's a CLI tool designed to import official exported data, CSV and JSON files, from popular services into Hugo. It offers: - Automatic CSV and JSON conversion into Hugo data files - Support for all Hugo data file formats - Easy data management in a human-readable format (YAML) - Automatic generation of Hugo shortcodes for imported data - Optional scrape capabilities for fetching missing fields - Flags suited for scripting and pipelines - Persistent configuration Feedback, suggestions, constructive criticism, and contributions are welcome! I've just submitted a patch implementing support for Instapaper. If you have ideas for additional services that would be a good fit to add support to, let me know. This is my first released FOSS project Below is a link containing a plaintext report of my TODOS along with the time I've spent on each item while developing diego. (I think it helps getting an overview of the project internals). https://ift.tt/hCrcLfw... https://ift.tt/c9zGi47 April 6, 2024 at 02:18AM
Show HN: PredicateKit – A type-safe replacement for NSPredicate for CoreData Hi, I really like CoreData. I think it's a great piece of software (I know this is a controversial opinion in some circles ;)). My only pet peeve with it has been the string-based querying API based on NSPredicate. It is a major source of bugs/crashes and doesn't really fit nicely in the modern strongly-typed world of Swift. I built PredicateKit as a lightweight replacement for NSPredicateKit (specifically for CoreData) that makes writing predicates as safe and pleasant as writing native Swift code. https://ift.tt/fmdFkUb April 5, 2024 at 05:03AM
Show HN: FizzBee – Formal methods in Python – Easiest Lang for everyday use GitHub: https://ift.tt/lZXVC3j Traditionally, formal methods are used only for highly mission critical systems to validate the software will work as expected before it's built. Recently, every major cloud vendor like AWS, Azure, Mongo DB, confluent, elastic and so on use formal methods to validate their design like the replication algorithm or various protocols doesn't have a design bug. I used TLA+ for billing and usage based metering applications. However, the current formal methods solutions like TLA+, Alloy or P and so on are incredibly complex to learn and use, that even in these companies only a few actually use. Now, instead of using an unfamiliar complicated language, I built formal methods model checker that just uses Python. That way, any software engineer can quickly get started and use. I've also created an online playground so you can try it without having to install on your machine. In addition to model checking like TLA+/PlusCal, Alloy, etc, FizzBee also has performance and probabilistic model checking that be few other formal methods tool does. (PRISM for example, but it's language is even more complicated to use) Please let me know your feedback. Url: https://FizzBee.io Git: https://ift.tt/lZXVC3j https://fizzbee.io/ April 2, 2024 at 04:15PM
Show HN: Managed GitHub Actions Runners for AWS Hey HN! I'm Jacob, one of the founders of Depot ( https://depot.dev ), a build service for Docker images, and I'm excited to show what we’ve been working on for the past few months: run GitHub Actions jobs in AWS, orchestrated by Depot! Here's a video demo: https://www.youtube.com/watch?v=VX5Z-k1mGc8 , and here’s our blog post: https://ift.tt/GXUeb0v . While GitHub Actions is one of the most prevalent CI providers, Actions is slow, for a few reasons: GitHub uses underpowered CPUs, network throughput for cache and the internet at large is capped at 1 Gbps, and total cache storage is limited to 10GB per repo. It is also rather expensive for runners with more than 2 CPUs, and larger runners frequently take a long time to start running jobs. Depot-managed runners solve this! Rather than your CI jobs running on GitHub's slow compute, Depot routes those same jobs to fast EC2 instances. And not only is this faster, it’s also 1/2 the cost of GitHub Actions! We do this by launching a dedicated instance for each job, registering that instance as a self-hosted Actions runner in your GitHub organization, then terminating the instance when the job is finished. Using AWS as the compute provider has a few advantages: - CPUs are typically 30%+ more performant than alternatives (the m7a instance type). - Each instance has high-throughput networking of up to 12.5 Gbps, hosted in us-east-1, so interacting with artifacts, cache, container registries, or the internet at large is quick. - Each instance has a public IPv4 address, so it does not share rate limits with anyone else. We integrated the runners with the distributed cache system (backed by S3 and Ceph) that we use for Docker build cache, so jobs automatically save / restore cache from this cache system, with speeds of up to 1 GB/s, and without the default 10 GB per repo limit. Building this was a fun challenge; some matrix workflows start 40+ jobs at once, then requiring 40 EC2 instances to launch at once. We’ve effectively gotten very good at starting EC2 instances with a "warm pool" system which allows us to prepare many EC2 instances to run a job, stop them, then resize and start them when an actual job request arrives, to keep job queue times around 5 seconds. We're using a homegrown orchestration system, as alternatives like autoscaling groups or Kubernetes weren't fast or secure enough. There are three alternatives to our managed runners currently: 1. GitHub offers larger runners: these have more CPUs, but still have slow network and cache. Depot runners are also 1/2 the cost per minute of GitHub's runners. 2. You can self-host the Actions runner on your own compute: this requires ongoing maintenance, and it can be difficult to ensure that the runner image or container matches GitHub's. 3. There are other companies offering hosted GitHub Actions runners, though they frequently use cheaper compute hosting providers that are bottlenecked on network throughput or geography. Any feedback is very welcome! You can sign up at https://ift.tt/eKW1Rat for a free trial if you'd like to try it out on your own workflows. We aren't able to offer a trial without a signup gate, both because using it requires installing a GitHub app, and we're offering build compute, so we need some way to keep out the cryptominers :) April 4, 2024 at 08:02PM
Show HN: AI generator for faceless TikTok videos Hi there Several months ago I launched an app called LogoPicture AI. It's a simple app that generates picture in the style of your logo. Last week I sold this app. It was my first app ever which generated some revenue. But I really struggled with marketing. I tried SEO, TikTok, Twitter - many other things. And one thing I realized that in marketing you need to be super consistent to win. You need to create a video every day, a post, etc. But I am too lazy for this :)) So, this is how the idea of Cliplama was born. I wanted to automate my marketing. I wanted to create a tool what will do at least some marketing for me. A tool, that will create a popular social account for me and I will be able to funnel its traffic in my apps. So yeah, this is that Cliplama is about. It helps you to create faceless TikTok or even YouTube channel and just post there videos daily. That's it. Right now Cliplama is in beta, so there are some bugs for sure. Some features are also missing. But I felt like it's a time to launch it. Would love to hear your feedback. Thank you! https://cliplama.com/ April 3, 2024 at 09:50PM
Show HN: Burr – A framework for building and debugging GenAI apps faster Hey HN, we're developing Burr (github.com/dagworks-inc/burr), an open-source python framework that makes it easier to build and debug GenAI applications. Burr is a lightweight library that can integrate with your favorite tools and comes with a debugging UI. If you prefer a video introduction, you can watch me build a chatbot here: https://www.youtube.com/watch?v=rEZ4oDN0GdU . Common friction points we’ve seen with GenAI applications include logically modeling application flow, debugging and recreating error cases, and curating data for testing/evaluation (see https://ift.tt/M9NQKIr ). Burr aims to make these easier. You can run Burr locally – see instructions in the repo. We talked to many companies about the pains they felt in building applications on top of LLMs and were surprised how many built bespoke state management layers and used printlines to debug. We found that everyone wanted the ability to pull up the state of an application at a given point, poke at it to debug/tweak code, and use for later testing/evaluation. People integrating with LLMOps tools fared slightly better, but these tend to focus solely on API calls to test & evaluate prompts, and left the problem of logically modeling/checkpointing unsolved. Having platform tooling backgrounds, we felt that a good abstraction would help improve the experience. These problems all got easier to think about when we modeled applications a state machines composed of “actions” designed for introspection (for more read https://ift.tt/PDZXpUx... ). We don’t want to limit what people can write, but we do want to constrain it just enough that the framework provides value and doesn’t get in the way. This led us to design Burr with the following core functionalities: 1. BYOF. Burr allows you to bring your own frameworks/delegate to any python code, like LangChain, LlamaIndex, Hamilton, etc. inside of “actions”. This provides you with the flexibility to mix and match so you’re not limited. 2. Pluggability. Burr comes with APIs to allow you to save/load (i.e. checkpoint) application state, run custom code before/after action execution, and add in your own telemetry provider (e.g. langfuse, datadog, DAGWorks, etc.). 3. UI. Burr comes with its own UI (following the python batteries included ethos) that you can run locally, with the intent to connect with your development/debugging workflow. You can see your application as it progresses and inspect its state at any given point. The above functionalities lend themselves well to building many types of applications quickly and flexibly using the tools you want. E.g. conversational RAG bots, text based games, human in the loop workflows, text to SQL bots, etc. Start with LangChain and then easily transition to your custom code or another framework without having to rewrite much of your application. Side note: we also see Burr as useful outside of interactive GenAI/LLMs applications, e.g. building hyper-parameter optimization routines for chunking and embeddings & orchestrating simulations. We have a swath of improvements planned. We would love feedback, contributions, & help prioritizing. Typescript support, more ergonomic UX + APIs for annotation and test/eval curation, as well as integrations with common telemetry frameworks and capture of finer grained information from frameworks like LangChain, LlamaIndex, Hamilton, etc… Re: the name Burr, you may recognize us as the authors of Hamilton (github.com/dagworks-inc/hamilton), named after Alexander Hamilton (the creator of the federal reserve). While Aaron Burr killed him in a duel, we see Burr being a complement, rather than killer to Hamilton ! That’s all for now. Please don’t hesitate to open github issues/discussions or join our discord https://ift.tt/RWlzhXG to chat with us there. We’re still very early and would love to get your feedback! https://ift.tt/3t6GjcY April 3, 2024 at 07:12PM
Show HN: I just open sourced my document/website extractor for Vision-LLMs Hi HackerNews, Lately, I have seen an explosion in posts offering paid APIs/services to get unstructured data into LLMs (i.e. langchain extract, ragflow, unstructured, unstract, just to name a few) and I have been largely disappointed by them, either because they fail to implement multimodal support, fail to give good context for "really tricky" PDFs / Word docs / Powerpoints, or are just plain difficult to use. In light of all these posts I figured I'd share my solution that has been working smoothly for me and my clients. I put it up on GitHub for free so you can check it out and hopefully offer some feedback / criticism or contribute to the code yourself. and BTW, I'm not trying to throw shade at any of the services mentioned, I'm just giving my honest experience in case there are others out there who feel the same way and want something that works Cheers! https://ift.tt/3kbCI5g April 3, 2024 at 12:10AM
Show HN: ghore, an Offline Renderer for GitHub flavoured Markdown Ghore is a simple cli tool to preview markdown files like they would appear on Github. It doesn't rely on Github's API and can be used completely offline. It also supports live preview of mermaid code blocks (Github's API doesn't process mermaid codeblocks). https://ift.tt/XoHZROq April 2, 2024 at 11:16PM
Show HN: RSS Filter Sometimes it's hard to keep up with multiple RSS feeds, and would be great to get some help filtering out the least relevant part of it. RSS filter is a side project that I created for myself to act as an RSS feed articles recommendation system, but self-hosted. It has a Python backend that is able to “proxy" your existing RSS feeds and after learning a bit about your interests, starts recommending similar articles and filtering out the rest. While, of course, always including some random articles for discovery. It uses LLM embeddings and machine learning to recommend similar articles. Embedding models allow for a new era of recommendation systems, where a large user base is not required, since allowing for self-hosting. Feel free to try it on my server or self-host it yourself. All feedback and contributions are welcome. https://ift.tt/bOvEsnw April 1, 2024 at 01:48AM
Show HN: Turn your work into a multiplayer adventure I was on the eternal quest of searching for a properly gamified productivity app that keeps me motivated throughout the day and in the long term. After a few months of a fruitless search, I made up my mind and decided to build it myself! I've since launched FocuMon a little over a month ago. Quite a few early users told me it has helped them stay focused, so I'm excited to share it on HN for ya'll to try it out and let me know how I can make it better! FocuMon is like an idle Pokémon game where you can collect, level up, and evolve cute monsters by being productive! It puts a reward layer on top of your work which helps to give you an initial boost to get into focus; it then gets out of your way once you are in the flow :) I'm actively developing more features & game modes for FocuMon. It is also a multiplayer-first app where you can adventure together with people from all over the world! Cheers and thanks for your time! Milton https://ift.tt/XLk6zym March 31, 2024 at 11:28PM
संदेश
