Show HN: ts-blank-space – fast TypeScript type-stripper https://bit.ly/4ev9zAu

Show HN: ts-blank-space – fast TypeScript type-stripper We've built a new type-stripping compiler for TypeScript source: `ts-blank-space` It's: * 5.6x faster than `tsc` due to zero-codegen * Written in TypeScript * Uses the official TypeScript parser * Accurate/preserved source locations * In used in production It uses the same novel approach that inspired Node 22.6.0's `--experimental-strip-types` functionality, released last month https://bit.ly/4gzx2m0 September 19, 2024 at 02:45PM

Show HN: LeanRL: Fast PyTorch RL with Torch.compile and CUDA Graphs https://bit.ly/3MRVf9B

Show HN: LeanRL: Fast PyTorch RL with Torch.compile and CUDA Graphs We're excited to announce that we've open-sourced LeanRL, a lightweight PyTorch reinforcement learning library that provides recipes for fast RL training using torch.compile and CUDA graphs. By leveraging these tools, we've achieved significant speed-ups compared to the original CleanRL implementations - up to 6x faster! Reinforcement learning is notoriously CPU-bound due to the high frequency of small CPU operations. PyTorch's powerful compiler can help alleviate these issues, but comes with its own costs. LeanRL addresses this challenge by providing simple recipes to accelerate your training loop and better utilize your GPU. Key results: - 6.8x speed-up with PPO (Atari) - 5.7x speed-up with SAC - 3.4x speed-up with TD3 - 2.7x speed-up with PPO (continuous actions) Why LeanRL? - Single-file implementations of RL algorithms with minimal dependencies in the spirit of gpt-fast - All optimization tricks are explained in the README - no heavy doc, just simple tricks - Forked from the popular CleanRL library Check out LeanRL on https://bit.ly/3XTi3MI now! https://bit.ly/3TC4lei September 20, 2024 at 01:27AM

Show HN: Real-time election betting odds and forecasts https://bit.ly/3zenI6x

Show HN: Real-time election betting odds and forecasts https://bit.ly/47x5mtT September 19, 2024 at 12:57PM

Show HN: ts-remove-unused – Remove unused code from your TypeScript project https://bit.ly/4deCOGK

Show HN: ts-remove-unused – Remove unused code from your TypeScript project ts-remove-unused is a command line tool for TypeScript projects that auto-fixes unused `export`s. It removes the export keyword from the declaration or the whole declaration based on its usage in the project. There are some similar tools but they are focused on "detecting" rather than "removing" so I've built one myself. I wanted a solution that's as minimal as possible; config files to specify the files in your project shouldn't be necessary because that info should be already configured in tsconfig.json. All you need to do is to specify your entrypoint file. Feedback is much appreciated! https://bit.ly/47Bl3k2 September 16, 2024 at 09:39AM

Show HN: High-Level Synthetic Data Generation from Verbal Descriptions https://bit.ly/4epH1sj

Show HN: High-Level Synthetic Data Generation from Verbal Descriptions Hi all! In statistics, synthetic data benchmarks are important for understanding the strengths and limitations of competing algorithms. For example, in clustering – the art of identifying groups of data points that are similar to each other – researchers typically study how algorithms perform on mock scenarios like “five oblong clusters in 2D with some overlap.” Unfortunately, creating these scenarios typically involves a lot of work. You have to design entire data sets so they match the scenario description. In clustering, this involves selecting cluster centers, tuning covariance matrices, etc. As part of my PhD at Caltech, I have developed a high-level synthetic data generator for clustering that automates this process. You only have to describe your desired scenario in English, and the algorithm takes care of creating data sets with suitable clusters. This means researchers can easily set up benchmarks by passing scenario descriptions as a list of strings. We have put up a demo here: https://bit.ly/4d9ip61 . Curious to hear your thoughts! Mike https://bit.ly/4d9ip61 September 19, 2024 at 04:17AM

Show HN: Poker over SSH https://bit.ly/47vP70m

Show HN: Poker over SSH I've been interested in making a TUI application and learning Rust for a while now. Since some friends and family started playing poker, I found it a good opportunity to make my "learning Rust" project a poker TUI app. Inspired by https://bit.ly/4gwM8Ja , I also wanted to make something that could be connected to and played with just using SSH. For me, poker is usually with friends and family, so I focused the app's target audience to be private, small-scale games, avoiding the headaches associated with scaling to some massive web-based platform. Big thanks again to the folks that've already taught me some cool things about Rust and UI design. https://bit.ly/4gJxOgD September 19, 2024 at 02:54AM

Show HN: Tipsy simulated tic tac toe https://bit.ly/4esuc0a

Show HN: Tipsy simulated tic tac toe Wanted to try out Cursor and was inspired by a recent podcast by Greg Isenberg/Jason Fried to make "weird" tech experiences. Fun little https://bit.ly/4ddlK3U September 14, 2024 at 05:43PM

Show HN: Parse your Postgres queries into a fully-typed AST in TypeScript https://bit.ly/4eL5k4j

Show HN: Parse your Postgres queries into a fully-typed AST in TypeScript Hey all, I'm the creator of @pg-nano/pg-parser. I'm using it in pg-nano[1] to statically analyze Postgres schemas spread across multiple SQL files for a couple of reasons: 1. Each CREATE statement needs to be in topological order, so pg-nano's dev command can execute them without issue. 2. pg-nano has a plugin system like Vite that allows SQL generation based on the parsed schema. Probably to the surprise of no one, working with an untyped AST feels like you're back in the days of JavaScript, because well... you are. Most of you know by now just how great TypeScript and static types in general are, especially if you appreciate SQL. So why is this project worth sharing with you? Well, writing the AST type definitions by hand would have taken me way too much time. It would also be a bear to keep up-to-date as Postgres continues to evolve. To my surprise, I discovered that libpg_query, the C library used under-the-hood, includes JSON definitions in their /srcdata/ folder. I figured I could use them to generate the type definitions. Genius, right? Okay... maybe not genius , but still cool, I think. You see, those JSON definitions provided by libpg_query? They don't exactly contain the TypeScript definitions (was that obvious?). No, no. I had to translate them into TypeScript definitions. (I'm sure you could have done it, yes yes. But did you? No siree bob) It was pain-staking, but overall really not too hard. Time-consuming? Yes, but not as much as writing the type definitions by hand. So... was it worth it? Only time will tell. I hope you find it as useful as I do. And that's all I've got, so thanks for reading. P.S. The build for Windows is broken, so if anyone could lend a hand, you would be a true hero. [1]: https://bit.ly/3zqP7SH (not ready for production use) https://bit.ly/4eL5kRR September 18, 2024 at 08:49AM

Show HN: Modern CI/CD Platform for Kubernetes https://bit.ly/4dgBNxT

Show HN: Modern CI/CD Platform for Kubernetes https://bit.ly/3wHU3Mt September 18, 2024 at 05:47AM

Show HN: I'm making a text editor for desktop (gonna build AI in it soon) https://bit.ly/4ejFUdS

Show HN: I'm making a text editor for desktop (gonna build AI in it soon) https://bit.ly/4ejFUuo September 18, 2024 at 03:55AM

Show HN: Exocortex – Encrypted Note-Taking Desktop App for Networked Thought https://bit.ly/3B8VCdq

Show HN: Exocortex – Encrypted Note-Taking Desktop App for Networked Thought https://bit.ly/3MUihfT September 16, 2024 at 11:51PM

Show HN: Fragaria – From 'R's in Strawberry to Complex Problem-Solving AI https://bit.ly/4e52Ai8

Show HN: Fragaria – From 'R's in Strawberry to Complex Problem-Solving AI I'm excited to share Fragaria, an open-source project that brings advanced AI reasoning capabilities to developers and researchers. Named after the botanical genus of strawberries (a nod to the classic "How many 'r's in strawberry?" problem), Fragaria is designed to tackle both simple queries and complex logical puzzles with increasing efficiency over time. Here are some features! * Multi-Provider Support: Seamlessly switch between OpenAI, Groq, and Together.ai as LLM providers. * Chain of Thought (CoT) Reasoning: Break down complex problems into step-by-step solutions. * Reinforcement Learning: Continuously improve problem-solving strategies through a novel approach combining CoT and RL. * Adaptive Learning: Utilize a SQLite-based scoring system to remember and refine successful strategies. * Easy Integration: RESTful API with OpenAI-compatible endpoints for drop-in replacement in existing projects. What makes Fragaria unique: * More than just another LLM wrapper. Fragaria implements a sophisticated reasoning process that grows more efficient with each solved problem. * The combination of CoT and RL allows Fragaria to develop novel problem-solving approaches that weren't explicitly programmed. * It's designed for researchers to easily experiment with different CoT strategies and RL algorithms. The project includes a sleek Svelte-based demo interface for easy visualization of the reasoning process. I built Fragaria to explore the intersection of language models, reasoning, and reinforcement learning. https://bit.ly/3XrFHhQ September 17, 2024 at 12:28AM

Show HN: React SaaS – Boilerplate with automated setup of dev/prod environments https://bit.ly/3XNERgN

Show HN: React SaaS – Boilerplate with automated setup of dev/prod environments After accidentally nuking the production database of my last side project, I wished I had used dev/prod environments with automated DB backups. When I looked for a SaaS template to start over with integrated dev/prod environments, I couldn’t find one. So I decided to build it myself. To help speed up the setup, I also built Bash and PowerShell scripts to guide users through the process. The current stack is Next.js, Firebase, and Stripe. However, I’m considering writing some Terraform modules for GCP resources to further speed up setup and provide more flexibility for power users. Would love to get your feedback! https://bit.ly/4ehkm1k September 17, 2024 at 12:12AM

Show HN: Sisi – Semantic Image Search CLI tool, locally without third party APIs https://bit.ly/47tl9Ke

Show HN: Sisi – Semantic Image Search CLI tool, locally without third party APIs I wrote this tool to get familiar with CLIP model, I know many people have written similar tools with CLIP before, but I'm new to machine learning and writing a classic tool helps my study. The unusual thing with my version is, it is in pure Node.js, with the power of node-mlx, a Node.js machine learning framework. The repo in the link is mostly about implementing indexing and CLI, the code of the model implementation lives as a Node.js module: https://bit.ly/3XGFSXL . Hope this helps other learners! https://bit.ly/3XJviiE September 16, 2024 at 11:59AM

Show HN: JAQT – JavaScript Queries and Transformations https://bit.ly/4e7u7zA

Show HN: JAQT – JavaScript Queries and Transformations Hi all, I've made a javascript library to simplify searching/sorting/filtering in arrays of objects. Its inspired by both GraphQL and SQL, but implemented using javascript Proxies. Instead of creating a new language, its all just javascript. I've made it as part of an experimental database, which uses javascript as the query engine. The normal javascript map/reduce/sort functions are quite difficult to master for junior developers. JAQT is easier to explain, and can still be used in combination with any existing array functions. Please let me know what you think of the API and its ease of use! https://bit.ly/4d6K0F8 September 16, 2024 at 10:08AM

Show HN: Looksmax AI – Receive Personalized Tips to Enhance Your Appearance https://bit.ly/47vSKn6

Show HN: Looksmax AI – Receive Personalized Tips to Enhance Your Appearance https://bit.ly/3XqLl3x September 16, 2024 at 05:36AM

Show HN: CAD with AI https://bit.ly/3ZxezQS

Show HN: CAD with AI https://bit.ly/3ZrAysD September 16, 2024 at 02:26AM

Show HN: I build Figma plugin for unlimited themes/modes https://bit.ly/4d4dHXb

Show HN: I build Figma plugin for unlimited themes/modes Hi everyone, I’ve created a new Token Swapper plugin for Figma to address limitations I encountered while working on a multi-brand mobile app UI. The Figma’s Pro plan’s 4 Modes (or even the Enterprise plan’s 40 Modes) weren’t sufficient for my needs. I made Token Swapper to quickly switch between themes (based on variables and styles) without relying on Modes. The only requirement is that you organize your themes in top-level folders within both the variables and styles sections. I’d love for you to try it out and share your thoughts. Your feedback is appreciated! https://bit.ly/3TvADrk September 14, 2024 at 08:01PM

Show HN: Wordllama – Things you can do with the token embeddings of an LLM https://bit.ly/3XIfooF

Show HN: Wordllama – Things you can do with the token embeddings of an LLM After working with LLMs for long enough, I found myself wanting a lightweight utility for doing various small tasks to prepare inputs, locate information and create evaluators. This library is two things: a very simple model and utilities that inference it (eg. fuzzy deduplication). The target platform is CPU, and it’s intended to be light, fast and pip installable — a library that lowers the barrier to working with strings semantically . You don’t need to install pytorch to use it, or any deep learning runtimes. How can this be accomplished? The model is simply token embeddings that are average pooled. To create this model, I extracted token embedding (nn.Embedding) vectors from LLMs, concatenated them along the embedding dimension, added a learnable weight parameter, and projected them to a smaller dimension. Using the sentence transformers framework and datasets, I trained the pooled embedding with multiple negatives ranking loss and matryoshka representation learning so they can be truncated. After training, the weights and projections are no longer needed, because there is no contextual calculations. I inference the entire token vocabulary and save the new token embeddings to be loaded to numpy. While the results are not impressive compared to transformer models, they perform well on MTEB benchmarks compared to word embedding models (which they are most similar to), while being much smaller in size (smallest model, 32k vocab, 64-dim is only 4MB). On the utility side, I’ve been adding some tools that I think it’ll be useful for. In addition to general embedding, there’s algorithms for ranking, filtering, clustering, deduplicating and similarity. Some of them have a cython implementation, and I’m continuing to work on benchmarking them and improving them as I have time. In addition to “standard” models that use cosine similarity for some algorithms, there are binarized models that use hamming distance. This is a slightly faster, similarity algorithm, with significantly less memory per embedding (float32 -> 1 bit). Hope you enjoy it, and find it useful. PS I haven’t figured out Windows builds yet, but Linux and Mac are supported. https://bit.ly/3XIfoFb September 15, 2024 at 04:25AM

Show HN: Insta2000 – Re-render Instagram locally, retro, and ad free https://bit.ly/4d1bMTi

Show HN: Insta2000 – Re-render Instagram locally, retro, and ad free I've been avoiding instagram for a few years, as it feels very noisy and addictive. I'm starting to feel out of touch from friends though, so got wondering if I could build a tool to get just my friends updates, without all the other junk. It turns out scraping my own profile and re-rendering a static webpage is quite a successful way to achieve that goal. It has no ads, no discovery injections, no trackers, no infinite scroll, no notifications. It does include stories and posts alongside each other. It does sort things based on recent updates. In the end, it proved better than I expected with a relatively small effort. I made it retro as well just for a fun throwback, and as a protest against the modern web! It leans very heavily on the [instaloader package]( https://bit.ly/47qNhhg I'm quite grateful for the contributors there! Repo is yours to use and build on, and feel free to contribute as well. https://bit.ly/3Txra30 September 15, 2024 at 02:01AM