The best Hacker News stories from Show from the past day

from

All Show

from the past*

day week month

with these search terms:

Create follow Go back

Latest posts:

Show HN: LearnGPT – Browse and share ChatGPT examples

Show HN: 0xFast – Faster Web3 APIs

Excited to showcase 0xFast to HN!<p>Built using a new indexing system designed for Web3 data, 0xFast outperforms the most popular web3 API platforms, while also being 3x cheaper.

Show HN: 0xFast – Faster Web3 APIs

Excited to showcase 0xFast to HN!<p>Built using a new indexing system designed for Web3 data, 0xFast outperforms the most popular web3 API platforms, while also being 3x cheaper.

Show HN: Wasp – DSL/framework for building full-stack web apps – now in beta

Hey HN! Wasp (<a href="https://wasp-lang.dev/" rel="nofollow">https://wasp-lang.dev/</a>) is a simple config language (DSL) and framework for building full-stack web apps. You describe the high-level features you want (auth, CRUD, async jobs, …) using the Wasp DSL, and write the rest of your logic in React, Node.js, and Prisma. We’re focused on simplifying developer experience and ensuring best practices. Everything is open source.<p>Why another full-stack framework? And why a config language/DSL? We were still experiencing a lot of boilerplate (repetitive tasks) using other frameworks—things like duplicating data models across database/server/client, implementing CRUD API, setting up auth, and choosing and stitching together all parts of the stack.<p>There are two main reasons for the DSL approach - 1) short-term: simpler and cleaner DX via a declarative language that helps avoid boilerplate, and 2) longer-tem: laying foundation for the stack & architecture independent system.<p>Since Wasp analyses the app’s requirements in compile time, it can decide how to generate the target code (React & Node.js currently). In the future it could support other stacks such as e.g. Vue/Svelte on the client and Python/Go on the server, even allowing for mixing’n’matching. The same goes for the architecture (dedicated server, serverless, …).<p>Our big vision for Wasp is to become a stable, stack-agnostic language for describing (web) app requirements (like SQL for databases or Terraform for infra) that interops with the existing stack. Wasp-lang stands for “Web Application SPecification language”.<p>Besides the DSL, another valid approach would be to offer an SDK in e.g. JS or Python to build Wasp AST (like Terraform and Pulumi now both offer). We see it as another “frontend” for constructing the AST and might also introduce it in the future.<p>Under the hood, everything is compiled to a client (React) and server (Node.js/Prisma) apps and we generate static files and a Docker image you can use for deploying to your platform of choice.<p>Wasp had an Alpha launch 1.5 years ago (<a href="https://news.ycombinator.com/item?id=26091956" rel="nofollow">https://news.ycombinator.com/item?id=26091956</a>). Now we are more stable and feature-full. We still expect things to change, so wouldn’t recommend using Wasp for heavy production or mission-critical systems just yet. But it has been used for hackathons, internal tools and even revenue-generating products (<a href="https://wasp-lang.dev/blog/2022/11/26/erlis-amicus-usecase" rel="nofollow">https://wasp-lang.dev/blog/2022/11/26/erlis-amicus-usecase</a>).<p>The current release is our biggest since we launched (<a href="https://wasp-lang.dev/blog/2022/11/29/wasp-beta" rel="nofollow">https://wasp-lang.dev/blog/2022/11/29/wasp-beta</a>). Besides general stability and DX improvements, it brings support for TypeScript (<a href="https://wasp-lang.dev/blog/2022/11/29/typescript-feature-announcement" rel="nofollow">https://wasp-lang.dev/blog/2022/11/29/typescript-feature-ann...</a>), Tailwind (<a href="https://wasp-lang.dev/blog/2022/11/16/tailwind-feature-announcement" rel="nofollow">https://wasp-lang.dev/blog/2022/11/16/tailwind-feature-annou...</a>), async jobs via pg-boss (<a href="https://wasp-lang.dev/blog/2022/06/15/jobs-feature-announcement" rel="nofollow">https://wasp-lang.dev/blog/2022/06/15/jobs-feature-announcem...</a>), full-stack authentication (now also with Google) (<a href="http://localhost:3000/blog/2022/11/15/auth-feature-announcement" rel="nofollow">http://localhost:3000/blog/2022/11/15/auth-feature-announcem...</a>), and by popular demand, Wasp LSP with VS Code integration (<a href="https://wasp-lang.dev/blog/2022/12/01/beta-ide-improvements" rel="nofollow">https://wasp-lang.dev/blog/2022/12/01/beta-ide-improvements</a>).<p>Our next focus will be on making Wasp even easier to use (examples, starter templates, UI helpers), and we’ll look into tighter weaving of data models with the rest of the stack and expanding the DSL with more functionalities.<p>We’re around to answer questions and look forward to hearing everything and anything you have to say!

Show HN: Wasp – DSL/framework for building full-stack web apps – now in beta

Hey HN! Wasp (<a href="https://wasp-lang.dev/" rel="nofollow">https://wasp-lang.dev/</a>) is a simple config language (DSL) and framework for building full-stack web apps. You describe the high-level features you want (auth, CRUD, async jobs, …) using the Wasp DSL, and write the rest of your logic in React, Node.js, and Prisma. We’re focused on simplifying developer experience and ensuring best practices. Everything is open source.<p>Why another full-stack framework? And why a config language/DSL? We were still experiencing a lot of boilerplate (repetitive tasks) using other frameworks—things like duplicating data models across database/server/client, implementing CRUD API, setting up auth, and choosing and stitching together all parts of the stack.<p>There are two main reasons for the DSL approach - 1) short-term: simpler and cleaner DX via a declarative language that helps avoid boilerplate, and 2) longer-tem: laying foundation for the stack & architecture independent system.<p>Since Wasp analyses the app’s requirements in compile time, it can decide how to generate the target code (React & Node.js currently). In the future it could support other stacks such as e.g. Vue/Svelte on the client and Python/Go on the server, even allowing for mixing’n’matching. The same goes for the architecture (dedicated server, serverless, …).<p>Our big vision for Wasp is to become a stable, stack-agnostic language for describing (web) app requirements (like SQL for databases or Terraform for infra) that interops with the existing stack. Wasp-lang stands for “Web Application SPecification language”.<p>Besides the DSL, another valid approach would be to offer an SDK in e.g. JS or Python to build Wasp AST (like Terraform and Pulumi now both offer). We see it as another “frontend” for constructing the AST and might also introduce it in the future.<p>Under the hood, everything is compiled to a client (React) and server (Node.js/Prisma) apps and we generate static files and a Docker image you can use for deploying to your platform of choice.<p>Wasp had an Alpha launch 1.5 years ago (<a href="https://news.ycombinator.com/item?id=26091956" rel="nofollow">https://news.ycombinator.com/item?id=26091956</a>). Now we are more stable and feature-full. We still expect things to change, so wouldn’t recommend using Wasp for heavy production or mission-critical systems just yet. But it has been used for hackathons, internal tools and even revenue-generating products (<a href="https://wasp-lang.dev/blog/2022/11/26/erlis-amicus-usecase" rel="nofollow">https://wasp-lang.dev/blog/2022/11/26/erlis-amicus-usecase</a>).<p>The current release is our biggest since we launched (<a href="https://wasp-lang.dev/blog/2022/11/29/wasp-beta" rel="nofollow">https://wasp-lang.dev/blog/2022/11/29/wasp-beta</a>). Besides general stability and DX improvements, it brings support for TypeScript (<a href="https://wasp-lang.dev/blog/2022/11/29/typescript-feature-announcement" rel="nofollow">https://wasp-lang.dev/blog/2022/11/29/typescript-feature-ann...</a>), Tailwind (<a href="https://wasp-lang.dev/blog/2022/11/16/tailwind-feature-announcement" rel="nofollow">https://wasp-lang.dev/blog/2022/11/16/tailwind-feature-annou...</a>), async jobs via pg-boss (<a href="https://wasp-lang.dev/blog/2022/06/15/jobs-feature-announcement" rel="nofollow">https://wasp-lang.dev/blog/2022/06/15/jobs-feature-announcem...</a>), full-stack authentication (now also with Google) (<a href="http://localhost:3000/blog/2022/11/15/auth-feature-announcement" rel="nofollow">http://localhost:3000/blog/2022/11/15/auth-feature-announcem...</a>), and by popular demand, Wasp LSP with VS Code integration (<a href="https://wasp-lang.dev/blog/2022/12/01/beta-ide-improvements" rel="nofollow">https://wasp-lang.dev/blog/2022/12/01/beta-ide-improvements</a>).<p>Our next focus will be on making Wasp even easier to use (examples, starter templates, UI helpers), and we’ll look into tighter weaving of data models with the rest of the stack and expanding the DSL with more functionalities.<p>We’re around to answer questions and look forward to hearing everything and anything you have to say!

Show HN: Ezy – open-source gRPC client, alternative to Postman and Insomnia

In this release I have concentrated on user experience:<p>- Full support of shortcuts - Reworked collections management - Notifications - Improved UX<p>gRPC clients I’ve worked with had drawbacks and didn’t fit my use-case in a way I was expecting, since I’ve started working with gRPC 3+ years ago.<p>Since then, I wanted a tool that fits any need in gRPC world. This is why I created ezy.<p>Compared to Insomnia and Postman, ezy offers better streams support, allows you to use Server-Side and Mutual TLS with custom TLS certificates, works with gRPC-Web and has a more slick UI/UX.<p>If you are looking for a gRPC/gRPC-Web client which fits your needs, give ezy a chance!<p>I’d love to hear your feedback and answer any questions regarding ezy.

Show HN: Ezy – open-source gRPC client, alternative to Postman and Insomnia

In this release I have concentrated on user experience:<p>- Full support of shortcuts - Reworked collections management - Notifications - Improved UX<p>gRPC clients I’ve worked with had drawbacks and didn’t fit my use-case in a way I was expecting, since I’ve started working with gRPC 3+ years ago.<p>Since then, I wanted a tool that fits any need in gRPC world. This is why I created ezy.<p>Compared to Insomnia and Postman, ezy offers better streams support, allows you to use Server-Side and Mutual TLS with custom TLS certificates, works with gRPC-Web and has a more slick UI/UX.<p>If you are looking for a gRPC/gRPC-Web client which fits your needs, give ezy a chance!<p>I’d love to hear your feedback and answer any questions regarding ezy.

Show HN: Ezy – open-source gRPC client, alternative to Postman and Insomnia

In this release I have concentrated on user experience:<p>- Full support of shortcuts - Reworked collections management - Notifications - Improved UX<p>gRPC clients I’ve worked with had drawbacks and didn’t fit my use-case in a way I was expecting, since I’ve started working with gRPC 3+ years ago.<p>Since then, I wanted a tool that fits any need in gRPC world. This is why I created ezy.<p>Compared to Insomnia and Postman, ezy offers better streams support, allows you to use Server-Side and Mutual TLS with custom TLS certificates, works with gRPC-Web and has a more slick UI/UX.<p>If you are looking for a gRPC/gRPC-Web client which fits your needs, give ezy a chance!<p>I’d love to hear your feedback and answer any questions regarding ezy.

Show HN: Web search using a ChatGPT-like model that can cite its sources

We’ve trained a generative AI model to browse the web and answer questions/retrieve code snippets directly. Unlike ChatGPT, it has access to primary sources and is able to cite them when you hover over an answer (click on the text to go to the source being cited). We also show regular Bing results side-by-side with our AI answer.<p>The model is an 11-billion parameter T5-derivative that has been fine-tuned on feedback given on hundreds of thousands of searches done (anonymously) on our platform. Giving the model web access lessens its burden to need to store a snapshot of human knowledge within its parameters. Rather, it knows how to piece together primary sources in a natural and informative way. Using our own model is also an order of magnitude cheaper than relying on GPT.<p>A drawback to aligning models to web results is that they are less inclined to generate complete solutions/answers to questions where good primary sources don’t exist. Answers generated without underlying citable sources can be more creative but are prone to errors. In the future, we will show both types of answers.<p>Examples:<p><a href="https://beta.sayhello.so/search?q=set+cookie+in+fastapi" rel="nofollow">https://beta.sayhello.so/search?q=set+cookie+in+fastapi</a><p><a href="https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn+from+users" rel="nofollow">https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn...</a><p><a href="https://beta.sayhello.so/search?q=How+to+get+command+line+parameters+in+Rust" rel="nofollow">https://beta.sayhello.so/search?q=How+to+get+command+line+pa...</a><p><a href="https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twitter" rel="nofollow">https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twit...</a><p>Would love to hear your thoughts.

Show HN: Web search using a ChatGPT-like model that can cite its sources

We’ve trained a generative AI model to browse the web and answer questions/retrieve code snippets directly. Unlike ChatGPT, it has access to primary sources and is able to cite them when you hover over an answer (click on the text to go to the source being cited). We also show regular Bing results side-by-side with our AI answer.<p>The model is an 11-billion parameter T5-derivative that has been fine-tuned on feedback given on hundreds of thousands of searches done (anonymously) on our platform. Giving the model web access lessens its burden to need to store a snapshot of human knowledge within its parameters. Rather, it knows how to piece together primary sources in a natural and informative way. Using our own model is also an order of magnitude cheaper than relying on GPT.<p>A drawback to aligning models to web results is that they are less inclined to generate complete solutions/answers to questions where good primary sources don’t exist. Answers generated without underlying citable sources can be more creative but are prone to errors. In the future, we will show both types of answers.<p>Examples:<p><a href="https://beta.sayhello.so/search?q=set+cookie+in+fastapi" rel="nofollow">https://beta.sayhello.so/search?q=set+cookie+in+fastapi</a><p><a href="https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn+from+users" rel="nofollow">https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn...</a><p><a href="https://beta.sayhello.so/search?q=How+to+get+command+line+parameters+in+Rust" rel="nofollow">https://beta.sayhello.so/search?q=How+to+get+command+line+pa...</a><p><a href="https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twitter" rel="nofollow">https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twit...</a><p>Would love to hear your thoughts.

Show HN: Web search using a ChatGPT-like model that can cite its sources

We’ve trained a generative AI model to browse the web and answer questions/retrieve code snippets directly. Unlike ChatGPT, it has access to primary sources and is able to cite them when you hover over an answer (click on the text to go to the source being cited). We also show regular Bing results side-by-side with our AI answer.<p>The model is an 11-billion parameter T5-derivative that has been fine-tuned on feedback given on hundreds of thousands of searches done (anonymously) on our platform. Giving the model web access lessens its burden to need to store a snapshot of human knowledge within its parameters. Rather, it knows how to piece together primary sources in a natural and informative way. Using our own model is also an order of magnitude cheaper than relying on GPT.<p>A drawback to aligning models to web results is that they are less inclined to generate complete solutions/answers to questions where good primary sources don’t exist. Answers generated without underlying citable sources can be more creative but are prone to errors. In the future, we will show both types of answers.<p>Examples:<p><a href="https://beta.sayhello.so/search?q=set+cookie+in+fastapi" rel="nofollow">https://beta.sayhello.so/search?q=set+cookie+in+fastapi</a><p><a href="https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn+from+users" rel="nofollow">https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn...</a><p><a href="https://beta.sayhello.so/search?q=How+to+get+command+line+parameters+in+Rust" rel="nofollow">https://beta.sayhello.so/search?q=How+to+get+command+line+pa...</a><p><a href="https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twitter" rel="nofollow">https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twit...</a><p>Would love to hear your thoughts.

Show HN: Web search using a ChatGPT-like model that can cite its sources

We’ve trained a generative AI model to browse the web and answer questions/retrieve code snippets directly. Unlike ChatGPT, it has access to primary sources and is able to cite them when you hover over an answer (click on the text to go to the source being cited). We also show regular Bing results side-by-side with our AI answer.<p>The model is an 11-billion parameter T5-derivative that has been fine-tuned on feedback given on hundreds of thousands of searches done (anonymously) on our platform. Giving the model web access lessens its burden to need to store a snapshot of human knowledge within its parameters. Rather, it knows how to piece together primary sources in a natural and informative way. Using our own model is also an order of magnitude cheaper than relying on GPT.<p>A drawback to aligning models to web results is that they are less inclined to generate complete solutions/answers to questions where good primary sources don’t exist. Answers generated without underlying citable sources can be more creative but are prone to errors. In the future, we will show both types of answers.<p>Examples:<p><a href="https://beta.sayhello.so/search?q=set+cookie+in+fastapi" rel="nofollow">https://beta.sayhello.so/search?q=set+cookie+in+fastapi</a><p><a href="https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn+from+users" rel="nofollow">https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn...</a><p><a href="https://beta.sayhello.so/search?q=How+to+get+command+line+parameters+in+Rust" rel="nofollow">https://beta.sayhello.so/search?q=How+to+get+command+line+pa...</a><p><a href="https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twitter" rel="nofollow">https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twit...</a><p>Would love to hear your thoughts.

Show HN: Tenebra game PC port of popular Commodore 64 game

Guide the hapless protagonist to the exit, while keeping in mind that he is afraid of darkness and refuses to walk in the dark areas.

Show HN: TromPhone, a Trombone for Your Phone

A few months ago I had a silly idea of making a mobile app that used the accelerometer to track the slide motion for playing a virtual trombone. Just wanted to share the story of bringing it to fruition here on hn.<p>I started out spending a couple days trying to get something cross-platform going in Flutter, but it soon became clear that wasn't the best fit, seeing as I'd need native hooks for most of what the app needed to do, and it wasn't yet clear it'd be possible at all. So I switched to making it an iOS app in Swift. The accelerometer data turned out to be not nearly accurate enough to do the job, so I switched to using the camera/AR using ARKit... and it worked instantly. Like the very first time I hooked up a slider UI element to the distance function. It felt a bit like magic. And also just ridiculous.<p>Here's a video I recorded to send to some friends at the time: <a href="https://youtu.be/6BIogfGH3IQ" rel="nofollow">https://youtu.be/6BIogfGH3IQ</a><p>Here's a video of it in action in it's current state: <a href="https://youtube.com/shorts/8kS2TRzV4I4?feature=share" rel="nofollow">https://youtube.com/shorts/8kS2TRzV4I4?feature=share</a><p>Apologies for the non-responsive website (using nextjs on cloudflare). It doesn't look great on mobile, which is kind of inexcusable, I'm working on it: <a href="https://www.tromphoneapp.com" rel="nofollow">https://www.tromphoneapp.com</a><p>Anyway, I'm not sure where I'll take it from here. I have some ideas for more AR content like hats, heart eyes, etc. Possibly a song editor so users can add songs that might have issues with copyright if I included them in the app? Any ideas you guys have would be fun to hear.

Show HN: Publish from GitHub Actions using multi-factor authentication

The backstory about this GitHub Action:<p>I discussed with an open-source maintainer why they publish npm packages from their local machine and do not use CI/CD pipelines.<p>They said publishing should require human intervention and want to continue using multi-factor authentication to publish to the npm registry.<p>This led to building the wait-for-secrets GitHub Action. It prints a URL in the build log and waits for secrets to be entered using a browser. Once entered, the workflow continues, and secrets can be used in future steps.<p>The latest release of "eslint-plugin-react" to the npm registry used a one-time password (OTP) from a GitHub Actions workflow! <a href="https://github.com/jsx-eslint/eslint-plugin-react/actions/runs/3498968497/jobs/5860126389#step:9:1" rel="nofollow">https://github.com/jsx-eslint/eslint-plugin-react/actions/ru...</a>

Show HN: Port of OpenAI's Whisper model in C/C++

Hi HN,<p>OpenAI recently released a model for automatic speech recognition called Whisper [0]. I decided to reimplement the inference of the model from scratch using C/C++. To achieve this I implemented a minimalistic tensor library in C and ported the high-level architecture of the model in C++. The entire code is less than 8000 lines of code and is contained in just 2 source files without any third-party dependencies. The Github project is here:<p><a href="https://github.com/ggerganov/whisper.cpp" rel="nofollow">https://github.com/ggerganov/whisper.cpp</a><p>With this implementation I can very easily build and run the model - <i>“make base.en”</i>. It also allows me to run it on a wide range of devices. For example, I have provided examples of running the model on an iPhone, Raspberry Pi 4 and even in a web page via WebAssembly!<p>The implementation runs fully on the CPU and utilizes FP16, AVX intrinsics on x86 architectures and NEON + Accelerate framework on Apple Silicon. The latter is especially efficient and I observe that the inference is about 2-3 times faster compared to the current PyTorch implementation provided by OpenAI when running it on my MacBook M1 Pro. The WASM port utilizes SIMD 128-bit intrinsics - a feature supported in some modern web browsers [1].<p>I am very happy with the performance that I observe on Apple Silicon devices. I didn’t expect that the Accelerate framework [2] (i.e. CBLAS) offers such a dramatic performance boost for matrix multiplications so I was very pleasantly surprised! To enable the framework in your C/C++ projects, all you have to do is add <i>`-framework Accelerate`</i> to your clang command-line flags.<p>This entire exercise of implementing the Whisper model was very interesting to me and helped me understand a lot about how the transformer architecture works. I also got a lot of positive feedback from people finding and using my project. We brainstormed on a lot of interesting tools that can potentially be created with this library (such as speech-to-text plugin for Vim, RPi4 voice assistant, WASM chat bot, etc). If interested, checkout the “Examples” section and the “Show and tell” discussions for some ideas!<p>Would love to know what you think about this project and about your experience with using the Accelerate framework in any of your projects. Cheers!<p>[0] <a href="https://github.com/openai/whisper" rel="nofollow">https://github.com/openai/whisper</a><p>[1] <a href="https://chromestatus.com/feature/6533147810332672" rel="nofollow">https://chromestatus.com/feature/6533147810332672</a><p>[2] <a href="https://developer.apple.com/documentation/accelerate" rel="nofollow">https://developer.apple.com/documentation/accelerate</a>

Show HN: Port of OpenAI's Whisper model in C/C++

Hi HN,<p>OpenAI recently released a model for automatic speech recognition called Whisper [0]. I decided to reimplement the inference of the model from scratch using C/C++. To achieve this I implemented a minimalistic tensor library in C and ported the high-level architecture of the model in C++. The entire code is less than 8000 lines of code and is contained in just 2 source files without any third-party dependencies. The Github project is here:<p><a href="https://github.com/ggerganov/whisper.cpp" rel="nofollow">https://github.com/ggerganov/whisper.cpp</a><p>With this implementation I can very easily build and run the model - <i>“make base.en”</i>. It also allows me to run it on a wide range of devices. For example, I have provided examples of running the model on an iPhone, Raspberry Pi 4 and even in a web page via WebAssembly!<p>The implementation runs fully on the CPU and utilizes FP16, AVX intrinsics on x86 architectures and NEON + Accelerate framework on Apple Silicon. The latter is especially efficient and I observe that the inference is about 2-3 times faster compared to the current PyTorch implementation provided by OpenAI when running it on my MacBook M1 Pro. The WASM port utilizes SIMD 128-bit intrinsics - a feature supported in some modern web browsers [1].<p>I am very happy with the performance that I observe on Apple Silicon devices. I didn’t expect that the Accelerate framework [2] (i.e. CBLAS) offers such a dramatic performance boost for matrix multiplications so I was very pleasantly surprised! To enable the framework in your C/C++ projects, all you have to do is add <i>`-framework Accelerate`</i> to your clang command-line flags.<p>This entire exercise of implementing the Whisper model was very interesting to me and helped me understand a lot about how the transformer architecture works. I also got a lot of positive feedback from people finding and using my project. We brainstormed on a lot of interesting tools that can potentially be created with this library (such as speech-to-text plugin for Vim, RPi4 voice assistant, WASM chat bot, etc). If interested, checkout the “Examples” section and the “Show and tell” discussions for some ideas!<p>Would love to know what you think about this project and about your experience with using the Accelerate framework in any of your projects. Cheers!<p>[0] <a href="https://github.com/openai/whisper" rel="nofollow">https://github.com/openai/whisper</a><p>[1] <a href="https://chromestatus.com/feature/6533147810332672" rel="nofollow">https://chromestatus.com/feature/6533147810332672</a><p>[2] <a href="https://developer.apple.com/documentation/accelerate" rel="nofollow">https://developer.apple.com/documentation/accelerate</a>

Show HN: Port of OpenAI's Whisper model in C/C++

Hi HN,<p>OpenAI recently released a model for automatic speech recognition called Whisper [0]. I decided to reimplement the inference of the model from scratch using C/C++. To achieve this I implemented a minimalistic tensor library in C and ported the high-level architecture of the model in C++. The entire code is less than 8000 lines of code and is contained in just 2 source files without any third-party dependencies. The Github project is here:<p><a href="https://github.com/ggerganov/whisper.cpp" rel="nofollow">https://github.com/ggerganov/whisper.cpp</a><p>With this implementation I can very easily build and run the model - <i>“make base.en”</i>. It also allows me to run it on a wide range of devices. For example, I have provided examples of running the model on an iPhone, Raspberry Pi 4 and even in a web page via WebAssembly!<p>The implementation runs fully on the CPU and utilizes FP16, AVX intrinsics on x86 architectures and NEON + Accelerate framework on Apple Silicon. The latter is especially efficient and I observe that the inference is about 2-3 times faster compared to the current PyTorch implementation provided by OpenAI when running it on my MacBook M1 Pro. The WASM port utilizes SIMD 128-bit intrinsics - a feature supported in some modern web browsers [1].<p>I am very happy with the performance that I observe on Apple Silicon devices. I didn’t expect that the Accelerate framework [2] (i.e. CBLAS) offers such a dramatic performance boost for matrix multiplications so I was very pleasantly surprised! To enable the framework in your C/C++ projects, all you have to do is add <i>`-framework Accelerate`</i> to your clang command-line flags.<p>This entire exercise of implementing the Whisper model was very interesting to me and helped me understand a lot about how the transformer architecture works. I also got a lot of positive feedback from people finding and using my project. We brainstormed on a lot of interesting tools that can potentially be created with this library (such as speech-to-text plugin for Vim, RPi4 voice assistant, WASM chat bot, etc). If interested, checkout the “Examples” section and the “Show and tell” discussions for some ideas!<p>Would love to know what you think about this project and about your experience with using the Accelerate framework in any of your projects. Cheers!<p>[0] <a href="https://github.com/openai/whisper" rel="nofollow">https://github.com/openai/whisper</a><p>[1] <a href="https://chromestatus.com/feature/6533147810332672" rel="nofollow">https://chromestatus.com/feature/6533147810332672</a><p>[2] <a href="https://developer.apple.com/documentation/accelerate" rel="nofollow">https://developer.apple.com/documentation/accelerate</a>

Show HN: Port of OpenAI's Whisper model in C/C++

Hi HN,<p>OpenAI recently released a model for automatic speech recognition called Whisper [0]. I decided to reimplement the inference of the model from scratch using C/C++. To achieve this I implemented a minimalistic tensor library in C and ported the high-level architecture of the model in C++. The entire code is less than 8000 lines of code and is contained in just 2 source files without any third-party dependencies. The Github project is here:<p><a href="https://github.com/ggerganov/whisper.cpp" rel="nofollow">https://github.com/ggerganov/whisper.cpp</a><p>With this implementation I can very easily build and run the model - <i>“make base.en”</i>. It also allows me to run it on a wide range of devices. For example, I have provided examples of running the model on an iPhone, Raspberry Pi 4 and even in a web page via WebAssembly!<p>The implementation runs fully on the CPU and utilizes FP16, AVX intrinsics on x86 architectures and NEON + Accelerate framework on Apple Silicon. The latter is especially efficient and I observe that the inference is about 2-3 times faster compared to the current PyTorch implementation provided by OpenAI when running it on my MacBook M1 Pro. The WASM port utilizes SIMD 128-bit intrinsics - a feature supported in some modern web browsers [1].<p>I am very happy with the performance that I observe on Apple Silicon devices. I didn’t expect that the Accelerate framework [2] (i.e. CBLAS) offers such a dramatic performance boost for matrix multiplications so I was very pleasantly surprised! To enable the framework in your C/C++ projects, all you have to do is add <i>`-framework Accelerate`</i> to your clang command-line flags.<p>This entire exercise of implementing the Whisper model was very interesting to me and helped me understand a lot about how the transformer architecture works. I also got a lot of positive feedback from people finding and using my project. We brainstormed on a lot of interesting tools that can potentially be created with this library (such as speech-to-text plugin for Vim, RPi4 voice assistant, WASM chat bot, etc). If interested, checkout the “Examples” section and the “Show and tell” discussions for some ideas!<p>Would love to know what you think about this project and about your experience with using the Accelerate framework in any of your projects. Cheers!<p>[0] <a href="https://github.com/openai/whisper" rel="nofollow">https://github.com/openai/whisper</a><p>[1] <a href="https://chromestatus.com/feature/6533147810332672" rel="nofollow">https://chromestatus.com/feature/6533147810332672</a><p>[2] <a href="https://developer.apple.com/documentation/accelerate" rel="nofollow">https://developer.apple.com/documentation/accelerate</a>

Show HN: Codeium – a free, fast AI codegen extension

I'm Varun, CEO of Exafunction, and we just released Codeium to open up access of generative AI to all developers for free. In the spirit of Show HN, we created a playground version for anyone to try this tech in the browser (click Try in Browser)!<p>We have built scalable, low-latency ML infra for many top AI companies in the past, and we are excited to leverage that tech into a product that we, as developers, would love. We hope that you do too, and we would appreciate any feedback that this community has for us!