The best Hacker News stories from Show from the past day
Latest posts:
Show HN: Desbordante 2.0 – A high-performance data profiler
Hi! We are excited to announce the second release of Desbordante — an open-source, high-performance data profiler that is capable of discovering and validating many different patterns in data using various algorithms.<p>Unlike existing data profilers, Desbordante focuses on discovering complex patterns in data, which are notoriously hard to extract. Since its inception in 2019, it has become the fastest open-source tool for these tasks. It also offers an array of patterns which have no alternative implementations. With this release, Desbordante now supports 17 types of patterns, such as: various types of functional dependencies, inclusion and order dependencies, fuzzy algebraic constraints and many others.<p>Some ways in which Desbordante can be helpful are:
1) Hypothesis generation for scientists that work with large volumes of data.
2) Business data owners and business analysts can benefit from hypothesis generation as well as data quality improvement: cleaning databases from errors, finding and removing inexact duplicates, and so on.
3) Found primitives can help data scientists in feature engineering and choosing the right direction for ablation studies.<p>Desbordante solves two types of tasks: Discovery and Validation. The Discovery task is designed to identify all instances of a specified pattern type of a given dataset. The Validation task is different: it is designed to check whether a specified pattern instance is present in a given dataset. This task not only returns True or False, but it also explains why the instance does not hold (e.g. it can list table rows with conflicting values).<p>Desbordante offers a CLI, a web application, and a Python library. The latter makes it possible to construct ad-hoc data analysis pipelines — essentially, your own applications for various data quality tasks: data cleaning, data deduplication, anomaly detection, data schema recovery and many others. You can check out example implementations here: <a href="https://github.com/Desbordante/desbordante-core/tree/main/examples">https://github.com/Desbordante/desbordante-core/tree/main/ex...</a>.<p>Check out some of our articles for more details:<p><a href="https://medium.com/@chernishev/exploratory-data-analysis-with-desbordante-4b97299cce07" rel="nofollow">https://medium.com/@chernishev/exploratory-data-analysis-wit...</a><p><a href="https://itnext.io/building-a-simple-data-cleaning-application-with-desbordante-e4897dcd4c5d" rel="nofollow">https://itnext.io/building-a-simple-data-cleaning-applicatio...</a><p><a href="https://levelup.gitconnected.com/checking-mining-and-exploring-functional-dependencies-in-python-903bb0e26d5d" rel="nofollow">https://levelup.gitconnected.com/checking-mining-and-explori...</a><p>This major release brings a lot of improvements: support for several novel patterns, support for novel data type — graphs, added python bindings for existing patterns, better guides and examples and more. The detailed changelog can be seen here (<a href="https://github.com/Desbordante/desbordante-core/releases/tag/v2.0.0">https://github.com/Desbordante/desbordante-core/releases/tag...</a>).
Show HN: Desbordante 2.0 – A high-performance data profiler
Hi! We are excited to announce the second release of Desbordante — an open-source, high-performance data profiler that is capable of discovering and validating many different patterns in data using various algorithms.<p>Unlike existing data profilers, Desbordante focuses on discovering complex patterns in data, which are notoriously hard to extract. Since its inception in 2019, it has become the fastest open-source tool for these tasks. It also offers an array of patterns which have no alternative implementations. With this release, Desbordante now supports 17 types of patterns, such as: various types of functional dependencies, inclusion and order dependencies, fuzzy algebraic constraints and many others.<p>Some ways in which Desbordante can be helpful are:
1) Hypothesis generation for scientists that work with large volumes of data.
2) Business data owners and business analysts can benefit from hypothesis generation as well as data quality improvement: cleaning databases from errors, finding and removing inexact duplicates, and so on.
3) Found primitives can help data scientists in feature engineering and choosing the right direction for ablation studies.<p>Desbordante solves two types of tasks: Discovery and Validation. The Discovery task is designed to identify all instances of a specified pattern type of a given dataset. The Validation task is different: it is designed to check whether a specified pattern instance is present in a given dataset. This task not only returns True or False, but it also explains why the instance does not hold (e.g. it can list table rows with conflicting values).<p>Desbordante offers a CLI, a web application, and a Python library. The latter makes it possible to construct ad-hoc data analysis pipelines — essentially, your own applications for various data quality tasks: data cleaning, data deduplication, anomaly detection, data schema recovery and many others. You can check out example implementations here: <a href="https://github.com/Desbordante/desbordante-core/tree/main/examples">https://github.com/Desbordante/desbordante-core/tree/main/ex...</a>.<p>Check out some of our articles for more details:<p><a href="https://medium.com/@chernishev/exploratory-data-analysis-with-desbordante-4b97299cce07" rel="nofollow">https://medium.com/@chernishev/exploratory-data-analysis-wit...</a><p><a href="https://itnext.io/building-a-simple-data-cleaning-application-with-desbordante-e4897dcd4c5d" rel="nofollow">https://itnext.io/building-a-simple-data-cleaning-applicatio...</a><p><a href="https://levelup.gitconnected.com/checking-mining-and-exploring-functional-dependencies-in-python-903bb0e26d5d" rel="nofollow">https://levelup.gitconnected.com/checking-mining-and-explori...</a><p>This major release brings a lot of improvements: support for several novel patterns, support for novel data type — graphs, added python bindings for existing patterns, better guides and examples and more. The detailed changelog can be seen here (<a href="https://github.com/Desbordante/desbordante-core/releases/tag/v2.0.0">https://github.com/Desbordante/desbordante-core/releases/tag...</a>).
Show HN: Desbordante 2.0 – A high-performance data profiler
Hi! We are excited to announce the second release of Desbordante — an open-source, high-performance data profiler that is capable of discovering and validating many different patterns in data using various algorithms.<p>Unlike existing data profilers, Desbordante focuses on discovering complex patterns in data, which are notoriously hard to extract. Since its inception in 2019, it has become the fastest open-source tool for these tasks. It also offers an array of patterns which have no alternative implementations. With this release, Desbordante now supports 17 types of patterns, such as: various types of functional dependencies, inclusion and order dependencies, fuzzy algebraic constraints and many others.<p>Some ways in which Desbordante can be helpful are:
1) Hypothesis generation for scientists that work with large volumes of data.
2) Business data owners and business analysts can benefit from hypothesis generation as well as data quality improvement: cleaning databases from errors, finding and removing inexact duplicates, and so on.
3) Found primitives can help data scientists in feature engineering and choosing the right direction for ablation studies.<p>Desbordante solves two types of tasks: Discovery and Validation. The Discovery task is designed to identify all instances of a specified pattern type of a given dataset. The Validation task is different: it is designed to check whether a specified pattern instance is present in a given dataset. This task not only returns True or False, but it also explains why the instance does not hold (e.g. it can list table rows with conflicting values).<p>Desbordante offers a CLI, a web application, and a Python library. The latter makes it possible to construct ad-hoc data analysis pipelines — essentially, your own applications for various data quality tasks: data cleaning, data deduplication, anomaly detection, data schema recovery and many others. You can check out example implementations here: <a href="https://github.com/Desbordante/desbordante-core/tree/main/examples">https://github.com/Desbordante/desbordante-core/tree/main/ex...</a>.<p>Check out some of our articles for more details:<p><a href="https://medium.com/@chernishev/exploratory-data-analysis-with-desbordante-4b97299cce07" rel="nofollow">https://medium.com/@chernishev/exploratory-data-analysis-wit...</a><p><a href="https://itnext.io/building-a-simple-data-cleaning-application-with-desbordante-e4897dcd4c5d" rel="nofollow">https://itnext.io/building-a-simple-data-cleaning-applicatio...</a><p><a href="https://levelup.gitconnected.com/checking-mining-and-exploring-functional-dependencies-in-python-903bb0e26d5d" rel="nofollow">https://levelup.gitconnected.com/checking-mining-and-explori...</a><p>This major release brings a lot of improvements: support for several novel patterns, support for novel data type — graphs, added python bindings for existing patterns, better guides and examples and more. The detailed changelog can be seen here (<a href="https://github.com/Desbordante/desbordante-core/releases/tag/v2.0.0">https://github.com/Desbordante/desbordante-core/releases/tag...</a>).
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.<p>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:<p>• bulk actions are performed in parallel<p>• syncing the entire library locally provides incredible search performances and the ability to query descriptions and other previously unaccessible fields<p>• maintenance scripts run automatically for you<p>• last but not least... app icons (or favicons when not available) let you quickly identify what you're looking for<p>The app is written in Swift and uses only AppKit with Nib files for top performance, pixel perfect design, and maximum flexibility.<p>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].<p>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:<p>• Import/Export<p>• Automatic replacement of apps installed without Homebrew<p>• Notifications about available updates<p>• Finder actions<p>• Spotlight integration<p>I hope you find Brewer X interesting. We're happy to answer any question!<p>[^1]: <a href="https://www.sketch.com/blog/how-we-redesigned-our-toolbar-icons-for-big-sur-and-monterey/" rel="nofollow">https://www.sketch.com/blog/how-we-redesigned-our-toolbar-ic...</a><p>[^2]: <a href="https://www.appiconbook.com/" rel="nofollow">https://www.appiconbook.com/</a>
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.<p>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:<p>• bulk actions are performed in parallel<p>• syncing the entire library locally provides incredible search performances and the ability to query descriptions and other previously unaccessible fields<p>• maintenance scripts run automatically for you<p>• last but not least... app icons (or favicons when not available) let you quickly identify what you're looking for<p>The app is written in Swift and uses only AppKit with Nib files for top performance, pixel perfect design, and maximum flexibility.<p>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].<p>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:<p>• Import/Export<p>• Automatic replacement of apps installed without Homebrew<p>• Notifications about available updates<p>• Finder actions<p>• Spotlight integration<p>I hope you find Brewer X interesting. We're happy to answer any question!<p>[^1]: <a href="https://www.sketch.com/blog/how-we-redesigned-our-toolbar-icons-for-big-sur-and-monterey/" rel="nofollow">https://www.sketch.com/blog/how-we-redesigned-our-toolbar-ic...</a><p>[^2]: <a href="https://www.appiconbook.com/" rel="nofollow">https://www.appiconbook.com/</a>
Show HN: Vapi – Convince our voice AI to give you the secret code
Show HN: Vapi – Convince our voice AI to give you the secret code
Show HN: Vapi – Convince our voice AI to give you the secret code
Show HN: A no-watermark video downloader for Any social media
Hi HN reader,<p>I'm a solo-founder navigating product market fit. I've spent 3 years building Double Subtitles which I am very proud of, but I don't know how to do marketing and my growth is somewhat stagnant. Through learning about SEO, I discovered this opportunity that I want to share:<p>There are a lot of websites that let you download social media videos, which have extremely high traffic. But they are all disjoint, in the sense that some download TikTok videos, while others download Instagram videos, others YouTube, etc...<p>I quickly prototyped software which downloads any video from any link. Currently the live version of FastDownload only supports TikTok and YouTube videos, since they are the most stable. But I will slowly start rolling out Instagram and other services weekly.<p>ps. Personally, what I think is really cool is this Apple Shortcut, which lets you download videos directly from within whatever app you're in, using Share Action Extensions.<p>pps. Another thing I'm very excited about is that while doing analysis with Ahrefs and SEMRush I have been able to identify many high value keywords which I'm deploying pSEO strategies for now.<p>I hope this product is helpful and any feedback (even negative) would bring me joy.<p>All the best,
Bojan
Show HN: A no-watermark video downloader for Any social media
Hi HN reader,<p>I'm a solo-founder navigating product market fit. I've spent 3 years building Double Subtitles which I am very proud of, but I don't know how to do marketing and my growth is somewhat stagnant. Through learning about SEO, I discovered this opportunity that I want to share:<p>There are a lot of websites that let you download social media videos, which have extremely high traffic. But they are all disjoint, in the sense that some download TikTok videos, while others download Instagram videos, others YouTube, etc...<p>I quickly prototyped software which downloads any video from any link. Currently the live version of FastDownload only supports TikTok and YouTube videos, since they are the most stable. But I will slowly start rolling out Instagram and other services weekly.<p>ps. Personally, what I think is really cool is this Apple Shortcut, which lets you download videos directly from within whatever app you're in, using Share Action Extensions.<p>pps. Another thing I'm very excited about is that while doing analysis with Ahrefs and SEMRush I have been able to identify many high value keywords which I'm deploying pSEO strategies for now.<p>I hope this product is helpful and any feedback (even negative) would bring me joy.<p>All the best,
Bojan
Show HN: A self-published art book about Google's first 25 years
This took me 3 years to finish. (It is 100% self-published, not endorsed by Google.)<p>So… I wrote a book. It’s a different book with a unique approach. It’s not a novel or a technical book. It’s a biography, a company’s biography. My hope is that it serves two purposes: to inspire founders and to captivate interior designers.<p>It all started three years ago. My wife and I were discussing interior design for our living room and she brought up what beautiful coffee table books should we have. After looking at her favorites I realized there’s nothing like this for tech startups. Digital products are intangible, how cool would it be to make them tangible and a beautiful art/decor book for your home office or living room?<p>This idea of consolidating my particular interests with my wife’s decorative goals got stuck in the back of my head for a while. I searched and searched. I could only find these kinds of books related to design, architecture, fashion, or travel. Nothing about tech startups. The next logical step for me was to do it myself…<p>And oh… what a “mistake” it was. This whole project was done during nights and weekends - the first half while working on my startup and the last half while helping create Webflow Labs. You have no idea how many times I “quit” the project. In the end, perseverance and discipline were key to making it happen.<p>During my founder journey, I collected several notes about successful (and unsuccessful) tech startups to help me learn from them. I decided to write about Google’s fascinating story: a true generational company and one of the most valuable companies in the world.<p>It’s with great pleasure and satisfaction that I’m introducing you to “Google - First 25 Years”. A book celebrating the minds behind the tech giant's incredible journey from Stanford to global dominance.<p>I have absolutely no idea if this book’s format and concept make sense or if it has an audience willing to invest ($169). Here are some of the unique features:<p>▪ Silk hardcover<p>▪ Limited edition<p>▪ Handpainted edges<p>▪ More than 30 handmade illustrations<p>I’m producing only 1,000 copies. If the book concept seems interesting enough, leave a comment, I’d love to know your thoughts!<p>Either way, this project was a super fun ride where being outside my comfort zone was the best way to learn storytelling, graphic design, and product manufacturing.<p>At least, I hope this inspires you to go back to your side projects and finish them!
Show HN: A self-published art book about Google's first 25 years
This took me 3 years to finish. (It is 100% self-published, not endorsed by Google.)<p>So… I wrote a book. It’s a different book with a unique approach. It’s not a novel or a technical book. It’s a biography, a company’s biography. My hope is that it serves two purposes: to inspire founders and to captivate interior designers.<p>It all started three years ago. My wife and I were discussing interior design for our living room and she brought up what beautiful coffee table books should we have. After looking at her favorites I realized there’s nothing like this for tech startups. Digital products are intangible, how cool would it be to make them tangible and a beautiful art/decor book for your home office or living room?<p>This idea of consolidating my particular interests with my wife’s decorative goals got stuck in the back of my head for a while. I searched and searched. I could only find these kinds of books related to design, architecture, fashion, or travel. Nothing about tech startups. The next logical step for me was to do it myself…<p>And oh… what a “mistake” it was. This whole project was done during nights and weekends - the first half while working on my startup and the last half while helping create Webflow Labs. You have no idea how many times I “quit” the project. In the end, perseverance and discipline were key to making it happen.<p>During my founder journey, I collected several notes about successful (and unsuccessful) tech startups to help me learn from them. I decided to write about Google’s fascinating story: a true generational company and one of the most valuable companies in the world.<p>It’s with great pleasure and satisfaction that I’m introducing you to “Google - First 25 Years”. A book celebrating the minds behind the tech giant's incredible journey from Stanford to global dominance.<p>I have absolutely no idea if this book’s format and concept make sense or if it has an audience willing to invest ($169). Here are some of the unique features:<p>▪ Silk hardcover<p>▪ Limited edition<p>▪ Handpainted edges<p>▪ More than 30 handmade illustrations<p>I’m producing only 1,000 copies. If the book concept seems interesting enough, leave a comment, I’d love to know your thoughts!<p>Either way, this project was a super fun ride where being outside my comfort zone was the best way to learn storytelling, graphic design, and product manufacturing.<p>At least, I hope this inspires you to go back to your side projects and finish them!
Show HN: A self-published art book about Google's first 25 years
This took me 3 years to finish. (It is 100% self-published, not endorsed by Google.)<p>So… I wrote a book. It’s a different book with a unique approach. It’s not a novel or a technical book. It’s a biography, a company’s biography. My hope is that it serves two purposes: to inspire founders and to captivate interior designers.<p>It all started three years ago. My wife and I were discussing interior design for our living room and she brought up what beautiful coffee table books should we have. After looking at her favorites I realized there’s nothing like this for tech startups. Digital products are intangible, how cool would it be to make them tangible and a beautiful art/decor book for your home office or living room?<p>This idea of consolidating my particular interests with my wife’s decorative goals got stuck in the back of my head for a while. I searched and searched. I could only find these kinds of books related to design, architecture, fashion, or travel. Nothing about tech startups. The next logical step for me was to do it myself…<p>And oh… what a “mistake” it was. This whole project was done during nights and weekends - the first half while working on my startup and the last half while helping create Webflow Labs. You have no idea how many times I “quit” the project. In the end, perseverance and discipline were key to making it happen.<p>During my founder journey, I collected several notes about successful (and unsuccessful) tech startups to help me learn from them. I decided to write about Google’s fascinating story: a true generational company and one of the most valuable companies in the world.<p>It’s with great pleasure and satisfaction that I’m introducing you to “Google - First 25 Years”. A book celebrating the minds behind the tech giant's incredible journey from Stanford to global dominance.<p>I have absolutely no idea if this book’s format and concept make sense or if it has an audience willing to invest ($169). Here are some of the unique features:<p>▪ Silk hardcover<p>▪ Limited edition<p>▪ Handpainted edges<p>▪ More than 30 handmade illustrations<p>I’m producing only 1,000 copies. If the book concept seems interesting enough, leave a comment, I’d love to know your thoughts!<p>Either way, this project was a super fun ride where being outside my comfort zone was the best way to learn storytelling, graphic design, and product manufacturing.<p>At least, I hope this inspires you to go back to your side projects and finish them!
Show HN: BiTE – Cross-platform executable viewer and reverse engineering tool
Hey everyone!<p>I’m excited to share a project I’ve been working on throughout my university studies. It’s called BiTE (<a href="https://github.com/WINSDK/bite">https://github.com/WINSDK/bite</a>) and it's a tool primarily focused on being an executable viewer with reverse engineering capabilities.<p>BiTE supports Windows, MacOS, and Linux, along with their associated executable formats. It’s also capable of parsing and displaying debug information using DWARF/PDB formats, which I hope will be useful even for just comparing codegen.<p>I’ve put a lot of effort into this and it's the first time I'm releasing something like this publicly. Any feedback, bug reports, or feature suggestions would be greatly appreciated!
Show HN: BiTE – Cross-platform executable viewer and reverse engineering tool
Hey everyone!<p>I’m excited to share a project I’ve been working on throughout my university studies. It’s called BiTE (<a href="https://github.com/WINSDK/bite">https://github.com/WINSDK/bite</a>) and it's a tool primarily focused on being an executable viewer with reverse engineering capabilities.<p>BiTE supports Windows, MacOS, and Linux, along with their associated executable formats. It’s also capable of parsing and displaying debug information using DWARF/PDB formats, which I hope will be useful even for just comparing codegen.<p>I’ve put a lot of effort into this and it's the first time I'm releasing something like this publicly. Any feedback, bug reports, or feature suggestions would be greatly appreciated!
Show HN: BiTE – Cross-platform executable viewer and reverse engineering tool
Hey everyone!<p>I’m excited to share a project I’ve been working on throughout my university studies. It’s called BiTE (<a href="https://github.com/WINSDK/bite">https://github.com/WINSDK/bite</a>) and it's a tool primarily focused on being an executable viewer with reverse engineering capabilities.<p>BiTE supports Windows, MacOS, and Linux, along with their associated executable formats. It’s also capable of parsing and displaying debug information using DWARF/PDB formats, which I hope will be useful even for just comparing codegen.<p>I’ve put a lot of effort into this and it's the first time I'm releasing something like this publicly. Any feedback, bug reports, or feature suggestions would be greatly appreciated!
Show HN: Open Source TailwindCSS UI Components
Free Tailwind html UI Components - built to create landing pages and websites. Easyfrontend UI components are free and open-source. Copy paste the components to update your existing site or create a new site from it.
Show HN: Speeding up LLM inference 2x times (possibly)
Here's a project I've been working on for the last few months.<p>It's a new (I think) algorithm, that allows to adjust smoothly - and in real time - how many calculations you'd like to do during inference of an LLM model.<p>It seems that it's possible to do just 20-25% of weight multiplications instead of all of them, and still get good inference results.<p>I implemented it to run on M1/M2/M3 GPU. The mmul approximation itself can be pushed to run 2x fast before the quality of output collapses.<p>The inference speed is just a bit faster than Llama.cpp's, because the rest of implementation could be better, but with a better development I think it can be a new method to speed up inference - in addition to quantization.<p>You could call it ad-hoc model distillation :)<p>You can change the speed / accuracy of a model at will, in real time.<p>Oh, and as a side effect, the data format allows to also choose how much of the model you want to load into the memory. You can decide to skip say 10-20-40% of the least important weights.<p>It's implemented for Mistral, it was also tested slightly on Mixtral and Llama. It's for FP16 for now, but Q8 is in the works.<p>The algorithm is described here, and the implementation is open source.<p><a href="https://kolinko.github.io/effort/" rel="nofollow">https://kolinko.github.io/effort/</a><p>I know these are bold claims, but I hope they survive the scrutiny :)
Show HN: Speeding up LLM inference 2x times (possibly)
Here's a project I've been working on for the last few months.<p>It's a new (I think) algorithm, that allows to adjust smoothly - and in real time - how many calculations you'd like to do during inference of an LLM model.<p>It seems that it's possible to do just 20-25% of weight multiplications instead of all of them, and still get good inference results.<p>I implemented it to run on M1/M2/M3 GPU. The mmul approximation itself can be pushed to run 2x fast before the quality of output collapses.<p>The inference speed is just a bit faster than Llama.cpp's, because the rest of implementation could be better, but with a better development I think it can be a new method to speed up inference - in addition to quantization.<p>You could call it ad-hoc model distillation :)<p>You can change the speed / accuracy of a model at will, in real time.<p>Oh, and as a side effect, the data format allows to also choose how much of the model you want to load into the memory. You can decide to skip say 10-20-40% of the least important weights.<p>It's implemented for Mistral, it was also tested slightly on Mixtral and Llama. It's for FP16 for now, but Q8 is in the works.<p>The algorithm is described here, and the implementation is open source.<p><a href="https://kolinko.github.io/effort/" rel="nofollow">https://kolinko.github.io/effort/</a><p>I know these are bold claims, but I hope they survive the scrutiny :)
Show HN: Speeding up LLM inference 2x times (possibly)
Here's a project I've been working on for the last few months.<p>It's a new (I think) algorithm, that allows to adjust smoothly - and in real time - how many calculations you'd like to do during inference of an LLM model.<p>It seems that it's possible to do just 20-25% of weight multiplications instead of all of them, and still get good inference results.<p>I implemented it to run on M1/M2/M3 GPU. The mmul approximation itself can be pushed to run 2x fast before the quality of output collapses.<p>The inference speed is just a bit faster than Llama.cpp's, because the rest of implementation could be better, but with a better development I think it can be a new method to speed up inference - in addition to quantization.<p>You could call it ad-hoc model distillation :)<p>You can change the speed / accuracy of a model at will, in real time.<p>Oh, and as a side effect, the data format allows to also choose how much of the model you want to load into the memory. You can decide to skip say 10-20-40% of the least important weights.<p>It's implemented for Mistral, it was also tested slightly on Mixtral and Llama. It's for FP16 for now, but Q8 is in the works.<p>The algorithm is described here, and the implementation is open source.<p><a href="https://kolinko.github.io/effort/" rel="nofollow">https://kolinko.github.io/effort/</a><p>I know these are bold claims, but I hope they survive the scrutiny :)