Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

I updated my quantization compiler to use e-means as a replacement for k-means in codebook generation and got Qwen3.5-4B from 8.41 GB to 1.23 GB while retaining 91.9% of original performance which I think is approximately SOTA, at least comparable to bonsai 2 but I think we could go... show more

𝞍 Shin Megami Boson 𝞍

13,089 subscribers

27,667 views • 2 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Hardware is a big constraint in robotics, that's why we don't think the answer will be increasingly larger VLM models as action backbones. We tried to solve that issue with WorldDiT, which is at the pareto frontier for size-performance. Now I want to see how far we can scale it without losing this tradeoff. Here's a simulation demo.

Hardware is a big constraint in robotics, that's why we don't think the answer will be increasingly larger VLM models as action backbones. We tried to solve that issue with WorldDiT, which is at the pareto frontier for size-performance. Now I want to see how far we can scale it without losing this tradeoff. Here's a simulation demo.

Bidhan Roy

11,405 views • 3 months ago

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!

Daily Dose of Data Science

23,748 views • 6 months ago

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!

Akshay 🚀

36,317 views • 7 months ago

Spring is coming which means summer is knock at our door. That means it's time to get that beach body ready so you can get your surf on. 🏄 That said the board that the girl in the video has like too small to be a standard size board. I'm no expert on this, so if anyone knows please share in the comments. 👇

Spring is coming which means summer is knock at our door. That means it's time to get that beach body ready so you can get your surf on. 🏄 That said the board that the girl in the video has like too small to be a standard size board. I'm no expert on this, so if anyone knows please share in the comments. 👇

Giggling Gannon

19,450 views • 7 months ago

Ok so I got access to Seedream 2.0 and... it's a huge step up. But to balance expecations, the quality isn't quite at the level I'm seeing in some of these promo vids. Don't get me wrong, this is MUCH better than existing models. But it's not Hollywood-level (yet). 2 months?

Ok so I got access to Seedream 2.0 and... it's a huge step up. But to balance expecations, the quality isn't quite at the level I'm seeing in some of these promo vids. Don't get me wrong, this is MUCH better than existing models. But it's not Hollywood-level (yet). 2 months?

Matt Shumer

33,755 views • 9 months ago

i'm writing my own gui framework in c++ and rewriting my image editor using it, which is going good so far, i wanted to move from swiftui, qt, tauri, electron, react native, and flutter so i can use metal and directx gpu apis on macos and windows as much as possible,so i think going full metal/directx is the way to go i know i have a very hard and long way to go though as there are many many flaws,bugs in this very very early version and doesnt utilize the gpu as much as i want to but i hope we will get there soon ,i also need to make the ui look somewhat decent instead of whatever mess it is right now

i'm writing my own gui framework in c++ and rewriting my image editor using it, which is going good so far, i wanted to move from swiftui, qt, tauri, electron, react native, and flutter so i can use metal and directx gpu apis on macos and windows as much as possible,so i think going full metal/directx is the way to go i know i have a very hard and long way to go though as there are many many flaws,bugs in this very very early version and doesnt utilize the gpu as much as i want to but i hope we will get there soon ,i also need to make the ui look somewhat decent instead of whatever mess it is right now

Ruben Veidt

85,234 views • 1 year ago

It's spooky season, so it means it's time to post spooky characters 😋 I think Makima is one of them! I just recently updated my kofi webshop with some new photoshoots. So if anyone missed the Makima set on my Patreon, now you can get it in the webshop!☺️ Check my link in bio!

It's spooky season, so it means it's time to post spooky characters 😋 I think Makima is one of them! I just recently updated my kofi webshop with some new photoshoots. So if anyone missed the Makima set on my Patreon, now you can get it in the webshop!☺️ Check my link in bio!

Enji Night

115,025 views • 2 years ago

Run Ternary Bonsai 2 locally in Atomic Chat 🪴 PrismML shrank Qwen3.5 7x so the 9B goes from 17.9 GB to just 2.42 GB and it keeps up or outperforms the original model on benchmarks Run AI models locally ->

Run Ternary Bonsai 2 locally in Atomic Chat 🪴 PrismML shrank Qwen3.5 7x so the 9B goes from 17.9 GB to just 2.42 GB and it keeps up or outperforms the original model on benchmarks Run AI models locally ->

atomic.chat

78,238 views • 29 days ago

JADE gets emotional reflecting on her North American tour in a new TikTok: “I’m having so much fun on this tour. I just feel so grateful that I’m getting to do this at this point in my career. The fact that I get to tour [North America] after being in the industry for 15 years, and only now just getting to do my own headline tour, is incredible. It’s been a long time coming. What’s really beautiful about these shows is that when I look into the audience, I recognise so many of the fans here from back in the day, who’ve literally waited for years – like me! – for this moment. It just makes me feel so loved and supported to know people have believed in me enough to stick around for years waiting for this to happen. Anyone that’s bought a ticket, dressed up, made their own costumes… It’s just such a lovely, beautiful thing. I hope they can see when I’m on stage just how much that means to me, and how much I love performing and putting on the best show that I possibly can every single night. I will never ever take it for granted. I’m just so chuffed that I get to do this for a living… be a silly pop girlie, write and create music that brings people – and myself – a lot of joy… Thank you for believing in me. I literally get on the bunk on my bus most nights from the tour and just lie there like, ‘Oh my god! As if this is my life!’ It means a lot. I hope I get to do this for the rest of my life… Full of gratitude and lots of all the lovely emotions. Thank you so much.”

JADE gets emotional reflecting on her North American tour in a new TikTok: “I’m having so much fun on this tour. I just feel so grateful that I’m getting to do this at this point in my career. The fact that I get to tour [North America] after being in the industry for 15 years, and only now just getting to do my own headline tour, is incredible. It’s been a long time coming. What’s really beautiful about these shows is that when I look into the audience, I recognise so many of the fans here from back in the day, who’ve literally waited for years – like me! – for this moment. It just makes me feel so loved and supported to know people have believed in me enough to stick around for years waiting for this to happen. Anyone that’s bought a ticket, dressed up, made their own costumes… It’s just such a lovely, beautiful thing. I hope they can see when I’m on stage just how much that means to me, and how much I love performing and putting on the best show that I possibly can every single night. I will never ever take it for granted. I’m just so chuffed that I get to do this for a living… be a silly pop girlie, write and create music that brings people – and myself – a lot of joy… Thank you for believing in me. I literally get on the bunk on my bus most nights from the tour and just lie there like, ‘Oh my god! As if this is my life!’ It means a lot. I hope I get to do this for the rest of my life… Full of gratitude and lots of all the lovely emotions. Thank you so much.”

JADE tea room ☕️

47,114 views • 8 months ago

Abdul El-Sayed: I'm a professional. I work with anybody if it means that we can get something done for the state of Michigan. That's the responsibility. I'm not there to go play politics. I think people are sick and tired of the retribution politics: they did this, so we do this. It is not about politicians. It is about the public we serve.

Abdul El-Sayed: I'm a professional. I work with anybody if it means that we can get something done for the state of Michigan. That's the responsibility. I'm not there to go play politics. I think people are sick and tired of the retribution politics: they did this, so we do this. It is not about politicians. It is about the public we serve.

Clash Report

13,868 views • 8 days ago

I feel like there is some kind of stagnation in my soul. I don't feel hopeless or tired, but suddenly everything became less interesting. I have to do something to wake up my soul so I can go back to how I was. I am the one who insisted on living, even if life was only room for my feet. I wonder what happened to me. I have become just a mass of flesh without a soul. I lost my passion and love for life.. My dear friends and loved ones .. This is your friend Mo from Gaza and this is an update The military operation is still continuing in northern Gaza... and there are more evacuation orders for the north and more displacement. I heard from one of the merchants that he has some cat food in Jabalia... but the area is dangerous there... Approximately he has 15 to 20 boxes Each box contains 24 cans Despite that, I said let's try to go, but we have to be careful... I went with my cousin Animal world We went to the furthest place we could reach. The picture below shows us at the Qurm Junction, and anyone who is in Gaza knows that it is the entrance to Jabalia... Unfortunately, we did not dare to approach any closer than that... We heard the sounds of explosions and saw plumes of smoke and the sounds of gunfire. Drones were flying densely over our heads. So we backed down and decided to go again today.. but the weather is very, very rainy today.. We will go again on Wednesday.. I wish we had access to cat food. It is more beneficial for them.. My friends and loved ones.. There is still the possibility of donating to buy the largest possible amount of food for them Or on the PayPal link It appears that the talks have reached a dead end. In this case, I should buy as much as possible. Today we can buy.. Who knows? Perhaps we will not be able to go and search in a few days... as I am doing now... I am grateful to someone who stood by my side during these difficult times. Your loving friend Mo

I feel like there is some kind of stagnation in my soul. I don't feel hopeless or tired, but suddenly everything became less interesting. I have to do something to wake up my soul so I can go back to how I was. I am the one who insisted on living, even if life was only room for my feet. I wonder what happened to me. I have become just a mass of flesh without a soul. I lost my passion and love for life.. My dear friends and loved ones .. This is your friend Mo from Gaza and this is an update The military operation is still continuing in northern Gaza... and there are more evacuation orders for the north and more displacement. I heard from one of the merchants that he has some cat food in Jabalia... but the area is dangerous there... Approximately he has 15 to 20 boxes Each box contains 24 cans Despite that, I said let's try to go, but we have to be careful... I went with my cousin Animal world We went to the furthest place we could reach. The picture below shows us at the Qurm Junction, and anyone who is in Gaza knows that it is the entrance to Jabalia... Unfortunately, we did not dare to approach any closer than that... We heard the sounds of explosions and saw plumes of smoke and the sounds of gunfire. Drones were flying densely over our heads. So we backed down and decided to go again today.. but the weather is very, very rainy today.. We will go again on Wednesday.. I wish we had access to cat food. It is more beneficial for them.. My friends and loved ones.. There is still the possibility of donating to buy the largest possible amount of food for them Or on the PayPal link It appears that the talks have reached a dead end. In this case, I should buy as much as possible. Today we can buy.. Who knows? Perhaps we will not be able to go and search in a few days... as I am doing now... I am grateful to someone who stood by my side during these difficult times. Your loving friend Mo

help cats

15,121 views • 1 year ago

This shouldn't be news to anyone. If I win, we will have a divided government. But that means we have to work together to get things done, and I think that's a good thing.

This shouldn't be news to anyone. If I win, we will have a divided government. But that means we have to work together to get things done, and I think that's a good thing.

Rob Sand

17,823 views • 5 months ago

We knew this was going to happen at some point, somewhere. We now have politicians officially trying to dictate what hardware autonomous companies like Tesla need to use (lol) Sen. Andrew Zwicker: "At this point, I don't think the evidence is sufficient that a single sensor with software can handle situations that humans can." Apparently Tesla's nearly 12B real world miles on FSD and Unsupervised Robotaxi operations in four cities with cameras only isn't enough evidence? This will surely not be the last time this happens Be sure to oppose this ridiculous bill with Tesla at the link below $TSLA

We knew this was going to happen at some point, somewhere. We now have politicians officially trying to dictate what hardware autonomous companies like Tesla need to use (lol) Sen. Andrew Zwicker: "At this point, I don't think the evidence is sufficient that a single sensor with software can handle situations that humans can." Apparently Tesla's nearly 12B real world miles on FSD and Unsupervised Robotaxi operations in four cities with cameras only isn't enough evidence? This will surely not be the last time this happens Be sure to oppose this ridiculous bill with Tesla at the link below $TSLA

Dillon Loomis

300,219 views • 4 months ago

I'm loving so much of what I saw in Fable but there were two noticable things that I couldn't overlook that I was hoping to bring to the attention to the ultra talented team at Playground for clarification or correction. Ralph Fulton Playground Games 1. The framerate looks to be 30FPS and I'm hoping it'll get better optimization at least for the PC release. 2. Our male character has a fairly feminine walk which you can see below and while I'm happy to know we can choose whichever sex, I do have an expectation of a more masculine hero character's strut and this for me misses the mark.

I'm loving so much of what I saw in Fable but there were two noticable things that I couldn't overlook that I was hoping to bring to the attention to the ultra talented team at Playground for clarification or correction. Ralph Fulton Playground Games 1. The framerate looks to be 30FPS and I'm hoping it'll get better optimization at least for the PC release. 2. Our male character has a fairly feminine walk which you can see below and while I'm happy to know we can choose whichever sex, I do have an expectation of a more masculine hero character's strut and this for me misses the mark.

𝕀𝕟𝕗𝕚𝕟𝕚𝕥𝕖 𝕌𝕞𝕓𝕣𝕒

121,370 views • 9 months ago

I finally updated my design portfolio on Framer 🙂‍↕️🎉 I’ve been sitting on this for a while because ideation is forever but I think it is in a good enough place to share.

I finally updated my design portfolio on Framer 🙂‍↕️🎉 I’ve been sitting on this for a while because ideation is forever but I think it is in a good enough place to share.

Adanna

33,592 views • 1 year ago

I'm struggling to describe how I feel with words right now I got woken up by a call from a friend at 02:56 in the morning, I decline, and for a second I see in my notifications "bro please for the love of god sell", then I get another call but I pick up - "bro your up 50k on my mother your up 50k" I was shaking for at least 30 minutes, I couldn't believe my eyes, for the first time in my life I felt like I was dreaming and I had to ask my friends if this was reality While I was journaling this moment I also found that in my journal I wrote "I will run my port up to 6 figs" three times in a row, which I think I wrote when I was about to zero out, I was manifesting this moment when it seemed impossible I've never felt like this in my entire life I can now say, that i've hit 6 figs. It's not the first time, but this time it feels way different I want to say a lot more things but I guess ill end this off by saying that anything is possible, I had 1.7k left 44 days ago, and with the correct mindset and discipline I was able to turn things around in a way I still can't believe I will be posting recaps on my YouTube as soon as i'm back home which is in around 8-9 days This is genuinely surreal, anything is possible +$37.4K

I'm struggling to describe how I feel with words right now I got woken up by a call from a friend at 02:56 in the morning, I decline, and for a second I see in my notifications "bro please for the love of god sell", then I get another call but I pick up - "bro your up 50k on my mother your up 50k" I was shaking for at least 30 minutes, I couldn't believe my eyes, for the first time in my life I felt like I was dreaming and I had to ask my friends if this was reality While I was journaling this moment I also found that in my journal I wrote "I will run my port up to 6 figs" three times in a row, which I think I wrote when I was about to zero out, I was manifesting this moment when it seemed impossible I've never felt like this in my entire life I can now say, that i've hit 6 figs. It's not the first time, but this time it feels way different I want to say a lot more things but I guess ill end this off by saying that anything is possible, I had 1.7k left 44 days ago, and with the correct mindset and discipline I was able to turn things around in a way I still can't believe I will be posting recaps on my YouTube as soon as i'm back home which is in around 8-9 days This is genuinely surreal, anything is possible +$37.4K

ke3d

47,434 views • 1 month ago

tho i’m not sure if i’ll be able to do it this year but i really wanted to try a new hobby. and one thing i’ve been thinking about for a while is learning how to play drums. i think it would be a great start having a hobby like that for this year

tho i’m not sure if i’ll be able to do it this year but i really wanted to try a new hobby. and one thing i’ve been thinking about for a while is learning how to play drums. i think it would be a great start having a hobby like that for this year

마끄리 🩵

101,679 views • 1 year ago

While learning Unreal Engine 5, I decided to port over most of Hyperion to it to see if I could potentially move my workflows over to UE5. Although it worked decently well, it didn't work well enough to allow me to change over entirely. Hyperion is simply too complex of an asset for real-time workflows (or at least is limited due to my knowledge of said workflows). 131 4K-8K UDIMs at over 8 channels and a solid amount of shaders with layering being the biggest bottleneck. I couldn't even match the shading setup to what I have in RenderMan, UE gave me a warning that the material was too complex. I was using the new Substrate materials, which mimic RenderMan LAMA shaders. The render below is using the pathtracer. As you can see there are some random flickering parts which I couldn't fix during my tests. This render of course doesn't have any displacement either (which Hyperion heavily relies on). I think Unreal Engine has a very bright future, but it is still limited by todays hardware technology for high end VFX work. Maybe in a few years I will give it another try when hardware improves and the engine becomes a bit more versatile in feature-set.

While learning Unreal Engine 5, I decided to port over most of Hyperion to it to see if I could potentially move my workflows over to UE5. Although it worked decently well, it didn't work well enough to allow me to change over entirely. Hyperion is simply too complex of an asset for real-time workflows (or at least is limited due to my knowledge of said workflows). 131 4K-8K UDIMs at over 8 channels and a solid amount of shaders with layering being the biggest bottleneck. I couldn't even match the shading setup to what I have in RenderMan, UE gave me a warning that the material was too complex. I was using the new Substrate materials, which mimic RenderMan LAMA shaders. The render below is using the pathtracer. As you can see there are some random flickering parts which I couldn't fix during my tests. This render of course doesn't have any displacement either (which Hyperion heavily relies on). I think Unreal Engine has a very bright future, but it is still limited by todays hardware technology for high end VFX work. Maybe in a few years I will give it another try when hardware improves and the engine becomes a bit more versatile in feature-set.

Rassoul Edji

25,133 views • 1 year ago

: Do u think there isn't enough done to promote (ILF) 🦮:Honestly I felt it was not enough since the last series, so for this series we requested that we wanted to go on more shows 🦋: yea 🦮🦋: We had already put in our request to do that #GrabThaiRaresxทททxLingOrm

: Do u think there isn't enough done to promote (ILF) 🦮:Honestly I felt it was not enough since the last series, so for this series we requested that we wanted to go on more shows 🦋: yea 🦮🦋: We had already put in our request to do that #GrabThaiRaresxทททxLingOrm

Tipsy 🇳🇿🇹🇭

169,240 views • 4 months ago