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.

got cloud inference working! here i have smolvla running on a remote 4090 on prime intellect with a thin client running on my local pc sending observations to the remote server and receiving action chunks back. round trip latency is varying between 320-470ms! the 4090 instance I'm using is... show more

Jack Vial

1,337 subscribers

12,992 views • 8 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

One of the coolest uses of LM Studio and AI at AMD is using LM Link to have a Framework Desktop be a remote inference server for a laptop, including something little like Framework Laptop 12. It feels native on the laptop, but gets the desktop’s performance.

One of the coolest uses of LM Studio and AI at AMD is using LM Link to have a Framework Desktop be a remote inference server for a laptop, including something little like Framework Laptop 12. It feels native on the laptop, but gets the desktop’s performance.

Framework

39,175 views • 7 months ago

Claude Code is running on a remote server, along with the web app it just built. You’re on your phone. You can still open localhost:3000 in Safari or Chrome. Just create a local port forwarding rule in Termius and securely tunnel the remote port to your device. No public ports. No extra setup.

Claude Code is running on a remote server, along with the web app it just built. You’re on your phone. You can still open localhost:3000 in Safari or Chrome. Just create a local port forwarding rule in Termius and securely tunnel the remote port to your device. No public ports. No extra setup.

Termius

132,867 views • 3 months ago

Got all 12 student plus 1 instructor Raspberry Pi Pico W with pimoroni GFX pack working and running my #redis streams #micropython workshop. Is this the most GFX Packs on a single project Hel Goblins ⚡🐍 vaporweave ? 😀 event is Thursday Oxford Python

Got all 12 student plus 1 instructor Raspberry Pi Pico W with pimoroni GFX pack working and running my #redis streams #micropython workshop. Is this the most GFX Packs on a single project Hel Goblins ⚡🐍 vaporweave ? 😀 event is Thursday Oxford Python

Simon Prickett 🇪🇺| simonprickett.dev

22,137 views • 3 years ago

GEMMA 4 26B ON AN RTX 4060 WITH A 248K TOKEN CONTEXT WINDOW 20 tokens per second and a context window so large you can feed it entire codebases, books and research papers in a single prompt this is not a cloud api and not a server rack, this is a regular consumer gpu running locally with llama.cpp and q4_k_xl quantization 248k context on an 8gb vram card was not supposed to be possible and here it is just running on someone’s desk the article below covers exactly which tools and configs make this kind of setup work in 2026 ↓

GEMMA 4 26B ON AN RTX 4060 WITH A 248K TOKEN CONTEXT WINDOW 20 tokens per second and a context window so large you can feed it entire codebases, books and research papers in a single prompt this is not a cloud api and not a server rack, this is a regular consumer gpu running locally with llama.cpp and q4_k_xl quantization 248k context on an 8gb vram card was not supposed to be possible and here it is just running on someone’s desk the article below covers exactly which tools and configs make this kind of setup work in 2026 ↓

leopardracer

56,865 views • 4 months ago

This is running in my web browser on my laptop. It's deterministic and running using webgpu Tonight; I have a /goal going where it is going to set up one of my 4090s as the authority server, and then test all the rollback code. Goal is 50k boxes simmed per user, up to 512 users. Massively multiplayer massively rigid body physics online games inside your web browser. Why the hell not, we've got the tokens to spare!

This is running in my web browser on my laptop. It's deterministic and running using webgpu Tonight; I have a /goal going where it is going to set up one of my 4090s as the authority server, and then test all the rollback code. Goal is 50k boxes simmed per user, up to 512 users. Massively multiplayer massively rigid body physics online games inside your web browser. Why the hell not, we've got the tokens to spare!

kache

39,217 views • 2 months ago

Streaming iPhone data in real-time directly to Rerun 🚀 The collection process is one of the most frustrating parts of building imitation-learning datasets. I’ve got a little army of sensors—📱 iPhone, iPad, Quest 3—but getting them temporally aligned, spatially aligned, AND seeing real-time feedback while recording is tough. I stumbled on a great library from Cake Lab (WPI) called ARFlow. It’s a thin client built on Unity’s ARFoundation that connects over gRPC to a server running Rerun for live data logging. I forked it to: - Log the SLAM translation poses, and - Upgrade rerun to v0.23 for my use case. So far, it works well, but there are still a few hitches: 1. Right now, it’s solid on iPhone and iPad; my Quest 3 client is still slow and not super reliable. 2. I’m using an older ARFlow branch focused on real-time streaming only—no spatial or temporal sync yet. Unity builds for iOS keep failing. 🛠️ 3. Nothing is saved locally to the client, so packet loss is a risk on shaky networks. There’s huge potential in tapping the ubiquitous sensors we carry around every day, and ARFlow is a big step toward making that easy

Streaming iPhone data in real-time directly to Rerun 🚀 The collection process is one of the most frustrating parts of building imitation-learning datasets. I’ve got a little army of sensors—📱 iPhone, iPad, Quest 3—but getting them temporally aligned, spatially aligned, AND seeing real-time feedback while recording is tough. I stumbled on a great library from Cake Lab (WPI) called ARFlow. It’s a thin client built on Unity’s ARFoundation that connects over gRPC to a server running Rerun for live data logging. I forked it to: - Log the SLAM translation poses, and - Upgrade rerun to v0.23 for my use case. So far, it works well, but there are still a few hitches: 1. Right now, it’s solid on iPhone and iPad; my Quest 3 client is still slow and not super reliable. 2. I’m using an older ARFlow branch focused on real-time streaming only—no spatial or temporal sync yet. Unity builds for iOS keep failing. 🛠️ 3. Nothing is saved locally to the client, so packet loss is a risk on shaky networks. There’s huge potential in tapping the ubiquitous sensors we carry around every day, and ARFlow is a big step toward making that easy

Pablo Vela

27,433 views • 1 year ago

Switched to the Deere 3D nozzles. Running them on A and B on the exact apply. One is facing forward one is facing back. 20gpa. 10oz veltyma and an insecticide. Doing a great job with coverage.

Switched to the Deere 3D nozzles. Running them on A and B on the exact apply. One is facing forward one is facing back. 20gpa. 10oz veltyma and an insecticide. Doing a great job with coverage.

Jacob Wade

34,220 views • 1 year ago

Working on a character as iconic as Optimus Prime was truly a dream come true. I'll always be grateful for the opportunity, especially as a huge Transformers fan. Here is a comparison between the viewport playblast and the final shot for one of his transformations.

Working on a character as iconic as Optimus Prime was truly a dream come true. I'll always be grateful for the opportunity, especially as a huge Transformers fan. Here is a comparison between the viewport playblast and the final shot for one of his transformations.

Rassoul Edji

43,400 views • 1 year ago

Running cold email campaigns just became a whole lot easier Smartlead now runs an MCP server, which in plain terms means Claude can read and act on your live campaign data directly instead of working off a spreadsheet that went stale the moment you exported it. The workflow is worth walking through properly, because it is shorter than people expect. You generate an API key inside your account, point Claude at the server once, and from then on you ask for what you want in a sentence. Here is a prompt worth stealing in full: "Fetch all Smartlead clients, then get today's performance for each: emails sent, replied, positive replies, unique lead count. Compute reply rate per client, run a top and bottom performer analysis, format it as a daily client performance report, and post it to Slack." One paste, and it pulls live figures for every account, does the arithmetic, ranks the strongest and the weakest, and delivers the finished thing into the channel your team already sits in, before anyone has logged on for the day. Be clear about the division of labour, because it is what makes this useful rather than a novelty. Smartlead is the engine holding the campaigns, the mailboxes, the warmup and the reply data, and Claude is simply the interface you operate all of it through, so nothing about your sending changes and everything about how you interrogate it does. The effect people underestimate is on the questions you start asking. Once a report costs you a sentence rather than an afternoon, you stop rationing the ones that used to feel like too much trouble, and problems that used to surface on a Friday start surfacing on a Tuesday. Connect it with Claude through MCP and run one prompt against your own account today.

Running cold email campaigns just became a whole lot easier Smartlead now runs an MCP server, which in plain terms means Claude can read and act on your live campaign data directly instead of working off a spreadsheet that went stale the moment you exported it. The workflow is worth walking through properly, because it is shorter than people expect. You generate an API key inside your account, point Claude at the server once, and from then on you ask for what you want in a sentence. Here is a prompt worth stealing in full: "Fetch all Smartlead clients, then get today's performance for each: emails sent, replied, positive replies, unique lead count. Compute reply rate per client, run a top and bottom performer analysis, format it as a daily client performance report, and post it to Slack." One paste, and it pulls live figures for every account, does the arithmetic, ranks the strongest and the weakest, and delivers the finished thing into the channel your team already sits in, before anyone has logged on for the day. Be clear about the division of labour, because it is what makes this useful rather than a novelty. Smartlead is the engine holding the campaigns, the mailboxes, the warmup and the reply data, and Claude is simply the interface you operate all of it through, so nothing about your sending changes and everything about how you interrogate it does. The effect people underestimate is on the questions you start asking. Once a report costs you a sentence rather than an afternoon, you stop rationing the ones that used to feel like too much trouble, and problems that used to surface on a Friday start surfacing on a Tuesday. Connect it with Claude through MCP and run one prompt against your own account today.

Tim

21,666 views • 21 days ago

The gold standard fluid interactions on iOS are built using in-process client animations (like SwiftUI does it) and not render-server animations (like CoreAnimation does it). This is common knowledge inside of Apple’s UI framework and system experience teams. A close inspection of iOS’s evolution reveals the gradual upgrading of system experiences written using render-server animations (that often disable user interaction) with ones using client animations (that are interruptible and interactive). The app switcher for iPhone X, the Dynamic Island, and Liquid Glass bars are prime examples of this trend. The other thing that makes these experiences feel so fluid is their use of spring animations that preserve velocity, rather than Bézier curve easing functions. To get velocity preservation, you need to be able to interrupt a spring animation and retarget it or drive it with a gesture. That’s a lot easier to do when the animation is running in the same process as the event handling. When we were first pitching SwiftUI, two of our go-to demos were a recreation of the iOS 9 navigation and Home Screen interactions which were fully interactive and interruptible, at a time when the actual implementations of those experiences in the OS disabled user interaction and ran server-side animations. CoreAnimation is great and the framework does much more than animation. You can build a fine app using its server-side animations. But if your UI framework can’t do efficient main thread client animations, there’s a ceiling to the quality of the experience your app can provide.

The gold standard fluid interactions on iOS are built using in-process client animations (like SwiftUI does it) and not render-server animations (like CoreAnimation does it). This is common knowledge inside of Apple’s UI framework and system experience teams. A close inspection of iOS’s evolution reveals the gradual upgrading of system experiences written using render-server animations (that often disable user interaction) with ones using client animations (that are interruptible and interactive). The app switcher for iPhone X, the Dynamic Island, and Liquid Glass bars are prime examples of this trend. The other thing that makes these experiences feel so fluid is their use of spring animations that preserve velocity, rather than Bézier curve easing functions. To get velocity preservation, you need to be able to interrupt a spring animation and retarget it or drive it with a gesture. That’s a lot easier to do when the animation is running in the same process as the event handling. When we were first pitching SwiftUI, two of our go-to demos were a recreation of the iOS 9 navigation and Home Screen interactions which were fully interactive and interruptible, at a time when the actual implementations of those experiences in the OS disabled user interaction and ran server-side animations. CoreAnimation is great and the framework does much more than animation. You can build a fine app using its server-side animations. But if your UI framework can’t do efficient main thread client animations, there’s a ceiling to the quality of the experience your app can provide.

Kyle Macomber

194,329 views • 23 days ago

Woow Tencent has released a model 100% open source which is better than Google Translate... And it runs locally on your phone?! - Only 440MB (!!) - 33 languages - On-device translation So we have now a local Google Translate which is running completely offline. They are using AngelSlim which their compression toolkit. That's why the model is so efficient.

Woow Tencent has released a model 100% open source which is better than Google Translate... And it runs locally on your phone?! - Only 440MB (!!) - 33 languages - On-device translation So we have now a local Google Translate which is running completely offline. They are using AngelSlim which their compression toolkit. That's why the model is so efficient.

Paul Couvert

142,678 views • 5 months ago

The B-Wing Cockpit animation and model are done! I've been at this one for a while, and I'm relieved to have it finished, so I can move onto something else. Enjoy! Rendered in 3dsmax 2024 using VRay 7 GPU on a 4090 RTX. It took about 3-3.5m per frame.

The B-Wing Cockpit animation and model are done! I've been at this one for a while, and I'm relieved to have it finished, so I can move onto something else. Enjoy! Rendered in 3dsmax 2024 using VRay 7 GPU on a 4090 RTX. It took about 3-3.5m per frame.

Howie Day

57,593 views • 1 year ago

Dom Hamel is asked about his role with the Mets and his future position on the team: "I'm trying not to leave them a choice, I don't want to go back [to AAA] - I'm thinking I'm up here for a reason"

Dom Hamel is asked about his role with the Mets and his future position on the team: "I'm trying not to leave them a choice, I don't want to go back [to AAA] - I'm thinking I'm up here for a reason"

SNY Mets

81,585 views • 1 year ago

MolmoAct2 is landing in LeRobot! Ai2's open Action Reasoning Model combines a Molmo2-ER vision-language backbone with a flow-matching continuous action expert to predict robot action chunks from images, language instructions, and proprioceptive state. An open robot foundation model built for real-world control, with strong out-of-the-box performance and easy fine-tuning in LeRobot. Pick-and-place inference running on NVIDIA DGX Spark! Blog: Paper: Thanks to Ai2 Jiafei Duan Haoquan Fang

MolmoAct2 is landing in LeRobot! Ai2's open Action Reasoning Model combines a Molmo2-ER vision-language backbone with a flow-matching continuous action expert to predict robot action chunks from images, language instructions, and proprioceptive state. An open robot foundation model built for real-world control, with strong out-of-the-box performance and easy fine-tuning in LeRobot. Pick-and-place inference running on NVIDIA DGX Spark! Blog: Paper: Thanks to Ai2 Jiafei Duan Haoquan Fang

LeRobot

25,163 views • 4 months ago

This is only 1 Niagara emitter to handle thousands of decals as particles using Data Channels! Significantly lower overhead on the GameThread and only a single component per grid cell. This also adds instancing to the decals which is normally not supported for individual decal components. Working on this as part of data-oriented projectiles that run without Actors, asynchronous collision and supporting multiplayer (incl. client side prediction). Removes a huge amount of overhead from the CPU, more on that soon!

This is only 1 Niagara emitter to handle thousands of decals as particles using Data Channels! Significantly lower overhead on the GameThread and only a single component per grid cell. This also adds instancing to the decals which is normally not supported for individual decal components. Working on this as part of data-oriented projectiles that run without Actors, asynchronous collision and supporting multiplayer (incl. client side prediction). Removes a huge amount of overhead from the CPU, more on that soon!

Tom Looman

31,566 views • 1 year ago

This guy bought a $400 Mac mini and walked into a coffee shop with $2,100 for installing an AI agent that never touches the internet. I had to rewatch this because the pitch is almost too simple. He shows up, plugs in the Mac mini, installs a local AI agent running entirely through Ollama, done in under an hour. Zero cloud dependency, no API bills, no monthly charges bleeding the client dry. The coffee shop owner gets a private AI system that keeps working even if the WiFi dies, and he walks out with $2,100 for an install that takes less time than a long lunch. The hardware cost him $400. One deployment and the machine's already paid for five times over. Then he locks in a monthly retainer for support, somewhere around $100-150, which means every client after the first is close to pure margin stacking on top of recurring revenue. Most people hear "AI business" and think they need to build a SaaS platform or learn to code or raise funding from someone. This skips all of that and goes straight to walking into coffee shops with a Mac mini under one arm. Not a company that scales to a billion dollars, obviously, but a solo operator clearing a few thousand a month from local installs while everyone else is still arguing about which LLM is best on Twitter.

This guy bought a $400 Mac mini and walked into a coffee shop with $2,100 for installing an AI agent that never touches the internet. I had to rewatch this because the pitch is almost too simple. He shows up, plugs in the Mac mini, installs a local AI agent running entirely through Ollama, done in under an hour. Zero cloud dependency, no API bills, no monthly charges bleeding the client dry. The coffee shop owner gets a private AI system that keeps working even if the WiFi dies, and he walks out with $2,100 for an install that takes less time than a long lunch. The hardware cost him $400. One deployment and the machine's already paid for five times over. Then he locks in a monthly retainer for support, somewhere around $100-150, which means every client after the first is close to pure margin stacking on top of recurring revenue. Most people hear "AI business" and think they need to build a SaaS platform or learn to code or raise funding from someone. This skips all of that and goes straight to walking into coffee shops with a Mac mini under one arm. Not a company that scales to a billion dollars, obviously, but a solo operator clearing a few thousand a month from local installs while everyone else is still arguing about which LLM is best on Twitter.

BrainRul

4,706,589 views • 2 months ago

OK, I have a definitive word on the CJ Abrams play from today's Pittsburgh Pirates at Washington Nationals game after talking with Elias Sports Bureau on this. This play will stay as a sacrifice fly. The originial ruling of NOT a sacrifice fly was for the exact same reason that I thought, which is that the infielder is not running into the outfield, which is year's past would have been correct As it was explained to me, in past years, an infielder had to be running almost in a straight line towards the outfield wall to be considered "running in the outfield". Here, since he is running, and he ends up further away from home (157 feet) than when he started (145 feet), this is going to count as a sacrifice fly. That definition is changing, in part from this play to help bring greater consistency, and to take some of the guesswork out of it (the argument that he is running into the outfield as opposed to more parallel). Now, folks all the time ask "why doesn't MLB publish the OS Manual" and I always say because it is a living document that can have the wording change, and the wording for this play will be modified to something like "more towards the outfield wall than towards home plate" to eliminate any confusion. The big key to this play is that he was running on a full sprint. Also, and this is helpful for me, but for all fly ball outs that score a run, Elias Saba reviews to ensure consistency. So, yes, it's a sacrifice fly, and now that I have that info from Elias themselves, that sort of settles this one. Sounds like the guidelines for this definition will be changing, either this season, or certainly for next season.

OK, I have a definitive word on the CJ Abrams play from today's Pittsburgh Pirates at Washington Nationals game after talking with Elias Sports Bureau on this. This play will stay as a sacrifice fly. The originial ruling of NOT a sacrifice fly was for the exact same reason that I thought, which is that the infielder is not running into the outfield, which is year's past would have been correct As it was explained to me, in past years, an infielder had to be running almost in a straight line towards the outfield wall to be considered "running in the outfield". Here, since he is running, and he ends up further away from home (157 feet) than when he started (145 feet), this is going to count as a sacrifice fly. That definition is changing, in part from this play to help bring greater consistency, and to take some of the guesswork out of it (the argument that he is running into the outfield as opposed to more parallel). Now, folks all the time ask "why doesn't MLB publish the OS Manual" and I always say because it is a living document that can have the wording change, and the wording for this play will be modified to something like "more towards the outfield wall than towards home plate" to eliminate any confusion. The big key to this play is that he was running on a full sprint. Also, and this is helpful for me, but for all fly ball outs that score a run, Elias Saba reviews to ensure consistency. So, yes, it's a sacrifice fly, and now that I have that info from Elias themselves, that sort of settles this one. Sounds like the guidelines for this definition will be changing, either this season, or certainly for next season.

MLB Scoring Changes

49,141 views • 3 months ago

This guy built a mini AI farm out of 4 Nvidia boxes It does not look like a data center. It looks like a stack of small machines sitting next to a laptop. But each box is a DGX Spark with Grace Blackwell inside, 128GB unified memory, and enough room to run models normal gaming GPUs cannot even open. Using the launch price from the article, 4 of them is almost $12,000 of local AI compute on one desk. That sounds expensive until you compare it to cloud GPUs. A serious AI builder can burn $1,500 to $3,000 a month renting A100s and H100s for client work, fine-tunes, agents and 70B models. He basically moved that bill from the cloud into hardware he owns. 4 Nvidia boxes. 512GB unified memory. No hourly meter running in the background. No rented GPUs eating the margin every time an agent runs too long. The funny part is most people still think local AI means a slow laptop running a toy model. Meanwhile guys like this are stacking compute at home. Save this, local AI is turning into the new mining farm.

This guy built a mini AI farm out of 4 Nvidia boxes It does not look like a data center. It looks like a stack of small machines sitting next to a laptop. But each box is a DGX Spark with Grace Blackwell inside, 128GB unified memory, and enough room to run models normal gaming GPUs cannot even open. Using the launch price from the article, 4 of them is almost $12,000 of local AI compute on one desk. That sounds expensive until you compare it to cloud GPUs. A serious AI builder can burn $1,500 to $3,000 a month renting A100s and H100s for client work, fine-tunes, agents and 70B models. He basically moved that bill from the cloud into hardware he owns. 4 Nvidia boxes. 512GB unified memory. No hourly meter running in the background. No rented GPUs eating the margin every time an agent runs too long. The funny part is most people still think local AI means a slow laptop running a toy model. Meanwhile guys like this are stacking compute at home. Save this, local AI is turning into the new mining farm.

Gipp 🦅

591,712 views • 4 months ago