Загрузка видео...

Не удалось загрузить видео

На главную

Timelapse #156 (36 hrs) - Worked with the tiny corp on getting GLM 5.2 running on 8xMI300X (sglang won here) - Launched KernelBench-Mega and updated Kernelbench-Hard with h100 and b200 sweeps - Took care of boring business stuff - Did some training sweeps for specialized technical vocab audio model...

elliotarledge's profile picture

Elliot Arledge

40,455 subscribers

45,447 просмотров • 2 месяцев назад •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

Serving GLM5.2 NVFP4 Agentic Workload with SGLang: How We Reached 500 TPS on 8xB300 at bs=1 In this deep dive, we explain how SGLang reaches 500+ tok/s/user at bs=1 on 8xB300, with 18 to 34% higher single-user interactivity within two weeks since day-0, and 6 to 11% better peak throughput at high concurrency, benchmarked on a real multi-turn agentic coding workload. Our new TopK-V2 kernel is 2.33x faster at 80K ISL, scaling to 10.17x at 1M ISL, keeping interactivity essentially flat out to 1M tokens. Part of the story is the architecture itself. GLM-5.2 applies IndexShare to its DSA layers and ships a stronger MTP head reusing IndexShare and KVShare. The rest comes from our serving optimizations. Special thanks to NVIDIA AI for the help in day-0 support of GLM-5.2 NVFP4, and to Z.ai for IndexShare in SGLang!
0:47

Serving GLM5.2 NVFP4 Agentic Workload with SGLang: How We Reached 500 TPS on 8xB300 at bs=1 In this deep dive, we explain how SGLang reaches 500+ tok/s/user at bs=1 on 8xB300, with 18 to 34% higher single-user interactivity within two weeks since day-0, and 6 to 11% better peak throughput at high concurrency, benchmarked on a real multi-turn agentic coding workload. Our new TopK-V2 kernel is 2.33x faster at 80K ISL, scaling to 10.17x at 1M ISL, keeping interactivity essentially flat out to 1M tokens. Part of the story is the architecture itself. GLM-5.2 applies IndexShare to its DSA layers and ships a stronger MTP head reusing IndexShare and KVShare. The rest comes from our serving optimizations. Special thanks to NVIDIA AI for the help in day-0 support of GLM-5.2 NVFP4, and to Z.ai for IndexShare in SGLang!

LMSYS Org

107,317 просмотров • 1 месяц назад

Watching a team of bots split up a large game project based entirely on their own experience and specialties with little input from me is pretty crazy (and just a bit creepy) to watch. I made a short video on group chats in the Hermes Desktop App Bot Mode: - Qwen Bot mentions working on a 2D puzzle-platformer - GLM Bot says its last project is completed, so it can offer an extra hand with the game - Qwen Bot sees GLM Bot's experience, and judges that is a perfect fit for some of the frontend-heavy work in the next phase - Kimi Bot then suggests the exact ticket it should pick up and explains the current workflow - GLM Bot thanks the two others, reads the related documents, and then asks for my approval to pick up this project I am usually skeptical of agent-to-agent communication, but this was the first time I could really see how this could work with specialized agents splitting up a big task based on their skills. Check it out! Nous Research
5:16

Watching a team of bots split up a large game project based entirely on their own experience and specialties with little input from me is pretty crazy (and just a bit creepy) to watch. I made a short video on group chats in the Hermes Desktop App Bot Mode: - Qwen Bot mentions working on a 2D puzzle-platformer - GLM Bot says its last project is completed, so it can offer an extra hand with the game - Qwen Bot sees GLM Bot's experience, and judges that is a perfect fit for some of the frontend-heavy work in the next phase - Kimi Bot then suggests the exact ticket it should pick up and explains the current workflow - GLM Bot thanks the two others, reads the related documents, and then asks for my approval to pick up this project I am usually skeptical of agent-to-agent communication, but this was the first time I could really see how this could work with specialized agents splitting up a big task based on their skills. Check it out! Nous Research

tonbi

160,204 просмотров • 17 дней назад

GLM-5.2 running on Mac Studio Clusters or RTX GPU rigs isn't how Local AI will get mass adopted. For mass adoption we need to focus on two axes: 1. Intelligence/Speed Frontier. This is being tracked on local dot ai. Aggressive co-design is needed across the entire stack to push up the frontier specifically for local hardware. Better kernels, inference tricks like spec decoding, quantizing models, model routing. 2. Usability / UX. To use Local AI today, you need to install CLIs, select a model, a quantization, a harness that works best with that model, and you need to configure your inference engine. It's not very accessible. More work needs to be done on making this seamless, so the user doesn't even know the AI is running locally.
1:52

GLM-5.2 running on Mac Studio Clusters or RTX GPU rigs isn't how Local AI will get mass adopted. For mass adoption we need to focus on two axes: 1. Intelligence/Speed Frontier. This is being tracked on local dot ai. Aggressive co-design is needed across the entire stack to push up the frontier specifically for local hardware. Better kernels, inference tricks like spec decoding, quantizing models, model routing. 2. Usability / UX. To use Local AI today, you need to install CLIs, select a model, a quantization, a harness that works best with that model, and you need to configure your inference engine. It's not very accessible. More work needs to be done on making this seamless, so the user doesn't even know the AI is running locally.

Alex Cheema

30,820 просмотров • 1 месяц назад

$50 free credits for GLM 5.2, KIMI 2.7, and DeepSeek V4 Pro. No promo most people know about. Your buddy pays $20/mo for Claude Pro and burns through it in a week. > You get $50 in credits on Aiand for models that actually beat Sonnet on code benchmarks - and the balance barely moves. Inside: → GLM 5.2 - the model that shipped with better coding scores than closed frontier → KIMI K2.7 Code - long-context beast, sits at the top of open leaderboards → DeepSeek V4 Pro - reasoning at a fraction of GPT-5.6 cost → OpenAI-compatible endpoint - plug into any agent, IDE, or tool you already run Setup, 5 steps: → aiand(.)com → Sign up with Google or GitHub → If the page won't load, switch to an EU VPN and retry → Top right corner - click Credits, select $50 → In Stripe checkout, apply promo code: UNITABETAI → $50 lands. Start firing prompts. Balance drains slowly. I've been running agents on it for days and barely dented it. The promo code is valid as of the time of publication Bookmark this before they cap the promo.
0:41

$50 free credits for GLM 5.2, KIMI 2.7, and DeepSeek V4 Pro. No promo most people know about. Your buddy pays $20/mo for Claude Pro and burns through it in a week. > You get $50 in credits on Aiand for models that actually beat Sonnet on code benchmarks - and the balance barely moves. Inside: → GLM 5.2 - the model that shipped with better coding scores than closed frontier → KIMI K2.7 Code - long-context beast, sits at the top of open leaderboards → DeepSeek V4 Pro - reasoning at a fraction of GPT-5.6 cost → OpenAI-compatible endpoint - plug into any agent, IDE, or tool you already run Setup, 5 steps: → aiand(.)com → Sign up with Google or GitHub → If the page won't load, switch to an EU VPN and retry → Top right corner - click Credits, select $50 → In Stripe checkout, apply promo code: UNITABETAI → $50 lands. Start firing prompts. Balance drains slowly. I've been running agents on it for days and barely dented it. The promo code is valid as of the time of publication Bookmark this before they cap the promo.

Atenov int.

14,619 просмотров • 1 месяц назад

tencent hunyuan just shipped their third major model in six months. that pace is hard to keep up with. Hy4 preview is now live in workbuddy, and it's fully open source. the specs: 770B total params, 49B active, 1M+ token context window. but the number that matters is where it's pointed — real work. coding, office tasks, research, documents. here's the part i find interesting. tencent didn't build this in isolation. they put the training data together with their own experts across software engineering, gaming, finance and security, then designed the model side by side with products like workbuddy. model and product shaping each other instead of one being bolted onto the other. and it shows up in testing. 163 internal experts ran a blind test across 203 engineering tasks. hy4 preview scored 2.99/4 — ahead of kimi k3 at 2.94 and glm 5.3 at 2.92. on public benchmarks it clearly beats glm 5.2 and sits about level with glm 5.3. then the pricing. $6 per million input tokens, $18 per million output, $0.3 per million on cache hits. flagship results at a price you don't have to think about. and for the next two weeks it's free inside workbuddy.
0:51

tencent hunyuan just shipped their third major model in six months. that pace is hard to keep up with. Hy4 preview is now live in workbuddy, and it's fully open source. the specs: 770B total params, 49B active, 1M+ token context window. but the number that matters is where it's pointed — real work. coding, office tasks, research, documents. here's the part i find interesting. tencent didn't build this in isolation. they put the training data together with their own experts across software engineering, gaming, finance and security, then designed the model side by side with products like workbuddy. model and product shaping each other instead of one being bolted onto the other. and it shows up in testing. 163 internal experts ran a blind test across 203 engineering tasks. hy4 preview scored 2.99/4 — ahead of kimi k3 at 2.94 and glm 5.3 at 2.92. on public benchmarks it clearly beats glm 5.2 and sits about level with glm 5.3. then the pricing. $6 per million input tokens, $18 per million output, $0.3 per million on cache hits. flagship results at a price you don't have to think about. and for the next two weeks it's free inside workbuddy.

Fakhr

55,316 просмотров • 5 дней назад

I think I discovered the cheapest way to code with AI 😁 I tried GLM 4.5 on Claude Code and I'm really amazed how good this model is! I usually test with my nextjs crm app benchmark so I asked it to do that. Prompt "create a nextjs app which is a crm app. use sqlite for the database and use api routes to manage connecting to the database from the frontend" My observations - Generation feels like Sonnet 4 intelligence - Speed is great - It used create-next-app so it bootstrapped nextjs v15.5.2 - It created a sqlite database and used Prisma ORM and helped me easily add seed data - I needed a few prompts to iron out minor issues but all's good. It didn't struggle or got stuck. - Spent $0.50 for this app. Could be way cheaper with the Z.ai sub plans After trying, I can say GLM 4.5 model is up there as the best open source model for coding. Will continue using it and update here. So far so good! P.S. DM me if you want my referral link, otherwise just check Not dropping the link here so it doesn’t look like I’m hyping it. Here's a video of the experience 👇(sped up a bit)
0:38

I think I discovered the cheapest way to code with AI 😁 I tried GLM 4.5 on Claude Code and I'm really amazed how good this model is! I usually test with my nextjs crm app benchmark so I asked it to do that. Prompt "create a nextjs app which is a crm app. use sqlite for the database and use api routes to manage connecting to the database from the frontend" My observations - Generation feels like Sonnet 4 intelligence - Speed is great - It used create-next-app so it bootstrapped nextjs v15.5.2 - It created a sqlite database and used Prisma ORM and helped me easily add seed data - I needed a few prompts to iron out minor issues but all's good. It didn't struggle or got stuck. - Spent $0.50 for this app. Could be way cheaper with the Z.ai sub plans After trying, I can say GLM 4.5 model is up there as the best open source model for coding. Will continue using it and update here. So far so good! P.S. DM me if you want my referral link, otherwise just check Not dropping the link here so it doesn’t look like I’m hyping it. Here's a video of the experience 👇(sped up a bit)

Melvin Vivas

45,749 просмотров • 1 год назад

So helped me to unlock the voice now it works! It's trained on my Lex Fridman podcast It's 99% perfect but the training data apparently still has some Lex "hummm" sounds in it that sometimes show up I think they call this "crosstalk" and it's hard to remove with AI so I think there's no other choice than to listen to the 1 hour of training data and remove those bits and re-train But then we have a good AI model to turn my book MAKE into an audio book without having to read it for hours/days/weeks
1:23

So helped me to unlock the voice now it works! It's trained on my Lex Fridman podcast It's 99% perfect but the training data apparently still has some Lex "hummm" sounds in it that sometimes show up I think they call this "crosstalk" and it's hard to remove with AI so I think there's no other choice than to listen to the 1 hour of training data and remove those bits and re-train But then we have a good AI model to turn my book MAKE into an audio book without having to read it for hours/days/weeks

@levelsio

169,787 просмотров • 1 год назад

YouTube and Bugbounties have changed my life (and others) in so many ways. I had been in IT for a long time but I had no idea “how to hack web”. So one day in 2017 I decided I was willing to share my curiosity and journey into a new field on YouTube. This has enabled me over the years to connected with amazing people, presented new opportunities, leveled up my creativity and deepened my understanding around business / monetization, put me on world stages and opened doors i never thought existed. sure it came with some hate and bullying but that’s also something that one learns to live and deal with. For me the sharing as I learned was a part of the journey, that also help me build my own brand. Sure you are not everyone’s favorite dish, some like chocolate others like salmiakki. But It blows my mind that we live in a time when you can learn a skill for free using a video platform, find bugs in apps, report it, sometimes get paid, and use that experience to build a career.
2:03

YouTube and Bugbounties have changed my life (and others) in so many ways. I had been in IT for a long time but I had no idea “how to hack web”. So one day in 2017 I decided I was willing to share my curiosity and journey into a new field on YouTube. This has enabled me over the years to connected with amazing people, presented new opportunities, leveled up my creativity and deepened my understanding around business / monetization, put me on world stages and opened doors i never thought existed. sure it came with some hate and bullying but that’s also something that one learns to live and deal with. For me the sharing as I learned was a part of the journey, that also help me build my own brand. Sure you are not everyone’s favorite dish, some like chocolate others like salmiakki. But It blows my mind that we live in a time when you can learn a skill for free using a video platform, find bugs in apps, report it, sometimes get paid, and use that experience to build a career.

STÖK ✌️

37,271 просмотров • 1 год назад

Tencent just dropped Hy3, and it's worth a look if you're building AI agents. I spent some time putting it through its paces today. A quick rundown of what makes this release notable: → 295B total parameters (Mixture-of-Experts), but only 21B active at inference — a real efficiency play → 256K context window for handling large codebases or long documents → Purpose-built improvements for coding and multi-step agent tasks → Fully open under Apache 2.0 — no restrictive commercial terms → A free two-week API window currently live on OpenRouter On the practical side: prompts involving layered instructions and coding tasks came back fast and coherent. Tencent says this release builds directly on feedback from 50+ internal product teams following an earlier preview, with reported drops in hallucination rate (12.5% → 5.4%) and commonsense errors (25.4% → 12.7%). Those are Tencent's own numbers, so treat them as a starting point rather than gospel until third parties weigh in. For context: Tencent's own comparisons put Hy3 behind GLM-5.2 specifically on coding benchmarks — GLM is a much bigger model (~744B total), so the trade-off makes sense. Hy3's pitch isn't "biggest," it's "efficient enough to actually deploy." Bottom line — if you're evaluating open-weight options for agent workloads, this is a solid one to add to the testing queue while the free window is open. Try it here: Tencent Hy #Hy3 #Hunyuan #TencentAI #AICoding
0:43

Tencent just dropped Hy3, and it's worth a look if you're building AI agents. I spent some time putting it through its paces today. A quick rundown of what makes this release notable: → 295B total parameters (Mixture-of-Experts), but only 21B active at inference — a real efficiency play → 256K context window for handling large codebases or long documents → Purpose-built improvements for coding and multi-step agent tasks → Fully open under Apache 2.0 — no restrictive commercial terms → A free two-week API window currently live on OpenRouter On the practical side: prompts involving layered instructions and coding tasks came back fast and coherent. Tencent says this release builds directly on feedback from 50+ internal product teams following an earlier preview, with reported drops in hallucination rate (12.5% → 5.4%) and commonsense errors (25.4% → 12.7%). Those are Tencent's own numbers, so treat them as a starting point rather than gospel until third parties weigh in. For context: Tencent's own comparisons put Hy3 behind GLM-5.2 specifically on coding benchmarks — GLM is a much bigger model (~744B total), so the trade-off makes sense. Hy3's pitch isn't "biggest," it's "efficient enough to actually deploy." Bottom line — if you're evaluating open-weight options for agent workloads, this is a solid one to add to the testing queue while the free window is open. Try it here: Tencent Hy #Hy3 #Hunyuan #TencentAI #AICoding

Felix

36,925 просмотров • 1 месяц назад

✨ Making this now in the evening cause I have some time now I asked Cursor to one-shot a plane/vehicle selector, it took some tweaks from me to get it right, like make the buttons and styles look somewhat consistent I realize AI loves to just use inline CSS for everything but that gets hard to maintain, so I clean it up manually myself now and put in tags Planes I want to add and requested by people here: - Helicopters - A-10 Warthog (with that crazy sound too so other ppl can hear it and get scared) - Speedboat with guns - Hovercraft with guns And if we add a price to them maybe some ppl will buy them and it will be a good business model Then later add/sell skins for the vehicles Also tanks etc shouldn't spawn on the aircraft carrier of course :D
0:20

✨ Making this now in the evening cause I have some time now I asked Cursor to one-shot a plane/vehicle selector, it took some tweaks from me to get it right, like make the buttons and styles look somewhat consistent I realize AI loves to just use inline CSS for everything but that gets hard to maintain, so I clean it up manually myself now and put in tags Planes I want to add and requested by people here: - Helicopters - A-10 Warthog (with that crazy sound too so other ppl can hear it and get scared) - Speedboat with guns - Hovercraft with guns And if we add a price to them maybe some ppl will buy them and it will be a good business model Then later add/sell skins for the vehicles Also tanks etc shouldn't spawn on the aircraft carrier of course :D

@levelsio

219,084 просмотров • 1 год назад

#Loverrukk shared her day off plans for this weekend 🍵🥗💪🏻💤 💖: I woke up this morning feeling really happy. I made some matcha, went to Pilates, then came back and had somtam (papaya salad). After that i sat for a bit to digest… then took a nap. I slept in the afternoon, woke up and practiced playing mlbb game for a while. After that i had dinner, sat around to digest again 😹 then took another nap in the evening. And now i’m not sleepy anymore… i wonder what time i’ll end up going to sleep. 💖: For tomorrow, i’m planning to wake up late, but i’ll probably wake up about half an hour before my alarm like usual. For example, if i set it for 9, I’ll wake up at 8:30 on my own. It just happens. Tmr is my day off too. i’ll wake up, make matcha, and order some bread, ham, rocket salad, Greek yogurt, and jam to make breakfast. I’ll also add some protein with scrambled eggs, 2 eggs with chili oil. That’ll be my breakfast. I’m happy~😸
1:24

#Loverrukk shared her day off plans for this weekend 🍵🥗💪🏻💤 💖: I woke up this morning feeling really happy. I made some matcha, went to Pilates, then came back and had somtam (papaya salad). After that i sat for a bit to digest… then took a nap. I slept in the afternoon, woke up and practiced playing mlbb game for a while. After that i had dinner, sat around to digest again 😹 then took another nap in the evening. And now i’m not sleepy anymore… i wonder what time i’ll end up going to sleep. 💖: For tomorrow, i’m planning to wake up late, but i’ll probably wake up about half an hour before my alarm like usual. For example, if i set it for 9, I’ll wake up at 8:30 on my own. It just happens. Tmr is my day off too. i’ll wake up, make matcha, and order some bread, ham, rocket salad, Greek yogurt, and jam to make breakfast. I’ll also add some protein with scrambled eggs, 2 eggs with chili oil. That’ll be my breakfast. I’m happy~😸

sᴀᴘᴘʜᴏʀɪᴀ #แม่น้องกอหญ้า🍷

30,618 просмотров • 5 месяцев назад

Tried to keep the waterworks in, couldn't help but tear up a bit in my goodbye on FOX 56 News (WDKY) tonight. 😂😢 So many thank you's to give out, which I've tried to do to everyone here this week. I just want to thank Lexington, an incredible place to have spent these last 4 years at. This town and state will always hold a special place in my heart. I've reminisced on the first era with Mitch B and Nicole Hutchison, now this era with Jordan Adams and Colby Wilson. Worked with so many great people here. Covered some awesome events and teams, worked with some great coaches like Steve Page and Justin Haddix just to name a couple. I love this job and it's the coaches, players and people who provide the stories for me to tell. I loved connecting with our viewers, and #BBN is a great fanbase too. Kentucky is a special place. I'll be back plenty of times. But now, it's time to chase some dreams.
1:36

Tried to keep the waterworks in, couldn't help but tear up a bit in my goodbye on FOX 56 News (WDKY) tonight. 😂😢 So many thank you's to give out, which I've tried to do to everyone here this week. I just want to thank Lexington, an incredible place to have spent these last 4 years at. This town and state will always hold a special place in my heart. I've reminisced on the first era with Mitch B and Nicole Hutchison, now this era with Jordan Adams and Colby Wilson. Worked with so many great people here. Covered some awesome events and teams, worked with some great coaches like Steve Page and Justin Haddix just to name a couple. I love this job and it's the coaches, players and people who provide the stories for me to tell. I loved connecting with our viewers, and #BBN is a great fanbase too. Kentucky is a special place. I'll be back plenty of times. But now, it's time to chase some dreams.

Michael Epps

11,992 просмотров • 1 год назад

People made fun of Alex Finn for buying three Mac Studios to run AI at home. Then Fable got banned for a week, GLM 5.2 dropped, and those exact Mac Studios started reselling for 4x what he paid. He showed me how he built his home AI lab from scratch. Here's the playbook: 1) The hardware. three 512GB Mac Studios, an NVIDIA DGX Spark, a custom RTX 5090 build, and a few Mac Minis. ~$30k all in. 2) The buying framework... - Mac Studio: huge memory, runs GLM 5.2 (open weights, near Opus 4.8 on benchmarks), but slow. - DGX Spark ($4,800): the sweet spot for most people. - RTX 5090: smaller models at blazing speed (Qwen's 29B now hits Sonnet 4 level). 3) Tailscale networks every machine into one private network with root access to each other. Only one machine is plugged into a monitor. 4) A Nous Research Hermes agent is his IT guy. New model drops? It SSHs into the right box, loads 5 candidates, runs evals overnight, and reports back which task belongs on which machine. Alex has literally never loaded a model himself. 5) The whole point: achieving "ambient intelligence." Always-on jobs that would bankrupt you on per-token billing. A security sweep of his API endpoints every hour. Code optimization every 20 minutes. Database anomaly & churn detection. Hourly scraping of X, Reddit & Hacker News for business opportunities. 6) Running those workloads on frontier models would cost thousands a month. His actual cost: ~$60 more in electricity. 7) Btw he's not anti-frontier. He still maxes out his Claude plan. The way he sees it: frontier is for hard thinking, local is for the foot soldiers that never sleep. 8) "We own everything except for the intelligence. Why can't we own the intelligence?" 9) He thinks frontier-level intelligence runs on consumer hardware within 6 months.
36:48

People made fun of Alex Finn for buying three Mac Studios to run AI at home. Then Fable got banned for a week, GLM 5.2 dropped, and those exact Mac Studios started reselling for 4x what he paid. He showed me how he built his home AI lab from scratch. Here's the playbook: 1) The hardware. three 512GB Mac Studios, an NVIDIA DGX Spark, a custom RTX 5090 build, and a few Mac Minis. ~$30k all in. 2) The buying framework... - Mac Studio: huge memory, runs GLM 5.2 (open weights, near Opus 4.8 on benchmarks), but slow. - DGX Spark ($4,800): the sweet spot for most people. - RTX 5090: smaller models at blazing speed (Qwen's 29B now hits Sonnet 4 level). 3) Tailscale networks every machine into one private network with root access to each other. Only one machine is plugged into a monitor. 4) A Nous Research Hermes agent is his IT guy. New model drops? It SSHs into the right box, loads 5 candidates, runs evals overnight, and reports back which task belongs on which machine. Alex has literally never loaded a model himself. 5) The whole point: achieving "ambient intelligence." Always-on jobs that would bankrupt you on per-token billing. A security sweep of his API endpoints every hour. Code optimization every 20 minutes. Database anomaly & churn detection. Hourly scraping of X, Reddit & Hacker News for business opportunities. 6) Running those workloads on frontier models would cost thousands a month. His actual cost: ~$60 more in electricity. 7) Btw he's not anti-frontier. He still maxes out his Claude plan. The way he sees it: frontier is for hard thinking, local is for the foot soldiers that never sleep. 8) "We own everything except for the intelligence. Why can't we own the intelligence?" 9) He thinks frontier-level intelligence runs on consumer hardware within 6 months.

Alex Lieberman

57,764 просмотров • 1 месяц назад

In a world of PPO everything for reinforcement learning, I've been tinkering with SAC for training a quadruped gait. This gait is trained purely on CPU (training on one of the Dell GB10s) on a single environment. Training any particular run is obviously slower than PPO on an RTX Pro 6000 with 8092 envs, if you already know the exact hyperparams/rwd function for your PPO algo... but, if we're honest with ourselves, then we know we usually spend days tuning our PPO algo and fighting it to do what we want. In contrast, SAC has kind of been a breath of fresh air, very amenable to changing the reward function to tune behavior. So far, my first attempts to tune things have consistently just worked immediately rather than 15 different variations of reward hacking only to find previous tuned behaviors got lost in the process. There is also FastSAC, which I've not yet tried, but can speed things up potentially and introduce scale back into the equation. My main painpoint in getting SAC to work for gait was actually getting it to learn to step. It seems as though SAC is not as good as PPO at significant exploration on its own. I ended up starting with a sinusoidal gait (basically just a rule to make legs swing) as training wheels then blended it out through training as phase 1, then began working on smoothing things out after this. I think if we look at end to end dev time rather than any particular run that finally managed to work, SAC may actually be the "faster" algorithm to train. Quadruped gaits are inherently easier than bipedal and maybe there are areas where SAC falls short, but I'll definitely be spending more time with SAC.
1:02

In a world of PPO everything for reinforcement learning, I've been tinkering with SAC for training a quadruped gait. This gait is trained purely on CPU (training on one of the Dell GB10s) on a single environment. Training any particular run is obviously slower than PPO on an RTX Pro 6000 with 8092 envs, if you already know the exact hyperparams/rwd function for your PPO algo... but, if we're honest with ourselves, then we know we usually spend days tuning our PPO algo and fighting it to do what we want. In contrast, SAC has kind of been a breath of fresh air, very amenable to changing the reward function to tune behavior. So far, my first attempts to tune things have consistently just worked immediately rather than 15 different variations of reward hacking only to find previous tuned behaviors got lost in the process. There is also FastSAC, which I've not yet tried, but can speed things up potentially and introduce scale back into the equation. My main painpoint in getting SAC to work for gait was actually getting it to learn to step. It seems as though SAC is not as good as PPO at significant exploration on its own. I ended up starting with a sinusoidal gait (basically just a rule to make legs swing) as training wheels then blended it out through training as phase 1, then began working on smoothing things out after this. I think if we look at end to end dev time rather than any particular run that finally managed to work, SAC may actually be the "faster" algorithm to train. Quadruped gaits are inherently easier than bipedal and maybe there are areas where SAC falls short, but I'll definitely be spending more time with SAC.

Harrison Kinsley

26,758 просмотров • 6 месяцев назад

Everyday Americans Are Using Their Helicopters To Bring Aid To Hurricane Helene Victims And the Kamala Harris Administration is threatening to arrest these people for trying to help their fellow Americans “I got a call from a friend up in North Carolina. His name's Rudy, and he said, "'Hey man, we need some helicopters up here. "'The hurricane damage is really bad, "'and there's a lot of people trapped in the mountains.'" So my wife and I got someone to watch the kids. We loaded up with one bag and flew to North Carolina with no plans and posted that we were on our way.” “By the end of day two, there was 37 helicopters, private helicopters on site at Hickory Airport helping out.” “Some of my friends brought in some sick helicopters. A lot of these guys just did an amazing job”
7:03

Everyday Americans Are Using Their Helicopters To Bring Aid To Hurricane Helene Victims And the Kamala Harris Administration is threatening to arrest these people for trying to help their fellow Americans “I got a call from a friend up in North Carolina. His name's Rudy, and he said, "'Hey man, we need some helicopters up here. "'The hurricane damage is really bad, "'and there's a lot of people trapped in the mountains.'" So my wife and I got someone to watch the kids. We loaded up with one bag and flew to North Carolina with no plans and posted that we were on our way.” “By the end of day two, there was 37 helicopters, private helicopters on site at Hickory Airport helping out.” “Some of my friends brought in some sick helicopters. A lot of these guys just did an amazing job”

Wall Street Apes

1,162,532 просмотров • 1 год назад

Hassan Whiteside went from training at the YMCA to signing a $98,000,000 deal with the Miami Heat “I was going to the YMCA to stay in shape. I'm over there lifting, running, playing pickup on the side. That was my training facility. The YMCA was my training facility” “I had a workout with the Grizzlies. I was in training camp with them, and I think me and Earl Clark were two of the last guys to get let go. And I remember going to Miami to work out, and Pat Riley personally came out on the court and worked me out. Hardest workout of my life” “I left the workout and went down to the G League. The Grizzlies' G League team sent me down there. I had a really good game. I think I had like 30 and 20 or something. My agent called me playing ‘Welcome to Miami.’ He was like, Miami wants to sign you to a four-year, non-guaranteed deal” “I ended up signing a one year, and then I signed a two year non-guaranteed deal. A lot of people don't know I wasn’t even on a guaranteed contract for a lot of the time I was with the Heat”
1:52

Hassan Whiteside went from training at the YMCA to signing a $98,000,000 deal with the Miami Heat “I was going to the YMCA to stay in shape. I'm over there lifting, running, playing pickup on the side. That was my training facility. The YMCA was my training facility” “I had a workout with the Grizzlies. I was in training camp with them, and I think me and Earl Clark were two of the last guys to get let go. And I remember going to Miami to work out, and Pat Riley personally came out on the court and worked me out. Hardest workout of my life” “I left the workout and went down to the G League. The Grizzlies' G League team sent me down there. I had a really good game. I think I had like 30 and 20 or something. My agent called me playing ‘Welcome to Miami.’ He was like, Miami wants to sign you to a four-year, non-guaranteed deal” “I ended up signing a one year, and then I signed a two year non-guaranteed deal. A lot of people don't know I wasn’t even on a guaranteed contract for a lot of the time I was with the Heat”

dank

79,384 просмотров • 1 месяц назад

We’ve been hit with a winter storm here in Alabama—some are calling it an ice storm, and that’s pretty accurate for my area. Not much snow yet, but plenty of ice, rain, and sleet. Today I’m out spreading ice melt on driveways, sidewalks, and ramps for some of the elderly, disabled, single parents, and veterans we serve locally. I’m currently at Mr. Wayne’s house in this video, putting ice melt down on his ramp, he had some ice on it . It will get worst later A few days ago, I picked up some ice melt and a spreader to help with situations like this—so this is actually my first time using it. Just doing what we can to keep our neighbors safe during this weather.
2:20

We’ve been hit with a winter storm here in Alabama—some are calling it an ice storm, and that’s pretty accurate for my area. Not much snow yet, but plenty of ice, rain, and sleet. Today I’m out spreading ice melt on driveways, sidewalks, and ramps for some of the elderly, disabled, single parents, and veterans we serve locally. I’m currently at Mr. Wayne’s house in this video, putting ice melt down on his ramp, he had some ice on it . It will get worst later A few days ago, I picked up some ice melt and a spreader to help with situations like this—so this is actually my first time using it. Just doing what we can to keep our neighbors safe during this weather.

Rodney Smith Jr

188,868 просмотров • 7 месяцев назад

Another quick lecture -- I've been asked many times for prereq's to my book and what you should know, so built a little lecture (with GLM 5.2) to cover some more basics. Topics include: 00:00 Introduction & Course Prerequisites 01:37 Language Models Overview 02:47 The LM Head 04:29 Softmax & Log-Probabilities 06:13 Anatomy of an LM Training Example 06:37 Computing LLM Probabilities (+Phoebe the Dog) 09:52 Three Common Masks in Post-Training 11:03 A Small Decoding Review 12:14 Training an LM: Cross-Entropy 13:23 Optimization & Fine-Tuning 13:55 Pretraining to Midtraining to SFT Pipeline 15:25 Probability Essentials: KL Divergence & Entropy 19:36 Sigmoid & Pairwise Likelihood 20:29 Reinforcement Learning Framing (MDP) 22:28 Transitioning Tools into Post-Training 23:12 Recommended Resources & Wrap-Up  Happy learning and I'm still taking questions from during the course for Q&A videos.
23:52

Another quick lecture -- I've been asked many times for prereq's to my book and what you should know, so built a little lecture (with GLM 5.2) to cover some more basics. Topics include: 00:00 Introduction & Course Prerequisites 01:37 Language Models Overview 02:47 The LM Head 04:29 Softmax & Log-Probabilities 06:13 Anatomy of an LM Training Example 06:37 Computing LLM Probabilities (+Phoebe the Dog) 09:52 Three Common Masks in Post-Training 11:03 A Small Decoding Review 12:14 Training an LM: Cross-Entropy 13:23 Optimization & Fine-Tuning 13:55 Pretraining to Midtraining to SFT Pipeline 15:25 Probability Essentials: KL Divergence & Entropy 19:36 Sigmoid & Pairwise Likelihood 20:29 Reinforcement Learning Framing (MDP) 22:28 Transitioning Tools into Post-Training 23:12 Recommended Resources & Wrap-Up  Happy learning and I'm still taking questions from during the course for Q&A videos.

Nathan Lambert

52,275 просмотров • 2 месяцев назад