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.

Daniel Han wrote Unsloth, the reason half of open-source can fine-tune a model on one GPU instead of a cluster. He didn't optimize the math. He rewrote the kernels by hand, found bugs in everyone else's code, and made training 2 to 3 times faster with zero accuracy loss.... show more

h100envy

3,744 subscribers

30,856 views • 3 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

12 Comments

hrabiapolski3 months ago

its really good info

zostaff3 months ago

this guy make our future

h100envy3 months ago

true 100%

Iana‧₊˚3 months ago

most infra work doesn't touch this many runs

0xSlyth3 months ago

rewrote the game no one saw it coming

The real Burgstall3 months ago

@chrisbward He's a legend

Nekt03 months ago

Good bro

💫ℹ️🐚▪️🌐💺🗨️🚨®️3 months ago

He is a genius. He single handedly done more work for the open-source Ai world .

David Roth 🇺🇸3 months ago

Excellent insight! Loved the questions and responses---genuine curiosity meets knowledgeable enthusiasm.

h100envy3 months ago

Ty my friend, i glad for you

AI Mastery Guide3 months ago

One guy quietly making everyone's GPU enough, that's a wild contribution 🔥

🧪Juice 🧃3 months ago

Unsloth works great. I built an auto training tool with it as my back end.

Related Videos

Ex-NVIDIA engineer who built Unsloth explained RL, kernels, reasoning, quantization, and agents in 2 hours 42 minutes - better than $5000 fine-tuning bootcamps. pick the base model -> write triton kernels for 2x faster fine-tune -> quantize to 4-bit -> run GRPO/DPO -> ship a reasoning model on your single GPU. That loop is why Unsloth is the default way to fine-tune Llama, Qwen, Gemma, and Phi on hardware you already own. Unsloth + Triton kernels + 4-bit quantization + GRPO/DPO + single-GPU fine-tuning - that's the stack. Watch and save it, then fine-tune your first model tonight.

Ex-NVIDIA engineer who built Unsloth explained RL, kernels, reasoning, quantization, and agents in 2 hours 42 minutes - better than $5000 fine-tuning bootcamps. pick the base model -> write triton kernels for 2x faster fine-tune -> quantize to 4-bit -> run GRPO/DPO -> ship a reasoning model on your single GPU. That loop is why Unsloth is the default way to fine-tune Llama, Qwen, Gemma, and Phi on hardware you already own. Unsloth + Triton kernels + 4-bit quantization + GRPO/DPO + single-GPU fine-tuning - that's the stack. Watch and save it, then fine-tune your first model tonight.

h100envy

496,659 views • 2 months ago

Many people are flat wrong about DeepSeek. You might think everyone freaked out about DeepSeek because the model is really good—which it is—or because it was from China, which is also true. But there's a more subtle reason: DeepSeek showed they could achieve those results using reinforcement learning instead of supervised fine-tuning. In other words, they didn't need an expensive phase where a bunch of people tuned the model by answering questions. They were able to do all of this faster and cheaper using reinforcement fine-tuning. This is a huge deal! I recorded the attached video with the following goals in mind: 1. Explain how reinforcement fine-tuning works 2. Show you how you can fine-tune your own models 3. Walk you through a complete code example

Many people are flat wrong about DeepSeek. You might think everyone freaked out about DeepSeek because the model is really good—which it is—or because it was from China, which is also true. But there's a more subtle reason: DeepSeek showed they could achieve those results using reinforcement learning instead of supervised fine-tuning. In other words, they didn't need an expensive phase where a bunch of people tuned the model by answering questions. They were able to do all of this faster and cheaper using reinforcement fine-tuning. This is a huge deal! I recorded the attached video with the following goals in mind: 1. Explain how reinforcement fine-tuning works 2. Show you how you can fine-tune your own models 3. Walk you through a complete code example

Santiago

105,589 views • 1 year ago

Ex-JPMorgan engineer who wrote the LLM Course explained everything about fine-tuning and merging in 18 minutes - better than $2500 fine-tuning bootcamps. pick the base -> LoRA or QLoRA -> then DPO or KTO for alignment -> merge two fine-tunes into one stronger model -> ship a model that beats the base on your task. That loop is why Labonne's merges sit at the top of the Hugging Face leaderboard. LoRA + QLoRA + DPO + KTO + mergekit - that's the stack. Watch and save it, then merge your first two fine-tunes this week.

Ex-JPMorgan engineer who wrote the LLM Course explained everything about fine-tuning and merging in 18 minutes - better than $2500 fine-tuning bootcamps. pick the base -> LoRA or QLoRA -> then DPO or KTO for alignment -> merge two fine-tunes into one stronger model -> ship a model that beats the base on your task. That loop is why Labonne's merges sit at the top of the Hugging Face leaderboard. LoRA + QLoRA + DPO + KTO + mergekit - that's the stack. Watch and save it, then merge your first two fine-tunes this week.

h100envy

13,856 views • 2 months ago

Anthropic Quant Andrej Karpathy: "Most people use tools that they don't understand - the ones who strip everything down to basics - end up faster than everyone else " "the best code is the code anyone can read " he couldn't fix a bug in 2 hours, so instead of googling - he rewrote the entire system from scratch no frameworks. no dependencies. it ended up faster that's the difference between using AI and understanding it 25-min masterclass - bookmark and watch

Anthropic Quant Andrej Karpathy: "Most people use tools that they don't understand - the ones who strip everything down to basics - end up faster than everyone else " "the best code is the code anyone can read " he couldn't fix a bug in 2 hours, so instead of googling - he rewrote the entire system from scratch no frameworks. no dependencies. it ended up faster that's the difference between using AI and understanding it 25-min masterclass - bookmark and watch

codila

380,157 views • 3 months ago

Boris Cherny, the creator of Claude Code: "Most people use Claude Code as a chatbot. It was built to run an entire engineering team." In 30 minutes on stage, he explains how he now works by directing agents instead of writing code by hand. Watch full video, then save the exact setup below👇

Boris Cherny, the creator of Claude Code: "Most people use Claude Code as a chatbot. It was built to run an entire engineering team." In 30 minutes on stage, he explains how he now works by directing agents instead of writing code by hand. Watch full video, then save the exact setup below👇

darkzodchi

35,506 views • 3 months ago

Ying Sheng co-wrote SGLang, the inference engine now serving Grok at xAI on a hundred thousand GPUs. She also built FlexGen, which made a 175-billion model run on a single consumer GPU, and helped build Chatbot Arena. Three artifacts the whole field uses, one researcher. SGLang hit a 5x cost cut over DeepSeek's own API, and a dozen teams reproduced it. Everyone argues about models. She builds the engines that actually serve them cheaply enough to survive.

Ying Sheng co-wrote SGLang, the inference engine now serving Grok at xAI on a hundred thousand GPUs. She also built FlexGen, which made a 175-billion model run on a single consumer GPU, and helped build Chatbot Arena. Three artifacts the whole field uses, one researcher. SGLang hit a 5x cost cut over DeepSeek's own API, and a dozen teams reproduced it. Everyone argues about models. She builds the engines that actually serve them cheaply enough to survive.

h100envy

40,959 views • 3 months ago

he made 405-billion-parameter models run on a single 8GB gaming GPU. instead of turning it into a paid service, he open-sourced the whole thing. Gavin Li Founder and CEO of Anima AI in the Bay Area. 'airllm' - it keeps one layer on the GPU at a time and streams the rest, so a 70B model fits in 4GB and DeepSeek's full 671B fits in about 12GB 22,700 stars. Apache 2.0. → / he has 808 followers. / his repo has 22,700 stars. bookmark it. that's what shipping something impossible looks like.

he made 405-billion-parameter models run on a single 8GB gaming GPU. instead of turning it into a paid service, he open-sourced the whole thing. Gavin Li Founder and CEO of Anima AI in the Bay Area. 'airllm' - it keeps one layer on the GPU at a time and streams the rest, so a 70B model fits in 4GB and DeepSeek's full 671B fits in about 12GB 22,700 stars. Apache 2.0. → / he has 808 followers. / his repo has 22,700 stars. bookmark it. that's what shipping something impossible looks like.

dunik

110,908 views • 2 months ago

THIS DEVELOPER JUST RAN A TRILLION PARAMETER MODEL ON 4 MAC STUDIOS - 10X FASTER AND 5X CHEAPER THAN CLOUD CODE 19:00 he says it out loud. "we just ran a trillion parameter model. 30 something tokens per second. wow." RDMA over Thunderbolt made the cluster 10x faster than before. tensor parallelism splits the model between machines in parallel instead of sequentially DeepSeek V3 full size without quantization. 183GB per machine. still 25 tokens per second. faster than most people can read Kimi K2 with trillion parameters and 256,000 context window. loaded, responded and adapted memory usage dynamically based on prompt size 4 machines consuming 66 watts total. cloud GPU for the same workload costs $1,900/month a cluster that used to cost millions now assembled from Mac Studios in a few hours

THIS DEVELOPER JUST RAN A TRILLION PARAMETER MODEL ON 4 MAC STUDIOS - 10X FASTER AND 5X CHEAPER THAN CLOUD CODE 19:00 he says it out loud. "we just ran a trillion parameter model. 30 something tokens per second. wow." RDMA over Thunderbolt made the cluster 10x faster than before. tensor parallelism splits the model between machines in parallel instead of sequentially DeepSeek V3 full size without quantization. 183GB per machine. still 25 tokens per second. faster than most people can read Kimi K2 with trillion parameters and 256,000 context window. loaded, responded and adapted memory usage dynamically based on prompt size 4 machines consuming 66 watts total. cloud GPU for the same workload costs $1,900/month a cluster that used to cost millions now assembled from Mac Studios in a few hours

Noisy

80,087 views • 3 months ago

AMD's Anush Elangovan explains why he thinks his company's open source ethos combined with agentic AI superpowers their leverage as a company: Because AMD publishes a lot of technical details about its hardware, when engineers use AI tools, the models already “understand” AMD’s systems and can help write code for them, debug them, or even generate new tools. And that makes developers more productive on AMD hardware without AMD having to do all the work internally. "AMD has had this ethos of open source, which really plays to our advantage. Every frontier model that I use has already seen every bit of AMD source code." "It'll rewrite my spec for me because it's already in the training data. Which you can't get from closed ecosystems." "In fact I built a virtual GPU simulator just based off our public specs, and now I'm running it on the GPU. So now I can run cross-generational GPU simulations on existing hardware." "We have that advantage. And we've run a Dev Day contest where we generated more tokens on AMD — Triton kernels and HIP kernels — than existed on the internet at the time." "So now that's all part of the pre-training data. It's a superpower because now you're open source, and you're agentically accelerating this process."

AMD's Anush Elangovan explains why he thinks his company's open source ethos combined with agentic AI superpowers their leverage as a company: Because AMD publishes a lot of technical details about its hardware, when engineers use AI tools, the models already “understand” AMD’s systems and can help write code for them, debug them, or even generate new tools. And that makes developers more productive on AMD hardware without AMD having to do all the work internally. "AMD has had this ethos of open source, which really plays to our advantage. Every frontier model that I use has already seen every bit of AMD source code." "It'll rewrite my spec for me because it's already in the training data. Which you can't get from closed ecosystems." "In fact I built a virtual GPU simulator just based off our public specs, and now I'm running it on the GPU. So now I can run cross-generational GPU simulations on existing hardware." "We have that advantage. And we've run a Dev Day contest where we generated more tokens on AMD — Triton kernels and HIP kernels — than existed on the internet at the time." "So now that's all part of the pre-training data. It's a superpower because now you're open source, and you're agentically accelerating this process."

TBPN

34,469 views • 4 months ago

You can now fine-tune Llama 3 without writing a single line of code! We are moving at breakneck speed. I recorded a video to show you how to fine-tune any open-source model in a few minutes. I'm using a GPT capable of taking a problem and turning it into a fine-tuned model that will solve it. You don't have to write any code. You only need to explain to a GPT what problem you want to solve and tell it you want to use Llama 3. For example, "fine-tune Llama 3" or "deploy zephyr." It feels magic. The system will recommend a dataset and fine-tune the model for you. I'm using Monster API, a platform that specializes in making fine-tuning and deploying open-source models easy and fast. Their stack is well-optimized to maximize fine-tuning efficiency using techniques like Q-Lora and vLLM. They are behind the GPT. Here is what you need to do: 1. Create an account at 2. Load the GPT with the link below This is as simple as it gets. When you are done, you can click a button to deploy the model and start using it. I have 10,000 free credits for anyone using the code "SANTIAGO" in the dashboard. You can use these credits to access, fine-tune, and deploy these open-source models. You can also keep up with their latest updates, and get free credits and special offers on their Discord server:

You can now fine-tune Llama 3 without writing a single line of code! We are moving at breakneck speed. I recorded a video to show you how to fine-tune any open-source model in a few minutes. I'm using a GPT capable of taking a problem and turning it into a fine-tuned model that will solve it. You don't have to write any code. You only need to explain to a GPT what problem you want to solve and tell it you want to use Llama 3. For example, "fine-tune Llama 3" or "deploy zephyr." It feels magic. The system will recommend a dataset and fine-tune the model for you. I'm using Monster API, a platform that specializes in making fine-tuning and deploying open-source models easy and fast. Their stack is well-optimized to maximize fine-tuning efficiency using techniques like Q-Lora and vLLM. They are behind the GPT. Here is what you need to do: 1. Create an account at 2. Load the GPT with the link below This is as simple as it gets. When you are done, you can click a button to deploy the model and start using it. I have 10,000 free credits for anyone using the code "SANTIAGO" in the dashboard. You can use these credits to access, fine-tune, and deploy these open-source models. You can also keep up with their latest updates, and get free credits and special offers on their Discord server:

Santiago

324,621 views • 2 years ago

A 12-year-old made $23,000 in a month from one Fortnite map while his classmates were just playing the same game. He understood one thing everyone missed - Epic pays for every minute a player spends on his island. He described the game mode to Claude Code and it wrote all the Verse code without a single line by hand. Currency, upgrades, leaderboard, daily bonus system - 10 hours of work and the map was live. 187,000 minutes of playtime in 30 days. One check from Epic. $722 million already paid out to creators. 58 people became millionaires in 2024. The top creator made $20,400,000 from one map inside a video game. Until the end of 2026 Epic gives 100% of direct item sales - after that it drops to 50% permanently. The only barrier was Verse. Claude Code removed it completely.

A 12-year-old made $23,000 in a month from one Fortnite map while his classmates were just playing the same game. He understood one thing everyone missed - Epic pays for every minute a player spends on his island. He described the game mode to Claude Code and it wrote all the Verse code without a single line by hand. Currency, upgrades, leaderboard, daily bonus system - 10 hours of work and the map was live. 187,000 minutes of playtime in 30 days. One check from Epic. $722 million already paid out to creators. 58 people became millionaires in 2024. The top creator made $20,400,000 from one map inside a video game. Until the end of 2026 Epic gives 100% of direct item sales - after that it drops to 50% permanently. The only barrier was Verse. Claude Code removed it completely.

Cortex

28,945 views • 4 months ago

He walked into a small shop and made it visible to the entire planet. One afternoon. 170 scans. A free app on his phone. He gets $400 for the scan. $99 every month for hosting. The result: a full 3D virtual tour anyone can walk through in a browser. Every shelf. Every product. Every detail of a store that most people would never find on a map. The owner gets more online visitors in a week than she got in-store visitors in a month. That gap is the business. Month one: $3,500. Month six: $18,000. The same model works for hotels, real estate listings, event venues, car dealerships, museums. Every physical space in every city is an untapped asset. Most owners have no idea this exists. He didn't wait for the right client. He walked into a shop and made one.

He walked into a small shop and made it visible to the entire planet. One afternoon. 170 scans. A free app on his phone. He gets $400 for the scan. $99 every month for hosting. The result: a full 3D virtual tour anyone can walk through in a browser. Every shelf. Every product. Every detail of a store that most people would never find on a map. The owner gets more online visitors in a week than she got in-store visitors in a month. That gap is the business. Month one: $3,500. Month six: $18,000. The same model works for hotels, real estate listings, event venues, car dealerships, museums. Every physical space in every city is an untapped asset. Most owners have no idea this exists. He didn't wait for the right client. He walked into a shop and made one.

Superior

16,952 views • 4 months ago

Dario Amodei was asked whether open source will eventually gut Anthropic's business. He didn't defend the moat. He didn't argue closed beats open. He said the whole question is a red herring. That is the reframe. And it flips how the industry keeps scoring this race. The conventional narrative is inherited from the last era of tech: open source wins because anyone can read the code, anyone improves it, contributions stack, and eventually the free thing catches the paid thing. Investors have a full lexicon for it. Commoditization. Which layer captures the value. Everyone repeats it. Amodei says the analogy breaks at the root. It's called open weights, not open source, for a reason: you can't see inside the model. So the thing that actually made open source powerful elsewhere, many people reading and additively improving shared code, never transfers. You just get a large file of numbers. Now here's where it gets interesting. The second engine isn't ideology. It's infrastructure. Free isn't free. Someone still has to host it. These are big models, and they're hard to run inference on. Someone has to make that fast. And the capabilities people assume only open weights unlock fine tuning, steering, inspecting activations labs are increasingly serving on their own clouds anyway. When DeepSeek shipped, he says he never asked whether it was open. He asked one thing: is it a good model, and is it better than us. That's the only axis he competes on. He even inverts the usual edge. Coming from outside that investor lexicon, he thinks knowing none of it lets him predict this better than the people fluent in it. He is not defending closed models. He is saying the scoreboard everyone is watching measures the wrong thing. The uncomfortable question if the free model still needs someone to run it, was the moat ever the weights, or always the machine underneath ?

Dario Amodei was asked whether open source will eventually gut Anthropic's business. He didn't defend the moat. He didn't argue closed beats open. He said the whole question is a red herring. That is the reframe. And it flips how the industry keeps scoring this race. The conventional narrative is inherited from the last era of tech: open source wins because anyone can read the code, anyone improves it, contributions stack, and eventually the free thing catches the paid thing. Investors have a full lexicon for it. Commoditization. Which layer captures the value. Everyone repeats it. Amodei says the analogy breaks at the root. It's called open weights, not open source, for a reason: you can't see inside the model. So the thing that actually made open source powerful elsewhere, many people reading and additively improving shared code, never transfers. You just get a large file of numbers. Now here's where it gets interesting. The second engine isn't ideology. It's infrastructure. Free isn't free. Someone still has to host it. These are big models, and they're hard to run inference on. Someone has to make that fast. And the capabilities people assume only open weights unlock fine tuning, steering, inspecting activations labs are increasingly serving on their own clouds anyway. When DeepSeek shipped, he says he never asked whether it was open. He asked one thing: is it a good model, and is it better than us. That's the only axis he competes on. He even inverts the usual edge. Coming from outside that investor lexicon, he thinks knowing none of it lets him predict this better than the people fluent in it. He is not defending closed models. He is saying the scoreboard everyone is watching measures the wrong thing. The uncomfortable question if the free model still needs someone to run it, was the moat ever the weights, or always the machine underneath ?

Vikram M

58,688 views • 2 months ago

Small Language Models (SML) are the future of AI. "Small" (SML) instead of "Large" (LLM). These small models are highly specialized models with superhuman abilities on specific tasks. Here are two techniques to build these models: • Spectrum • Model Merging I give you a short introduction in the attached video, but here is a quick summary: Spectrum helps us identify the most relevant layers to solve one specific task. We can ignore everything else and focus on fine-tuning these layers. Using Spectrum, we can fine-tune models in a heartbeat. Model Merging combines multiple models into a unique, much better model than any of the individual input models. You can also combine models specialized in different tasks and get a model with multiple abilities. This is the state of the art of productizing models. It's what Arcee.ai's platform does behind the scenes. Arcee collaborated with me on this post and is sponsoring it. There are three main steps to produce a model for your particular use case: 1. You create a dataset by uploading your data. 2. You train a model. At this step, Arcee uses Spectrum and Model Merging to produce a highly specialized model for your task. 3. You can deploy that model to any environment you want. Three important notes: • Training process is 2x faster and 2x cheaper than regular fine-tuning. • Resultant models are smaller and have higher accuracy. • They create these specialized models from open-source models. Check this site so you can fully appreciate how this works: If you want to fine-tune an open-source model, consider Arcee's platform. This is the state of the art.

Small Language Models (SML) are the future of AI. "Small" (SML) instead of "Large" (LLM). These small models are highly specialized models with superhuman abilities on specific tasks. Here are two techniques to build these models: • Spectrum • Model Merging I give you a short introduction in the attached video, but here is a quick summary: Spectrum helps us identify the most relevant layers to solve one specific task. We can ignore everything else and focus on fine-tuning these layers. Using Spectrum, we can fine-tune models in a heartbeat. Model Merging combines multiple models into a unique, much better model than any of the individual input models. You can also combine models specialized in different tasks and get a model with multiple abilities. This is the state of the art of productizing models. It's what Arcee.ai's platform does behind the scenes. Arcee collaborated with me on this post and is sponsoring it. There are three main steps to produce a model for your particular use case: 1. You create a dataset by uploading your data. 2. You train a model. At this step, Arcee uses Spectrum and Model Merging to produce a highly specialized model for your task. 3. You can deploy that model to any environment you want. Three important notes: • Training process is 2x faster and 2x cheaper than regular fine-tuning. • Resultant models are smaller and have higher accuracy. • They create these specialized models from open-source models. Check this site so you can fully appreciate how this works: If you want to fine-tune an open-source model, consider Arcee's platform. This is the state of the art.

Santiago

164,162 views • 2 years ago

Michael Jordan once played in front of 62,046 people, the biggest crowd in NBA history. Late in the game, with nothing on the line, he shot a free throw with his left hand and made it. There's a real reason he did it. Most people have never heard it.

Michael Jordan once played in front of 62,046 people, the biggest crowd in NBA history. Late in the game, with nothing on the line, he shot a free throw with his left hand and made it. There's a real reason he did it. Most people have never heard it.

OLDSKOOLBBALL

31,934 views • 21 days ago

Vibe coding is cool. But have you tried vibe-tuning? Here’s Claude Code using the Lightning Rod Labs SDK to turn messy Fed PDFs into a small, specialized economic forecaster. The agent: - parses raw historical PDFs - builds a training set - fine-tunes a smaller model - evaluates it against GPT-5 Result: a fine-tuned model that beats GPT-5 and runs anywhere, on a single GPU. Try it in your coding agent of choice.

Vibe coding is cool. But have you tried vibe-tuning? Here’s Claude Code using the Lightning Rod Labs SDK to turn messy Fed PDFs into a small, specialized economic forecaster. The agent: - parses raw historical PDFs - builds a training set - fine-tunes a smaller model - evaluates it against GPT-5 Result: a fine-tuned model that beats GPT-5 and runs anywhere, on a single GPU. Try it in your coding agent of choice.

Ben Turtel

423,542 views • 4 months ago

Jane Street brought in Horace He to explain how to squeeze the most out of a single GPU. And then how to scale it to tens of thousands. Graduated from Cornell in 2020. Works on the PyTorch team. Training Llama 3 takes 40 trillion trillion floating point operations. He wrote the two things those models run on: torch.compile and FlexAttention.

Jane Street brought in Horace He to explain how to squeeze the most out of a single GPU. And then how to scale it to tens of thousands. Graduated from Cornell in 2020. Works on the PyTorch team. Training Llama 3 takes 40 trillion trillion floating point operations. He wrote the two things those models run on: torch.compile and FlexAttention.

zostaff

51,764 views • 3 months ago

Bare imitation learning without robotic pre-training is an unfairly overlooked area. (it doesn't even have a name) When done right, it has the advantages of programming. It’s reliable, predictable, easy to modify and debug. The only difference is that the source of truth is a small dataset instead of code. And it is glued together by a vision encoder and a transformer model instead of being compiled. It's hard to imagine how Claude Code could appear without a ton of production-grade source code hardened over years of real use. Same for robotic models. Here is the robustness of such a model. It was trained on orange zip ties with no lighting changes.

Bare imitation learning without robotic pre-training is an unfairly overlooked area. (it doesn't even have a name) When done right, it has the advantages of programming. It’s reliable, predictable, easy to modify and debug. The only difference is that the source of truth is a small dataset instead of code. And it is glued together by a vision encoder and a transformer model instead of being compiled. It's hard to imagine how Claude Code could appear without a ton of production-grade source code hardened over years of real use. Same for robotic models. Here is the robustness of such a model. It was trained on orange zip ties with no lighting changes.

Igor Kulakov (MicroFactory)

20,234 views • 2 months ago

underrated gems in Kimi-K3 release: > an early K3 wrote the majority of the kernels in the late development stages > it built a triton-class compiler from scratch, MiniTriton, that delivers performance on par with or better than Triton and torch.compile > then it designed a chip, by a model, for a model, in one 48-hour autonomous run the model is rewriting and optimizing every layer of the stack it runs on: kernels for its own training. A compiler for its own kernels. silicon for its own weights. we're watching software build its own hardware

underrated gems in Kimi-K3 release: > an early K3 wrote the majority of the kernels in the late development stages > it built a triton-class compiler from scratch, MiniTriton, that delivers performance on par with or better than Triton and torch.compile > then it designed a chip, by a model, for a model, in one 48-hour autonomous run the model is rewriting and optimizing every layer of the stack it runs on: kernels for its own training. A compiler for its own kernels. silicon for its own weights. we're watching software build its own hardware

Alex Ker 🔭

311,363 views • 2 months ago