Loading video...

Video Failed to Load

Go Home

Luminal ( is creating PyTorch for Production – an ML compiler that generates blazingly fast CUDA kernels and makes deploying to production one line of code. Congrats on the launch, Jake Stevens, Joe Fioti, and Matthew Gunton!

ycombinator's profile picture

Y Combinator

1,537,083 subscribers

98,496 views • 1 year ago •via X (Twitter)

Science & TechnologyNews & PoliticsEducation
LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Introducing The AI CUDA Engineer: An agentic AI system that automates the production of highly optimized CUDA kernels. The AI CUDA Engineer can produce highly optimized CUDA kernels, reaching 10-100x speedup over common machine learning operations in PyTorch. Our system is also able to produce highly optimized CUDA kernels that are much faster than existing CUDA kernels commonly used in production. We believe that fundamentally, AI systems can and should be as resource-efficient as the human brain, and that the best path to achieve this efficiency is to use AI to make AI more efficient! We are excited to publish our paper, The AI CUDA Engineer: Agentic CUDA Kernel Discovery, Optimization and Composition. We also release a dataset of over 17,000 verified CUDA kernels produced by The AI CUDA Engineer. Paper: Kernel Archive Webpage: HuggingFace Dataset: The AI CUDA Engineer utilizes evolutionary LLM-driven code optimization to autonomously improve the runtime of machine learning operations. Our system is not only able to convert PyTorch code into CUDA kernels, but through the use of evolution, it can also optimize the runtime performance of CUDA kernels, fuse multiple operations, and even discover novel solutions for writing efficient CUDA operations by learning from past innovations! We believe The AI CUDA Engineer opens a new era of AI-driven acceleration of AI and automated inference time optimization. We (Robert Lange, Aaditya Prasad 🇺🇸, sssss, Maxence Faldor, Yujin Tang, hardmaru) are excited to continue Sakana AI's mission of leveraging AI to improve AI.
1:21

Introducing The AI CUDA Engineer: An agentic AI system that automates the production of highly optimized CUDA kernels. The AI CUDA Engineer can produce highly optimized CUDA kernels, reaching 10-100x speedup over common machine learning operations in PyTorch. Our system is also able to produce highly optimized CUDA kernels that are much faster than existing CUDA kernels commonly used in production. We believe that fundamentally, AI systems can and should be as resource-efficient as the human brain, and that the best path to achieve this efficiency is to use AI to make AI more efficient! We are excited to publish our paper, The AI CUDA Engineer: Agentic CUDA Kernel Discovery, Optimization and Composition. We also release a dataset of over 17,000 verified CUDA kernels produced by The AI CUDA Engineer. Paper: Kernel Archive Webpage: HuggingFace Dataset: The AI CUDA Engineer utilizes evolutionary LLM-driven code optimization to autonomously improve the runtime of machine learning operations. Our system is not only able to convert PyTorch code into CUDA kernels, but through the use of evolution, it can also optimize the runtime performance of CUDA kernels, fuse multiple operations, and even discover novel solutions for writing efficient CUDA operations by learning from past innovations! We believe The AI CUDA Engineer opens a new era of AI-driven acceleration of AI and automated inference time optimization. We (Robert Lange, Aaditya Prasad 🇺🇸, sssss, Maxence Faldor, Yujin Tang, hardmaru) are excited to continue Sakana AI's mission of leveraging AI to improve AI.

Sakana AI

1,160,190 views • 1 year ago

Attune (Attune) makes blazingly fast build tools for developers that require zero migration effort. Built by developers with a decade of experience in build systems and developer tools. Congrats on the launch, xin and Eliza Zhang!
0:59

Attune (Attune) makes blazingly fast build tools for developers that require zero migration effort. Built by developers with a decade of experience in build systems and developer tools. Congrats on the launch, xin and Eliza Zhang!

Y Combinator

16,787 views • 1 year ago

LineWise (LineWise) is building an AI assistant to help manufacturing teams get their production lines back up and running— fast. Congrats on the launch, Tanachart Kujareevanich, ZHICHU REN, and Wenbo Zhang!
0:50

LineWise (LineWise) is building an AI assistant to help manufacturing teams get their production lines back up and running— fast. Congrats on the launch, Tanachart Kujareevanich, ZHICHU REN, and Wenbo Zhang!

Y Combinator

28,665 views • 1 year ago

Subtrace lets developers resolve production issues faster. Unlike Sentry and other backend monitoring tools, Subtrace makes it easy to search for the requests that truly matter during an incident. Congrats on the launch, and Sachin Sridhar!
0:58

Subtrace lets developers resolve production issues faster. Unlike Sentry and other backend monitoring tools, Subtrace makes it easy to search for the requests that truly matter during an incident. Congrats on the launch, and Sachin Sridhar!

Y Combinator

15,048 views • 1 year ago

Embedder (Embedder) is building the Cursor for embedded: upload data sheets, chat with agents, and generate production-level driver code in seconds. Congrats on the launch, Ethan Gibbs and Bob Wei!
2:13

Embedder (Embedder) is building the Cursor for embedded: upload data sheets, chat with agents, and generate production-level driver code in seconds. Congrats on the launch, Ethan Gibbs and Bob Wei!

Y Combinator

55,557 views • 1 year ago

Solar (Solar AI) lets small teams collaborate with AI agents to build and deploy production apps. Combining the best parts of Figma + Cursor, Solar has shipped production apps with ML backends, gorgeous UIs, and automated QA — from concept to deployment in under an hour. Congrats on the launch, jack mcoxford, Vivek Hazari, @dandipietro1, and Ziray Hao!
1:35

Solar (Solar AI) lets small teams collaborate with AI agents to build and deploy production apps. Combining the best parts of Figma + Cursor, Solar has shipped production apps with ML backends, gorgeous UIs, and automated QA — from concept to deployment in under an hour. Congrats on the launch, jack mcoxford, Vivek Hazari, @dandipietro1, and Ziray Hao!

Y Combinator

25,639 views • 1 year ago

21st (21stdev) is the production layer for frontier AI agents. 1.4M developers already use their platform to build AI apps, 15K+ GitHub stars - now they're making it dead simple to deploy Claude Code, Codex, and other agents to production. Congrats on the launch, sergey bunas & serafim!
0:54

21st (21stdev) is the production layer for frontier AI agents. 1.4M developers already use their platform to build AI apps, 15K+ GitHub stars - now they're making it dead simple to deploy Claude Code, Codex, and other agents to production. Congrats on the launch, sergey bunas & serafim!

Y Combinator

33,400 views • 4 months ago

FlexAttention is a novel compiler-driven programming model that allows implementing the majority of attention variants in a few lines of idiomatic PyTorch code Boyuan Feng & Avik Chaudhuri show how many existing attention variants can be implemented via FlexAttention & that we achieve competitive performance compared to handwritten kernels. 📺 Watch the full video on YouTube:
0:57

FlexAttention is a novel compiler-driven programming model that allows implementing the majority of attention variants in a few lines of idiomatic PyTorch code Boyuan Feng & Avik Chaudhuri show how many existing attention variants can be implemented via FlexAttention & that we achieve competitive performance compared to handwritten kernels. 📺 Watch the full video on YouTube:

PyTorch

17,929 views • 1 year ago

Modelence (Modelence) is building the all-in-one TypeScript platform, making it easy for startups to ship production AI apps without boilerplate. Congrats on the launch, Aram Shatakhtsyan and Eduard Piliposyan!
1:43

Modelence (Modelence) is building the all-in-one TypeScript platform, making it easy for startups to ship production AI apps without boilerplate. Congrats on the launch, Aram Shatakhtsyan and Eduard Piliposyan!

Y Combinator

13,975 views • 1 year ago

in the lectures below, i hold your hand through low-level LLM systems engineering. it includes everything up to TODAY! 1) pytorch tensors 2) large matmul on cpu vs gpu 3) JAX (and why xAI uses it instead of pytorch) 4) raw cuda kernels and global threading indexing 5) triton design philosophy and softmax example 6) HIP kernels 7) mapping out the ENTIRE ecosystem + differences between CUDA and ROCm/HIP (BLAS, FFT, DNN) 8) cutlass and cute-dsl 9) pretraining, finetuning, rl, unsloth, axolotl, megatron-lm, deepspeed, nanogpt, nanochat 10) training vs inference, inference serving problems, throughput vs latency vs concurrency scaling, vllm, sglang, tensorrt-llm, tensorrt, llama.cpp, exllamav2, exllamav3, benchmark comparisons 11) projects/companies using llms to generate SOTA cuda/triton kernels 12) luminal inference 13) mojo/modular/max
2:23:25

in the lectures below, i hold your hand through low-level LLM systems engineering. it includes everything up to TODAY! 1) pytorch tensors 2) large matmul on cpu vs gpu 3) JAX (and why xAI uses it instead of pytorch) 4) raw cuda kernels and global threading indexing 5) triton design philosophy and softmax example 6) HIP kernels 7) mapping out the ENTIRE ecosystem + differences between CUDA and ROCm/HIP (BLAS, FFT, DNN) 8) cutlass and cute-dsl 9) pretraining, finetuning, rl, unsloth, axolotl, megatron-lm, deepspeed, nanogpt, nanochat 10) training vs inference, inference serving problems, throughput vs latency vs concurrency scaling, vllm, sglang, tensorrt-llm, tensorrt, llama.cpp, exllamav2, exllamav3, benchmark comparisons 11) projects/companies using llms to generate SOTA cuda/triton kernels 12) luminal inference 13) mojo/modular/max

Elliot Arledge

57,855 views • 9 months ago

YC S24's Polymet (YC S24) helps non-designers create production-ready designs and frontend code with AI. Explain what you want to build, or provide an image, and Polymet designs and implements it. Congrats on the launch, and Furkan Köseoğlu!
0:48

YC S24's Polymet (YC S24) helps non-designers create production-ready designs and frontend code with AI. Explain what you want to build, or provide an image, and Polymet designs and implements it. Congrats on the launch, and Furkan Köseoğlu!

Y Combinator

125,453 views • 1 year ago

cubic (cubic) is Cursor for code review. An AI-native platform that helps teams like n8n, and the Linux Foundation ship code 28% faster. Congrats on the launch, @NeuralPaul and Allis Yao!
1:06

cubic (cubic) is Cursor for code review. An AI-native platform that helps teams like n8n, and the Linux Foundation ship code 28% faster. Congrats on the launch, @NeuralPaul and Allis Yao!

Y Combinator

51,780 views • 1 year ago

Senso (YC W24) is an AI-Powered knowledge base that generates insights that improve the way Credit Unions, Banks, and traditional enterprises interact with customers. Congrats to Saroop Bharwani and Tom on the Senso launch!
0:54

Senso (YC W24) is an AI-Powered knowledge base that generates insights that improve the way Credit Unions, Banks, and traditional enterprises interact with customers. Congrats to Saroop Bharwani and Tom on the Senso launch!

Y Combinator

13,712 views • 2 years ago

Deeptrace (Deeptrace) is an AI SRE and debugging companion that automatically investigates every production alert and surfaces a clear root cause, so your engineers spend less time firefighting and more time building. Congrats on the launch, srinth & andy lee!
1:35

Deeptrace (Deeptrace) is an AI SRE and debugging companion that automatically investigates every production alert and surfaces a clear root cause, so your engineers spend less time firefighting and more time building. Congrats on the launch, srinth & andy lee!

Y Combinator

67,975 views • 9 months ago

Where others promise responsiveness, we deliver it with Electron - repeatedly. With reliable rockets, a steady production line, launch sites we control, and an experienced team that moves quickly, we deliver every piece that matters for responding fast to the nation's most important missions.
2:44

Where others promise responsiveness, we deliver it with Electron - repeatedly. With reliable rockets, a steady production line, launch sites we control, and an experienced team that moves quickly, we deliver every piece that matters for responding fast to the nation's most important missions.

Rocket Lab

81,386 views • 1 month ago

If AI can code 100x faster, why aren't you shipping 100x faster? Because AI code is not production-ready code, and definitely not code where you understand and can vouch for every line Introducing the Command Center alpha. Support our Product Hunt launch!
0:22

If AI can code 100x faster, why aren't you shipping 100x faster? Because AI code is not production-ready code, and definitely not code where you understand and can vouch for every line Introducing the Command Center alpha. Support our Product Hunt launch!

Jimmy Koppel

72,478 views • 9 months ago

“Initial production is always very slow and follows an S-curve. The speed of the production ramp is inversely proportionate to how many new parts and steps there are. For Cybercab and Optimus, almost everything is new, so the early production rate will be agonizingly slow, but eventually end up being insanely fast.” —Elon Musk
1:20

“Initial production is always very slow and follows an S-curve. The speed of the production ramp is inversely proportionate to how many new parts and steps there are. For Cybercab and Optimus, almost everything is new, so the early production rate will be agonizingly slow, but eventually end up being insanely fast.” —Elon Musk

James Stephenson

125,972 views • 1 month ago

Introducing Factory Describe your idea in one sentence and Factory builds the first draft of the production for you. It plans the scenes, chooses the models, generates the assets, and wires everything together on an editable infinite canvas.
1:08

Introducing Factory Describe your idea in one sentence and Factory builds the first draft of the production for you. It plans the scenes, chooses the models, generates the assets, and wires everything together on an editable infinite canvas.

Intelligent Internet

17,520 views • 3 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
0:22

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 🔭

310,260 views • 23 days ago