Media Summary: Speaker: Mengdi Wu Abstract We introduce Mirage, the first Talk by Mengdi Wu and Xinhao Cheng on Mirage. Mirage Persistent Kernel (MPK) is a compiler and runtime system that ... Compilers are caught in a tug-of-war between increasingly exotic architectures and instruction set extensions on one hand, and ...

A Multi Level Superoptimizer For - Detailed Analysis & Overview

Speaker: Mengdi Wu Abstract We introduce Mirage, the first Talk by Mengdi Wu and Xinhao Cheng on Mirage. Mirage Persistent Kernel (MPK) is a compiler and runtime system that ... Compilers are caught in a tug-of-war between increasingly exotic architectures and instruction set extensions on one hand, and ... This video took almost forever to make. Partly because of me learning Manim, partly because higher education nuked my ... In this video, I'll be talking about OpenRouter's new Fusion API, which claims to deliver Fable- In this video I am testing Qwen 3.6 27B MTP against Qwopus 3.6 27B v2 MTP. Let's see how the base Qwen model compares to ...

Xiaomi got a trillion-parameter model — one that trades blows with the frontier on coding benchmarks — to generate text at ... ... same thing to to do the same thing so I got curious about it and I wanted to try to uh write a Title: Unlocking the Power of Mixed-Precision Spatial Compute in the AMD Ryzen™ AI NPU Speakers: Gagandeep Singh, Kristof ... A 95% per-step accuracy sounds great — until you chain 10 steps together and your AI agent fails 40% of the time. This is the ... It's a software engineer's dream: A compiler that can take idiomatic high-

Photo Gallery

A Multi-Level Superoptimizer for Tensor Programs
OSDI '25 - Mirage: A Multi-Level Superoptimizer for Tensor Programs
Mirage (MPK): Compiling LLMs into Mega Kernels
Lecture 79 Mirage (MPK): Compiling LLMs into Mega Kernels
Archive: Superoptimizing LLVM
All about multistaging
FULLY FREE Unlimited API + OpenCode: MiniMax M3,Step 3.7 Flash,Nemotron 3 Ultra,GLM,Kimi!
Qwen 3.6 27B MTP vs Qwopus 3.6 27B v2 MTP - 16GB Local LLM setup
How to Run a Trillion-Parameter AI at 1,000 Tokens a Second
TIS 100 Superoptimization (Lightning Talk) — Daan van Berkel
SAFARI-EFCL Seminar: Unlocking the Power of Mixed-Precision Spatial Compute in the AMD Ryzen™ AI NPU
95% Accuracy Isn't Enough: Why Multi-Step AI Agents Fail
View Detailed Profile
A Multi-Level Superoptimizer for Tensor Programs

A Multi-Level Superoptimizer for Tensor Programs

https://egraphs.org/meeting/2025-10-16-mirage Speaker: Mengdi Wu Abstract We introduce Mirage, the first

OSDI '25 - Mirage: A Multi-Level Superoptimizer for Tensor Programs

OSDI '25 - Mirage: A Multi-Level Superoptimizer for Tensor Programs

Mirage:

Mirage (MPK): Compiling LLMs into Mega Kernels

Mirage (MPK): Compiling LLMs into Mega Kernels

Talk by Mengdi Wu and Xinhao Cheng on Mirage. Mirage Persistent Kernel (MPK) is a compiler and runtime system that ...

Lecture 79 Mirage (MPK): Compiling LLMs into Mega Kernels

Lecture 79 Mirage (MPK): Compiling LLMs into Mega Kernels

Talk by Mengdi Wu and Xinhao Cheng on Mirage. Mirage Persistent Kernel (MPK) is a compiler and runtime system that ...

Archive: Superoptimizing LLVM

Archive: Superoptimizing LLVM

Compilers are caught in a tug-of-war between increasingly exotic architectures and instruction set extensions on one hand, and ...

All about multistaging

All about multistaging

This video took almost forever to make. Partly because of me learning Manim, partly because higher education nuked my ...

FULLY FREE Unlimited API + OpenCode: MiniMax M3,Step 3.7 Flash,Nemotron 3 Ultra,GLM,Kimi!

FULLY FREE Unlimited API + OpenCode: MiniMax M3,Step 3.7 Flash,Nemotron 3 Ultra,GLM,Kimi!

In this video, I'll be talking about OpenRouter's new Fusion API, which claims to deliver Fable-

Qwen 3.6 27B MTP vs Qwopus 3.6 27B v2 MTP - 16GB Local LLM setup

Qwen 3.6 27B MTP vs Qwopus 3.6 27B v2 MTP - 16GB Local LLM setup

In this video I am testing Qwen 3.6 27B MTP against Qwopus 3.6 27B v2 MTP. Let's see how the base Qwen model compares to ...

How to Run a Trillion-Parameter AI at 1,000 Tokens a Second

How to Run a Trillion-Parameter AI at 1,000 Tokens a Second

Xiaomi got a trillion-parameter model — one that trades blows with the frontier on coding benchmarks — to generate text at ...

TIS 100 Superoptimization (Lightning Talk) — Daan van Berkel

TIS 100 Superoptimization (Lightning Talk) — Daan van Berkel

... same thing to to do the same thing so I got curious about it and I wanted to try to uh write a

SAFARI-EFCL Seminar: Unlocking the Power of Mixed-Precision Spatial Compute in the AMD Ryzen™ AI NPU

SAFARI-EFCL Seminar: Unlocking the Power of Mixed-Precision Spatial Compute in the AMD Ryzen™ AI NPU

Title: Unlocking the Power of Mixed-Precision Spatial Compute in the AMD Ryzen™ AI NPU Speakers: Gagandeep Singh, Kristof ...

95% Accuracy Isn't Enough: Why Multi-Step AI Agents Fail

95% Accuracy Isn't Enough: Why Multi-Step AI Agents Fail

A 95% per-step accuracy sounds great — until you chain 10 steps together and your AI agent fails 40% of the time. This is the ...

Compiler Optimization with Greta Yorsh

Compiler Optimization with Greta Yorsh

It's a software engineer's dream: A compiler that can take idiomatic high-