RT @alvinkcheung: And we just won the outstanding paper award at the MLArchSys workshop this year! Congrats Charles, @sahilb17 and @Sophia_….

RT @alvinkcheung: In this work, we show how to use off-the-shelf LLMs to generate code for accelerators. This is interesting as accelerator….

"Two kayakers paddling on a glowing lake under a breathtaking Milky Way sky, framed by dark forested mountains.". Create images and videos in seconds with Grok Imagine.

📝 Paper: 💻 Code: 📖 Blog: w/ Charles Hong, @alvinkcheung & Sophia Shao.4/4.

How?.✅ LLM-guided beam search.✅ A domain-specific “menu” of optimizations.✅ Tricks to explore better.✅ Performance feedback at every step.3/4.

We target Gemmini (the accelerator, not the model 😉), and achieve big speedups over hand-tuned baselines—using LLMs smartly! 2/4.

💥𝗖𝗮𝗻 𝗟𝗟𝗠𝘀 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗲 𝗹𝗼𝘄-𝗹𝗲𝘃𝗲𝗹 𝗮𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗼𝗿 𝗰𝗼𝗱𝗲?.Yes! (With some help.) In our paper Autocomp: LLM-Driven Code Optimization for Tensor Accelerators, we show how LLMs can optimize super low-resource accelerator code. 🧵 1/4.

RT @simonguozirui: LLMs for GPU kernel🌽generation have been getting Pop🍿ular since our preview last Dec; excited to announce 📢 our full pap….

We’re excited to present our work on Verified Code Transpilation using LLMs at #NeurIPS2024 today! 🤖🔍 Come visit our poster in the East Exhibit Hall A-C #2904 from 11AM-2PM.

RT @alvinkcheung: 📢 Please RT! 📢 Our group has a 1-2 yr postdoc opening starting this fall on using ML for code compilation and generation….

RT @ShadajL: Katara, our system for automatically synthesizing CRDT designs from sequential data types (, is now op….

RT @alvinkcheung: any suggestions for a parser that parses C, Java, Python, and Ruby into a common intermediate representation (like LLVM b….

RT @iamsrk: Sometimes we don’t land or arrive at the destination we want to. The important thing is we took off and had the Hope and Belief….