Today we’re releasing Meta LLM Compiler, a family of models built on Meta Code Llama with additional code optimization and compiler capabilities. The models achieve state-of-the-art results on optimization of code size and disassembly tasks. Hugging Face repo ➡️ https://go.fb.me/tdd3dw Research paper ➡️ https://go.fb.me/hvfnam LLM Compiler can emulate the compiler, predict optimal passes for code size, and disassemble code. It can be fine-tuned for new optimizations and compiler tasks. This work shows that AI is learning to optimize code and can assist compiler experts in identifying opportunities to apply optimizations. We believe this work could have an impact ranging from use in optimization for individual developer environments to inclusion in a compiler such as LLVM. We’re releasing LLM Compiler 7B & 13B models under a permissive license for both research and commercial use in the hopes of making it easier for developers and researchers alike to leverage this in their work and carry forward new research in this highly impactful space.
The question everybody is asking can we use it to make commercial products and is it MIT open source
What a horrible idea..... I have never heard anyone ask for a compiler that gives random results and is only 20% accurate.
This is an amazing step for developers.
This is wild! Soon AI will simply write machine code with no intermediate required (that's for humans). I imagine you can also go the other way, decompiling plausible Java or Python from the machine code, complete with usage of high level libraries (Also, 1000th like)
Meta's LLM Compiler sounds like it's flipping the script on code optimization! Emulating compilers and predicting optimal passes—impressive stuff. Who's diving into these models first-hand? Any early insights on how they're shaping up against traditional compiler techniques?
Exciting news from Meta! The release of the AI at Meta Meta LLM compiler is a significant step forward in the field of AI-driven code optimization. It's impressive to see how this model, built on Code Llama, can enhance compiler reasoning and code optimization tasks. The potential for improving code size and disassembly is particularly intriguing. Looking forward to seeing how developers and researchers will leverage this technology to push the boundaries of software optimization and compiler design. Great work on making this available for both research and commercial use!
AI at Meta, You never disappoint! This is going to change the software engineering for lower level code generation. It could improve performance and memory management. A meta-compiler for compilers.
Amazing work! We would be glad if you consider publishing a free open access video on our platform about your research.
Does this mean that it is the end of the CUDA moat for Nvidia and that LLMs can be used to optimize competing hardware platforms from AMD/Groq/Amazon and others to run LLM training and particularly inference in a more cost efficient manner soon?
Automatic Formal Verification, Compiler Optimization
2moDear AI at Meta , this is wonderful, something that the research community has been anticipating for ages. You may want to consider supplementing this with a formal translation validator to obtain a reliable end-to-end system. A recent validator from LLVM to x86 is described in https://dl.acm.org/doi/abs/10.1145/3649863 (sorry for advertising my own work, but this is exciting stuff and I couldn't resist). I will be very happy to help with integrating our validator if there is interest.Volker Seeker Dejan Grubisic Hugh Leather (tagging the authors I could find in my network)