Ke Yang

We (w/ Shansan Gong, Ruixiang ZHANG, Huangjie Zheng, Jiatao Gu, Navdeep Jaitly, Lingpeng Kong) released a family of 7B diffusion language models, DiffuCoder, that specializes on code generation, with a focus on understanding and improving masked diffusion models. A core analysis of DiffuCoder is the autoregressiveness (AR-ness) score, a novel metric that quantifies the causal patterns in decoding, revealing how diffusion models break from strict left-to-right generation for more flexible, non-linear code planning. Recent advances in autoregressive (AR) models dominate code generation, but diffusion-based LLMs (dLLMs) like DiffuCoder offer a promising alternative, especially for complex programming tasks. DiffuCoder explores how these models decode differently—showing less global AR-ness in code tasks compared to math—and how temperature affects both token selection and generation order, unlike traditional AR models. We also introduce coupled-GRPO, a post-training RL method with a coupled-sampling scheme, to reduce performance drops during accelerated decoding, boosting parallelism and efficiency. We use a self-improvement pipeline that leverages AR-ness analysis, coupled-GRPO optimization, and evaluation on benchmarks like AceCode-89k to refine decoding strategies. This approach enables DiffuCoder to navigate diverse code generation pathways and enhance performance with modest computational overhead. Looking ahead, we aim to further leverage Reinforcement Learning to steer code generation through these decoding patterns, with the discrete nature of AR-ness scores providing a foundation for search-based strategies—ideal for the sparse rewards of optimizing complex code structures. Check out our full paper and code for a deeper dive! Paper: https://lnkd.in/gVWU3BDJ Code: https://lnkd.in/gmXTZ_6n Models: https://lnkd.in/gTcKCDr9 #MachineLearning #AI #CodeGeneration #DiffusionModels #NLP

220

5 Comments

We developed an open-source AI tool to design concrete mixes that are stronger, more sustainable, and ready to build with faster—speeding up construction while reducing environmental impact. https://lnkd.in/gPCk8tCM But the impact of this AI tool is not just hypothetical! Amrize used Meta’s AI-based technologies to design a new low-carbon mix, and successfully deployed it in an at-scale slab-on-grade application at Meta's new data center in Rosemont, MN. Compared to the legacy mix, this new AI-designed mix is: 🦁 Stronger ⏱ Faster 🍃 Lower carbon ⏱ ️The ideal set time All this was achieved without needing any new materials, nor special equipment. Best of all, the AI is open-sourced. https://lnkd.in/g2KA7KZW This work was featured in a Meta engineering blog article published today! https://lnkd.in/gBU9HY8H

98

9 Comments

𝗧𝗵𝗲 𝗯𝗲𝘀𝘁 𝗔𝗜 𝗰𝗼𝗱𝗶𝗻𝗴 𝗹𝗲𝘀𝘀𝗼𝗻 𝗰𝗼𝘀𝘁 𝗼𝘂𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿 $𝟲𝟬𝟬 𝗮𝗻𝗱 𝗮 𝗺𝗮𝗿𝗿𝗶𝗮𝗴𝗲 𝗮𝗿𝗴𝘂𝗺𝗲𝗻𝘁. Our VP of Engineering, Xiaofan(James) Luan, was supposed to buy his wife a Dior bag for their anniversary.  Instead, he bought three Claude Code subscriptions and spent the holiday trying to cross-compile 2 million lines of C++. Every fix on one platform broke two others.  $600 later, the only output was "git reset --hard" — and a very cold dinner table.😂 "Make it compile on Windows" is a trap. The real goal was "compile everywhere without hacks" — no AI is going to figure that out for you at 2 am. What worked: constraints before code, review tests not code, bottom-up, one layer at a time. Same task, two days. Then he ran six parallel Claude sessions across three machines with git worktree. The bottleneck stopped being intelligence and started being how fast one person can alt-tab. AI solves exactly the problem you give it. Engineering is in knowing which one to give. His wife is still waiting for that bag. Full story: https://lnkd.in/gtsW_Wvk ——— Follow Milvus, created by Zilliz, for everything related to unstructured data

9

Search relevance is the backbone of any RAG system. If the retrieval is bad, the answers are bad — no matter how good the LLM is. We just published a deep dive on how we approach this at Dropbox Dash, and I think the core idea is worth sharing because it applies well beyond our product. The challenge: Dash searches across millions (sometimes billions) of enterprise documents to find the handful that matter for a given query. A relevance model ranks those results, and that model needs training data — lots of it. Human labeling is the gold standard, but it's expensive, slow, and can't touch sensitive customer data. Pure LLM labeling is cheap and fast, but uncalibrated LLMs make systematic mistakes. Our approach: use humans to teach the LLM, then use the LLM to scale. A small set of human-labeled examples calibrates the LLM's relevance judgments. Once the LLM meets quality thresholds, it generates hundreds of thousands of labels that train the production ranking model. The result is a 100x amplification of human labeling effort. A few things that made this work in practice: We focused evaluation on the hardest cases — documents users clicked that the LLM rated low, and documents users skipped that the LLM rated high. That's where the learning happens. We gave the LLM tools to research context before judging. Enterprise search is full of internal acronyms and jargon (at Dropbox, "diet sprite" is a performance management tool, not a soft drink). Without context, even strong models get these wrong consistently. We used DSPy (Community) to programmatically optimize prompts against human judgments, which turned prompt tuning from an art into something measurable and repeatable. The pattern — small human reference set, calibrated LLM evaluation, scaled labeling, continuous monitoring — is generalizable to any domain where you need reliable judgments at scale. Full our post on the Dropbox tech blog: https://lnkd.in/gB_5TECm

68

What I'm Reading This Week 📚 ⚡ Diffusion Coming to LLMs https://lnkd.in/gMrhRsvj Lots of announcements at Google IO this week. One that caught my eye is Gemini's new Diffusion model—applying image diffusion algorithms to text generation for faster speeds. This could be a game changer for latency-sensitive applications like IDE auto-completion and real-time code editing. 🔄 Real-Time Collaboration Architecture https://lnkd.in/ghwrBAxT Matthew Weidner delivers a brilliant deep dive into synchronizing state between central servers and multiple clients. If you've ever wondered how Google Docs manages concurrent edits without chaos, this is your answer. His exploration of operational transformation and server reconciliation patterns gives me new appreciation for the complexity behind seamless collaboration. 📈 Economic Reality Check https://lnkd.in/gfTf2TG9 Thomas Klitgaard from the NY Fed provides a down-to-earth explanation of why the US consistently runs trade deficits. While the current administration's trade war rhetoric continues, one thing is clear: rebalancing trade won't happen without painful adjustments for American consumers. The macroeconomic forces at play are more complex than political soundbites suggest. 🎯 Algebraic Effects Explained https://lnkd.in/g8zeFzWX A fascinating introduction to algebraic effects—essentially resumable exceptions that enable elegant control flow. While this cutting-edge programming language feature might be hard to wrap your head around, the Ante team does an excellent job explaining when and why you'd need such abstractions. 🤖 Google's Coding Agent https://jules.google/ Google launches Jules, their take on cloud-hosted agentic coding loops. Between AI-powered IDEs, local agents like Claude Code, and these cloud solutions, we're clearly still figuring out the right form factor for agentic programming. The overloaded "Codex" branding from OpenAI isn't helping clarity either, but the underlying trend toward autonomous code generation is undeniable. What's your take on the current state of AI coding assistants? Are you finding them more helpful or distracting in your development workflow? #TechArchitecture #AICodeGeneration #DistributedSystems #EconomicPolicy #FutureOfProgramming #AgenticDevelopment #DeveloperTools

15

This is one of those things that just makes sense. The Linux Foundation (and by extension the PyTorch Foundation too) is a great home to nurture open-source projects.

4

** Big update on the roadmap for Meta's MTIA roadmap (our Meta-designed AI chips) ** We just shared today a big update on the MTIA (Meta Training & Inference Accelerator) roadmap. The main takeaway is we're accelerating the whole set of programs to get rapid iterations on chips (given how fast the model needs are evolving). 👉 Overall blog is here: https://lnkd.in/gaHuxUfg For the networking aficionados, you can see a whole lot of networking that we've been working on inside MTIA - we're excited to finally be able talk about it! 👉 The scale-up and scale-out networks we've already talked about. It's sort of buried in there, but we are already running distributed training clusters with MTIA 300 for core ranking & recommendation (even as we focus future generations on Inference). 👉 The communication library (HCCL) that is purpose-built to take advantage of networking-specific capabilities we've built into the chips to offload the main compute. 👉 The in-built support for fused compute/communication kernels to minimize latency (key for both training and inference workloads). Super-proud of the whole MTIA team!

143

4 Comments