The team's approach: Introducing Self-Distilled Sparse Drafters (SD²), a novel methodology that leverages self-data distillation and fine-grained weight sparsity to produce highly efficient and well-aligned draft models.
CODE
-

SD² Enhances Draft Token Acceptance Reducing MACs
By
–
SD² systematically enhances draft token acceptance rates while significantly reducing Multiply-Accumulate operations (MACs), even in the Universal Assisted Generation (UAG) setting, where draft and target models originate from different model families.
-

Hugging Face Science Team Showcases Year of AI Contributions
By
–
The @huggingface Science team now has an org on the hub and it's really cool to look back at all their contributions of the past 12 months! Fineweb, LeRobot, TheStack, StarCoder2, SmolLM2, Big Science, LLMleaderboard, TRL, Nanotron, or some of the courses that have been followed
-

SmolVLA: Accessible Vision-Language AI for Consumer Robot Projects
By
–
SmolVLA: Affordable & Efficient VLA Robotics on Consumer GPUs Want to build a robot that sees, understands, and acts, without needing a PhD or a fat wallet?
Meet SmolVLA: the most accessible way to bring vision-language intelligence to your robotics projects. See how it -
MongoDB Text-to-MQL with LangChain: Natural Language Queries
By
–
@MongoDB supports Text-to-MQL with LangChain, allowing you to convert natural language queries directly into MongoDB queries. This makes it easier to build apps and agents that interact with your data seamlessly. Read the blog: https://
mongodb.com/blog/post/prod
uct-release-announcements/introducing-text-to-mql-langchain-query-mongodb-using-natural-language?utm_campaign=langchain_text_to_mql&utm_source=mongodb&utm_medium=blog&utm_term=langchain
… See the tutorial: -

Building an Agent with Authentication and Payment Processing
By
–
New Video: How to Build an Agent with Auth and Payments AI Services often track LLM requests via credits. We just released a video showing how to build a LangGraph.js app that does exactly this! It has:
– Stripe for payments
– Supabase for auth & credits
– LangGraph -
Importing Design Systems into Magicpath AI
By
–
Here's how easy it is to import and use any design system into @Magicpathai.
— Pietro Schirano (@skirano) 30 juin 2025
In this video we are importing one from tweakcn.
The generation follows the system so well. pic.twitter.com/nzqfIHNqVpHere's how easy it is to import and use any design system into @Magicpathai
. In this video we are importing one from tweakcn. The generation follows the system so well. -

nanoGPT: From Teaching Tool to Self-Improvement Benchmark
By
–
Love this project: nanoGPT -> recursive self-improvement benchmark. Good old nanoGPT keeps on giving and surprising 🙂 – First I wrote it as a small little repo to teach people the basics of training GPTs.
– Then it became a target and baseline for my port to direct C/CUDA -
Technical discussion on AI agent modes and MCP interaction
By
–
Hein ? Mais n’importe quoi c’est l’agent mode qui appelle le MCP. Et d’où ça fait exploser les tokens ? Faut vraiment éviter de dire des conneries avec autant d’aplomb.
-
Cursor Launches AI Agents Platform for Developers
By
–
Give it a try at http://
cursor.com/agents! And read the full announcement here: https://
cursor.com/blog/agent-web