🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

You can now run Qwen3-235B-A22B-2507 with our Dynamic 2-bit GGUFs! The full 250GB model gets reduced to just 88GB (-65% size). Achieve >5 tokens/s on 89GB unified memory or 80GB RAM + 8GB VRAM. GGUFs: huggingface.co/unsloth/Qwen3-…

Fascinating work bridging cognitive science + NLP! PrefPalette decomposes preferences into interpretable attributes (humor, empathy, formality) with dynamic weighting. 46.6% better than GPT-4o with explainability. This opens new directions in alignment and personalization.

🚨 The era of infinite internet data is ending, So we ask: 👉 What’s the right generative modelling objective when data—not compute—is the bottleneck? TL;DR: ▶️Compute-constrained? Train Autoregressive models ▶️Data-constrained? Train Diffusion models Get ready for 🤿 1/n

🚀 We’re excited to introduce Qwen3-235B-A22B-Thinking-2507 — our most advanced reasoning model yet! Over the past 3 months, we’ve significantly scaled and enhanced the thinking capability of Qwen3, achieving: ✅ Improved performance in logical reasoning, math, science & coding

Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀 📄 huggingface.co/papers/2507.18…

Open weight models + open science is the way to go and will win in the long term. Congrats to the team for this awesome release!

StanfordのMixture of Experts (MoE)の講義は超おすすめ（5回ぐらいヘビロテした）/ Stanford CS336 Language Modeling from Scratch | Spring 2025 | Lecture 4:... youtu.be/LPv1KfUXLCo?si… via YouTube

📢🧵 Thread | Celebrating Ibero-American NLP at #ACL2025! 🇪🇸🇵🇹🇲🇽🇨🇴🇧🇷🇦🇷 ... 🌎 The SomosNLP @ACL  community is showing up strong at ACL 2025 ! We want to highlight all the papers from our vibrant community 👇 #NLProc #NLP #IberoAmerica

What if the future of AI was fundamentally inequitable for an entire continent? This is the critical question we pose in our latest work (#ACL2025). We undertake a comprehensive empirical evaluation of leading LLMs on Sahara, our extensive benchmark that we collect using mostly

Today, we are presenting 3 papers #ACL2025: 1) Injongo: A multicultural intent detection for African lang. Room 1.15-16 (Multilingualism)@ 14:00 2) BRIGHTER: Emotion classification(Nedjma Ousidhoum نجمة أوسيدهم @ACL2025) Hall B (resources) @ 14:00 3) Global MMLU (Shivalika Singh Sara Hooker) poster @ 10:30

It’s day 4 at ACL 2025 and we’re excited that MURI will be showcased today! 🎉 Work led by: Abdullatif Köksal Marion Thaler  Ayyoob Imani Ahmet Üstün Anna Korhonen Peiqin Lin  📜arxiv.org/abs/2409.12958

6th AfricaNLP Workshop ACL 2025 #ACL2025NLP #ACL2025 @masakhane Opening remark by Shamsuddeen Hassan Muhammad, PhD

Great talk by Malihe Alikhani on AI sycophancy 🚀

La inclusión digital no es un lujo, es una necesidad.  indigital.surcooaxaca.org Cerremos la brecha.

🦥 Qwen3-Coder-Flash: Qwen3-Coder-30B-A3B-Instruct 💚 Just lightning-fast, accurate code generation. ✅ Native 256K context (supports up to 1M tokens with YaRN) ✅ Optimized for platforms like Qwen Code, Cline, Roo Code, Kilo Code, etc. ✅ Seamless function calling & agent

Pleased to see this new Alignment Project, where I serve as an expert advisor, launched by the UK's AI Security Institute and supported by the Canadian AI Safety Institute and many others. I encourage my fellow researchers to apply for funding, compute and support from int’l experts.

Long-form AI reading is back and we’ve just dropped the ultimate summer read. Inspired by the likes of Stripe Press, we’re proud to announce the first book from HF Press: a carefully crafted, book-length PDF edition of the Ultra-Scale Playbook. Over 200 dense pages to learn the

AI Industry Made $57 Billion Mistake and No One’s Talking About It. While GPT-5 headlines kept you distracted... NVIDIA quietly released a bold claim: → Small Language Models (SLMs) are the future of AI agents Cheaper, faster and just as capable for 80% of real-world tasks.

So, I did some coding this week... - Qwen3 Coder Flash (30B-A3B) - Mixture-of-Experts setup with 128 experts, 8 active per token - In pure PyTorch (optimized for human readability) - in a standalone Jupyter notebook - Runs on a single A100

Has anyone actually tried this 27M param model that beats o3?