🔥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

(6/n) Download #Xwen #LLM now! BF16 and all kinds of GGUF models are provided. HuggingFace🤗: huggingface.co/collections/sh…

由清华大学等高校成员组成的 Xwen Team 开源的 Xwen 模型！ 基于Qwen base模型训练而成，包含Xwen-72B-Chat与Xwen-7B-Chat两种大小，表现不错，有关注本地部署小模型的可以关注一下

Xwen 🔥 a series of open models based on Qwen2.5 models, developed by a brilliant research team of PhD students from the Chinese community. huggingface.co/collections/sh… ✨ 7B/72B ✨ Apache 2.0 ✨ Xwen-72B-Chat outperformed DeepSeek V3 on Arena Hard Auto

lmarena.ai (formerly lmsys.org) OpenAI Hey lmarena.ai (formerly lmsys.org) team, we'd love to see our open-sourced model, Xwen-72B-Chat, included in the Chatbot Arena! 🥹 It supports both English and Chinese, with strong performance across multiple benchmarks. We have sufficient inference compute for API integration. We've sent several

lmarena.ai (formerly lmsys.org) OpenAI Xwen-72B-Chat achieves 86.1 on Arena-Hard-Auto, surpassing DeepSeek-V3 (671B) with only ~1/10th the parameters! 😆We believe Xwen-72B-Chat can perform well on Chatbot Arena, possibly ranking in the top 10.

🔥 Welcome to try our EasyR1!

Microsoft releases ART Anonymous Region Transformer for Variable Multi-Layer Transparent Image Generation

In modern RLHF frameworks, we see many "batch size" configs. They are designed for maximizing the GPU utilization. However, these configs also confuse users who are not familiar with systems. To provide a clear view for those who want to train their reasoning models, we explain

Today, we release QwQ-32B, our new reasoning model with only 32 billion parameters that rivals cutting-edge reasoning model, e.g., DeepSeek-R1. Blog: qwenlm.github.io/blog/qwq-32b HF: huggingface.co/Qwen/QwQ-32B ModelScope: modelscope.cn/models/Qwen/Qw… Demo: huggingface.co/spaces/Qwen/Qw… Qwen Chat:

🚀 Recently, I've been focusing on RL for LLM, and I'm excited to introduce QwQ-32B—the best open-source reasoning model under 100B scale. RL indeed holds some fascinating yet unexplored mysteries. You're all welcome to continue building more interesting things based on Qwen!

LLaMA-Factory now supports the fine-tuning (Full/LoRA/QLoRA) of the QwQ-32B model. Time to release the power of reasoning models on your personal data 💥

🚨Ever wonder why diffusion models generate nonsensical text? Our latest study at #ICLR2025 uncovers "Local Generation Bias"—a hidden training bias causing textual hallucinations! 🧠 Key finding: Diffusion models independently generate symbols locally without global context.

Cooragent by our Tsinghua LeapLab LeapLab@THU : Open-source multi-agent collaboration framework! 🚀 Tell it to "build an AI intelligence secretary" → auto-scans, curates updates, delivers daily reports. MIT Licensed | Dev-friendly. Try ⬇️ github.com/LeapLabTHU/coo… #AgenticAI

Does RL Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? Our new paper investigate the question and has sparked active discussions. In video, freq Q&A starts at 1:28, covering common questions on pass@k, the takeaway and etc. see limit-of-RLVR.github.io

Introducing Qwen3! We release and open-weight Qwen3, our latest large language models, including 2 MoE models and 6 dense models, ranging from 0.6B to 235B. Our flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark evaluations of coding, math, general

❄️Introducing Absolute Zero Reasoner: Our reasoner learns to both propose tasks that maximize learnability and improve reasoning by solving them, entirely through self-play—with no external data! It overall outperforms other "zero" models in math & coding domains. 🧵 1/

🔥 Excited to introduce our work: Absolute Zero—training reasoning LLMs with NO DATA via RLVR! 🚀 A new “Absolute Zero” paradigm: models learn to propose and solve tasks, evolving through self‑play. 🏆 AZ Reasoner: SoTA overall performance in math & coding with no human data.

🔥 Checkout our new survey on scaffolded language models that learn beyond parametric update! 🚀Applications could be in software agents like Codex to continuously learn and adapt to an user’s needs