@profcelsofontes @ZenMuxAI The model is open-source, but ZenMux is a commercial platform, so you may need to pay for access. The ZenMux platform offers two payment options: a subscription plan and a pay-as-you-go option.
English
Ant Open Source
191 posts
Ant Open Source
@ant_oss
All things open source at Ant Group. We aim to bring high caliber infrastructure FinTech OSS to the community.
Katılım Mart 2024
93 Takip Edilen1.6K Takipçiler
⚡️ 892 tokens/s — our 100B diffusion LLM, LLaDA2.1-flash, is now live on @ZenMuxAI!
With Token Editing, LLaDA 2.1 goes from research breakthrough to production-ready speed. Diffusion models just got real.
Try it via API or Chat 👇 zenmux.ai/inclusionai/ll… #LLaDA #ZenMux #AI #dLLM
ZenMux@ZenMuxAI
⚡️New on ZenMux: LLaDA2.1-flash 100B diffusion LLM from @TheInclusionAI . → Error-correcting editable generation → Speed Mode: ultra-fast inference → Quality Mode: competitive performance → RL tailored for 100B-scale dLLM 🔗 zenmux.ai/inclusionai/ll… 🔗 huggingface.co/inclusionAI/LL…
English
🚀 AReaL v1.0 is here! Evolve your 🦞#OpenClaw agents (or any agent) with RL—zero changes to agent required.
We add a transparent proxy that shadows your agent's base_url, capturing all agent–LLM interactions for RL training. Just swap the URL and go.
✅ One-click agentic RL for any existing agent—no modifications needed
✅ Full #opencode recipe: training code, data, infra, and models all open
✅ Archon Engine: PyTorch-native 5D parallelism via pure PyTorch; uv sync and go (zero manual compilation)
✅ torch.compile by default: instant 10% performance boost out of the box
📊 SOTA on tau2Bench: 73.0% pass@1 (Airline) / 98.3% (Telecom)
#opensource #inclusionAI #RL
📄 Paper: arxiv.org/abs/2601.22607
🐙 GitHub: github.com/inclusionAI/AR…
Train your OpenClaw agent: github.com/inclusionAI/AR…
English
Meet Ling-2.5-1T! 📊1T params (63B active), 29T pre-training corpus & 1M context #LLMs #OpenSource
As the latest flagship instant model in the @AntLingAGI family, Ling-2.5-1T delivers comprehensive upgrades across model architecture, token efficiency, and preference alignment.
🤗 Hugging Face: huggingface.co/inclusionAI/Li…
📷 ModelScope: modelscope.cn/models/inclusi…
Ant Ling@AntLingAGI
Introducing Ling-2.5-1T, the new flagship with 1T params (63B active), 29T pre-training corpus & 1M context - Efficiency: Matches reasoning of frontier thinking models using only 1/4 tokens via composite rewards. - Alignment: Bidirectional RL + Agent-based verification for precise instruction following. - Agentic: Native Agentic RL training; SOTA on BFCL-V4 & ready for Claude Code/OpenCode. It leads in tool use & alignment, balancing speed with intelligence. ⚖️
English
🔍MLLMs are nearsighted: zooming helps but kills speed. What if we teach them to "see" fine details without zooming?
Introducing Region-to-Image Distillation(R2I): we train models to internalize "zooming".
🏆 ZwZ-8B achieves SOTA performance on fine-grained perception with zero tool calls.
📊 Plus ZoomBench: human-AI collaboration construction, dual-view "zooming gap" metric
Dive into our joint efforts:
📑Paper: huggingface.co/papers/2602.11…
🔗Code: github.com/inclusionAI/Zo…
⚙️ Model&Data: huggingface.co/collections/in…
#OpenSource #MLLM #LLMs #inclusionAI
English
Ring-1T-2.5 is released📣🚀with high efficiency in planning and multi-step tool collaboration.
🔧Try out and explore how this 1:7 MLA + Lightning Linear Attention boosts reasoning speed and exploration! #OpenSource #LRM
🤗 Hugging Face: huggingface.co/inclusionAI/Ri…
📷 ModelScope: modelscope.cn/models/inclusi…
Ant Ling@AntLingAGI
🚀 Unveiling Ring-1T-2.5 The first hybrid linear-architecture 1T thinking model. -Efficient: Hybrid linear breakthrough (10x lower memory) -Gold Tier: IMO25 (35/42) & CMO25 (105/126) -Agentic: Natively with Claude Code & OpenClaw -Open SOTA: IMOAnswerBench，GAIA2-search & more!
English
🚀 The official version of Ming-flash-omni 2.0 is released!
📷 Dive in: github.com/inclusionAI/Mi…
#opensource #LLM #inclusionAI
Ant Ling@AntLingAGI
One for all, and all for one 🧧 Introducing Ming-flash-omni-2.0: A specialist in every domain, unified as a capable generalist. A gift from Ling =) - Unified Acoustic Synthesis: Speech, audio, and music combined for unbounded creativity; - "Seeing" to "Knowing": Moving beyond input to true deep semantic understanding; - Native Visual Fusion: Seamless generation, editing, and segmentation;
English
@aiktp_com @_akhaliq Hi, thanks for your interest! For the early results, please check out our earlier paper: huggingface.co/papers/2512.15…
English
LLaDA2.1
Speeding Up Text Diffusion via Token Editing
huggingface.co/papers/2602.08…
English
Ant Open Source retweetledi
Congrats to @ant_oss on releasing LLaDA 2.1, a 100B discrete diffusion LLM that breaks the speed–quality tradeoff. Day-0 support is live in SGLang!
⚡ Unified decoding: fast parallel generation & on-the-fly token correction
🎛️ User-controllable modes: ultra-fast decoding & high-fidelity reasoning
🧩 Mask-to-Token + Token-to-Token editing under one framework
🧠 Trained with large-scale block-level RL for SOTA efficiency and performance
Related PR: github.com/sgl-project/sg…
Ant Open Source@ant_oss
What if an LLM could EDIT its own tokens in real-time, not just generate them? 🤯 Introducing LLaDA2.1 — a diffusion model that breaks from autoregressive dominance. It drafts fast, then fixes its own mistakes on the fly with Token-to-Token editing. The result? 892 tokens/sec on a 100B model. 🔥 ⚡ 892 TPS on HumanEval+ (coding) ⚡ 801 TPS on BigCodeBench 🧠 Real-time self-correction via T2T editing ✅ @lmsysorg SGLang Day 0 support — production-ready now A "non-consensus" architecture now challenging the mainstream. Open-sourced TODAY. 👇 #LLaDA #TokenEditing #OpenSource #LLM #dLLM
English
Ready to experience the future? We're releasing two versions:
🔹 LLaDA2.1-Mini (16B) — Fast and efficient
🔹 LLaDA2.1-Flash (100B) — Maximum performance
Both ready to revolutionize software development, content creation, and beyond. 🚀
🤗 HuggingFace: huggingface.co/collections/in…
📖 Technical Report: huggingface.co/papers/2602.08…
💻 GitHub: github.com/inclusionAI/LL…
The era of editable, self-correcting LLMs starts now. Join us in redefining what's possible. ✍️
English
The results? Mind-blowing. 🤯
LLaDA2.1-Flash (100B) hits a peak speed of 892 tokens/second on complex coding tasks. This isn't just incremental — it's a leap forward in generation efficiency.
Across 33 rigorous benchmarks, LLaDA2.1 proves that diffusion models can challenge autoregressive dominance.
English
What if an LLM could EDIT its own tokens in real-time, not just generate them? 🤯
Introducing LLaDA2.1 — a diffusion model that breaks from autoregressive dominance. It drafts fast, then fixes its own mistakes on the fly with Token-to-Token editing.
The result? 892 tokens/sec on a 100B model. 🔥
⚡ 892 TPS on HumanEval+ (coding)
⚡ 801 TPS on BigCodeBench
🧠 Real-time self-correction via T2T editing
✅ @lmsysorg SGLang Day 0 support — production-ready now
A "non-consensus" architecture now challenging the mainstream. Open-sourced TODAY. 👇
#LLaDA #TokenEditing #OpenSource #LLM #dLLM
English
Happy to share more Kata use cases in agentic infra!
Kata Containers@katacontainers
This article, complete with @AntGroup use case, highlights how Kata Containers adds VM-level isolation to Kubernetes, helping cloud providers and enterprises run sensitive and AI workloads with greater security, without changing developer workflows. thestack.technology/kata-container…
English