💥疯了！Qwen自己长出了一颗真GPU，在权重里跑真实CUDA！
This is a dev post classified by AGK Intelligence as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
💥疯了！Qwen自己长出了一颗真GPU，在权重里跑真实CUDA！ 牛逼之处在于把一颗完整的 4-lane SIMT GPU 直接编译进了 Qwen3.5-2B 的权重里！ 45层全新编译层插在Qwen原有24层前面，用它自己的 Gated DeltaNet 状态当寄存器、程序ROM和全局内存。 最炸的点： 🔹参考解释器对标真实 NVIDIA Tesla T4，49/49 kernel完全一致 🔹对原模型语言能力几乎无影响（KL散度仅约7e-4） 🔹支持真实CUDA特性：warp shuffle、atomicAdd、__syncthreads、分支发散…… 🔹高级模式还能让Qwen自己写CUDA kernel 有哪些用处？ 1️⃣精确计算内嵌：LLM自己做加减乘除/矩阵运算，再也不用外部工具调用出错 2️⃣自包含Agent：模型内部就能跑并行计算，不依赖外部GPU/Python 3️⃣硬件级神经计算研究：证明DeltaNet可以当真实寄存器+内存用 4️⃣极致可验证：每一条launch都能token级复盘对比 5️⃣教育/演示神器：直接看见“模型在跑GPU”的全过程 GitHub：https://github.com/vixhal-baraiya/qwenGPU #Qwen #qwenGPU #CUDA #LLM #AIHardware #DeltaNet
Posted by SuSu_酥酥👅 (25.6k followers) 1 h ago · 17 likes · 1.5k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: qwengpu.
More dev work like this
- 💥惊了！中国国产1.3B开源模型OXR横扫全球三大文档解析榜单！精度、速度双SOTA！ — @NFT_Chen
- Meet Qwen3.8-27b-noslop-v2-i1-GGUF: a 27B parameter model fine-tuned for writing and… — @HuggingModels
- 🔥Kimi just released an end-to-end game development and creative solution built on Kimi… — @Chinazhidx
- Building agents that handle payments? Start with the plumbing. — @DanKornas
- 🤯Dario无敌了！Fable 5.5 灰测炸裂：纯代码一发入魂，直接干翻所有生图/视频模型！吉卜力级动漫短片诞生！灰测指南务必码住！ — @NFT_Chen
- 美团发布 LongCat-2.5 正式版，距离今年 9 月预览版上线已有一个多月。 — @0xLogicrw
- Meet Ministral 3 14B Instruct 2512, abliterated and GGUF-ready. This uncensored 14B… — @HuggingModels
- Made another tiny Claude Code mod: every GitHub PR link in the chat gets its status… — @adrianthedev
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.3k posts from 5k X accounts over the last 21 days, 9k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-02 11:07 UTC. Full method.