qwenpu
qwenpu is An LLM running on an NPU inside an LLM. Karpathy's stories260K writing stories on a vector unit compiled into Qwen3.5-4B's weights. - divagr18/QwenPU. It is ranked #1624 on the AI Radar, in Open & local models, first seen 4 days ago and shared in 2 posts (45.2k views).
GitHub - divagr18/QwenPU: An LLM running on an NPU inside an LLM. Karpathy's stories260K writing stories on a vector unit compiled into Qwen3.5-4B's weights. · GitHub Skip to content Navigation Menu Sign in Search / Sign in Sign up Appearance settings You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session. Dismiss alert {{ message }} Repository navigation QwenPU Public Notifications You must be signed in to change notification settings Fork 8 ( 8 ) Star 86 ( 86 ) You must…
What people said about qwenpu on X
🫨绝了啊！有人把完整CPU + 另一个LLM塞进Qwen3.5-4B权重里，还跑出了故事！ 有人用Qwen3.5-4B做了件疯狂的事：直接把完整32位RISC-V CPU编译进模型权重，再用它跑出另一个LLM（Karpathy的stories260K）。 全程在模型自己的forward pass里完成，零外部运行时、零工具调用，数学精度直接从55%干到97.5%+，还是bit-exact的。 核心亮点： 🔹CPU硬核嵌入：11层编译出来的RV32IM（+Zba+大数单元）直接塞在Qwen前面，权重全是编译器算出来的，没训练。 🔹RAM就是模型自己：用Gated…
— @NFT_Chen, 4 days ago · 136 likes · see the post
💥逆天！史诗级魔改！仅需24G内存跑通！Qwen3.8-27B全64层变纯线性注意力，262K长文秒杀级神器来了！ 原版是48层Gated DeltaNet + 16层全注意力的混合架构。这个自定义版本直接把那16层全注意力也转成Gated DeltaNet，做成100%纯GDN模型！ 核心亮点： 🔹全层线性注意力：彻底干掉KV Cache爆炸，显存几乎不随上下文增长（相关版本262K仅~55GB且不OOM） 🔹逐层蒸馏+LoRA(r32)恢复精度，不是粗暴替换 🔹专为书籍/论文/超长报告设计，英日双语，262K位置嵌入配置好 🔹4bit约17GB（24GB卡可上），bf16约54GB…
— @NFT_Chen, 4 days ago · 100 likes · see the post
Alternatives to qwenpu
- kimi-k3.1-pro — Releasing the model weights of Kimi K3.1 Pro
- unsloth.ai — Guide to running Mistral Large 4 GGUFs locally with Unsloth and llama.cpp.
- niklaslenz.com — Play the world in your browser. Every number, the exact command and the raw output
- qwen3.8-flash-next-gguf — Image-Text-to-Text model with 177B parameters. Download, explore files and run it on Hugging Face.
- project-maya — Plus a round of fixes, UI improvements and community PRs.
- twil-8b — Text Generation model with 9B parameters. Download, explore files and run it on Hugging Face.
qwenpu in numbers
- Rank on the AI Radar: #1624 of 10412
- Shared in 2 posts by 1 account: @NFT_Chen
- 45.2k views on those posts
- First seen 4 days ago, last shared 4 days ago
- Pricing seen by AGK Intelligence: open source
- Market: Open & local models
FAQ
What is qwenpu?
An LLM running on an NPU inside an LLM. Karpathy's stories260K writing stories on a vector unit compiled into Qwen3.5-4B's weights. - divagr18/QwenPU It was first shared on X 4 days ago and is ranked #1624 on the AI Radar.
Is qwenpu free?
It is open source.
Who shared qwenpu?
1 account on X, including @NFT_Chen, in 2 posts totalling 45.2k views.
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. 65.5k posts from 5k X accounts over the last 14 days, 10.4k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 16:50 UTC. Full method.