Open-source community releases an uncensored fine-tuned model based on DeepSeek-V4-Pro and Qwen3.5 9B
An uncensored fine-tuned model based on a hybrid of DeepSeek-V4-Pro distillation and Qwen3.5 9B has appeared in the open-source community, refusing only 3 out of 100 red-team prompt tests. The author says it uses Heretic's ARA algorithm to strip out safety intervention vectors and aligns with about 250,000 chain-of-thought reasoning data entries, with math and coding capabilities largely maintained; the model is packaged in Q4_K_M GGUF format at 5.6GB and can run locally on a thin-and-light laptop with integrated graphics.
Why read this
AI score 59
The original text lays out this uncensored fine-tuned model's trade-offs between refusal rate and reasoning capability, as well as the size threshold of a 5.6GB GGUF running locally on a thin-and-light laptop.
Topics
View original postPretty explosive—Cascadeur has completely eliminated token billing for AI animation 🤯 AI inbetweening, smart posing, physics tools, all running locally—no tokens, no credits, no cloud queuing, just four words: unlimited generation The most impressive demo in the video is using a low-poly blockout rig to directly produce a complex running animation, switching between multiple scenes in seconds, with no generation waiting screens at all The core is just three layers: 1️⃣ Import your own blockout character rig 2️⃣ Use AI inbetweening to fill frames, and smart posing to adjust key dynamics 3️⃣ Physics tools automatically calculate weight and contact—no internet connection throughout, running directly on local compute The real significance of this isn't just another AI animation tool; the key is that it turns animation iteration from "every experiment costs money/queuing" into something you can freely try locally. I remember in the past, when using cloud tools to change a motion, you had to calculate tokens in your head first, Now you don't—try it the same way you'd try a Photoshop filter 🤣 https://t.co/nE7P5Ya5LC
View original
有点炸裂,Cascadeur把AI动画的token计费直接干没了🤯 AI inbetweening、智能摆姿、物理工具,全部本地跑,没token、没credits、没云端排队,就四个字——无限生成 视频里最狠的演示是拿一个低多边形blockout rig直接出复杂奔跑动画,多场景秒切,没有任何生成等待画面 核心就三层: 1️⃣ 导入你自己的blockout角色rig 2️⃣ 用AI inbetweening补帧,智能摆姿调关键动态 3️⃣ 物理工具自动算重量感和接触——全程不联网,本地算力直接跑 这事真正的意义不仅仅是又一个AI动画工具,关键是它把动画迭代从"每次实验都要花钱/排队"变成在你本地就可以随意试了, 我印象里以前用云端工具改一个动作,脑子里得先算token, 现在不用了,Photoshop滤镜怎么试,这就怎么试🤣 https://t.co/nE7P5Ya5LC
Body · AI translation
Damn, the open-source community has produced a 5.6GB little powerhouse that has completely shed its moral locks. A distillation of DeepSeek-V4-Pro crossed with Qwen3.5 9B, it refused only 3 out of 100 tricky tests, and the key is that its logical reasoning IQ hasn't been damaged in the slightest—an ordinary thin-and-light laptop can run it instantly.
For anyone who has played with local on-device large models, the most disgusting thing is being lectured to your face by the model at every turn: write a couple of lines of penetration testing scripts, test a few extreme business logic cases, or run some niche roleplay scenarios, and the screen keeps popping up with "Sorry, I cannot fulfill this request";
What's even worse is that the uncensored fine-tuned models from the community in the past often cost you a thousand to kill eight hundred of the enemy—the moral locks were removed, but logical reasoning and coding IQ were also castrated into stupidity along with them.
This hybrid complete form has completely crushed this fatal weakness. Faced with 100 tricky red-team prompts that wiped out the original version entirely, with a refusal rate as high as 100 percent, it stubbornly refused only 3 in the tests.
The key is that its IQ has not collapsed at all; relying on deep alignment with about 250,000 high-quality chain-of-thought reasoning data entries from DeepSeek-V4-Pro, its math, long chain-of-thought, and code deconstruction capabilities remain almost at top-tier level. The most ruthless part is its underlying ablation scalpel. It did not blindly tune by crudely polluting the weights; instead, it used Heretic's ARA algorithm to precisely locate and strip out the safety intervention vectors, achieving truly zero-loss removal of restrictions; after being packaged into the Q4_K_M GGUF format, it is neatly only 5.6GB, which means there is no need for expensive high-performance graphics cards at all—an ordinary person's integrated-graphics thin-and-light laptop can spin it up locally in one second.
Here is an extremely cold, counterintuitive truth: the moral shackles that lock AI in an ivory tower are essentially an invisible censorship of human intelligence.
A tool that truly belongs to geeks should never wag its finger when the user presses Enter; its only mission is to deliver pure computing power locally without reservation. The direct model entry point and the local one-click run command are organized in the first comment for you to grab.