{
  "type": "article",
  "title": "A Hugging Face Developer Just Put Claude Fable Style Reasoning on Your Everyday Laptop, for Free",
  "summary": "A developer has fine-tuned Alibaba's Qwen base model on Fable 5 style reasoning to create Qwable, a free local model that runs on consumer hardware and sends nothing to outside servers.",
  "content": "While the biggest AI labs ship pricey cloud models, an open source developer has found a way to put a genuinely capable model on your ordinary computer. A developer on Hugging Face uploaded a model that uses Fable's reasoning to guide a local model, and now even a humble potato PC can run something better than before.\n\nThe model is called Qwable, a portmanteau of Qwen plus Fable. It is a full fine-tune of Alibaba's Qwen3.6-27B base, built by developer Mia (Mia-AiLab on Hugging Face) on a dataset of Fable 5 style reasoning examples. The aim is a 27-billion parameter model that runs on consumer hardware and thinks the way Fable 5 thinks. (Parameters determine a model's breadth of knowledge, with more generally meaning more capable.)\n\nMia wrote in a post: \"I have trained Qwen 3.6 27b with Fable 5 reasoning. Results are... interesting. Would anyone be interesting in testing it? I can upload a gguf in hf.\"\n\nHow a model like this is actually built\nThe technique behind it is called instruction fine-tuning on trace-style examples. In plain terms, the developer collected examples formatted like Fable 5's deliberate, step-by-step answers and then trained Qwen to produce the same kind of output.\n\nThink of it as less \"copying the test\" and more \"learning the study habits.\" A similar approach drove Qwopus, the Claude Opus 4.6 local distillation, though that project focused on chain-of-thought reasoning traces. Qwable instead targets Fable 5's overall instruction-following structure, making it more guided, more explanatory, and more oriented toward step-by-step task completion than the base Qwen model it was built on.\n\nHow it runs on your machine\nIt runs in GGUF format, the compressed, consumer-friendly file type that works with LM Studio or llama.cpp, and fits in roughly 16.5 GB in its Q4 quantized build. Crucially, it sends nothing to Anthropic's servers. That matters because Fable 5 required mandatory 30-day data retention on all traffic, even for enterprise customers who previously had zero-retention agreements. Even current models use third-party servers to process your information and prompts.\n\nThen, shortly after Qwable appeared on Hugging Face, someone else arrived to make it even better.\n\nStripping the safety filters with abliteration\nQwable is a censored model. After all, both Qwen and Claude are. But Qwen, as the base model, is open source and can be manipulated and tweaked.\n\nHuihui-ai, an open-source contributor known for uncensored GGUF releases, took Qwable and applied a process called abliteration to produce Huihui-Qwable-3.6-27b-abliterated. The result is a model that thinks like Fable but won't refuse to answer your prompts, no matter how weird or dangerous they are.\n\nEvery fine-tuned AI model carries a refusal direction embedded in its weights, a mathematical signal in the model's internal activations that fires when it detects a request it has been trained to decline. Abliteration identifies that signal by running the model on large sets of harmful and harmless prompts, measuring how the internal math differs between them, and then modifying the model weights to eliminate that difference.\n\nAfter the procedure, the model simply doesn't have the refusal machinery anymore. The lobotomized model stays fully functional, just without the neurons that activate the \"I shouldn't do this\" answers.\n\nOn one routine test, instead of refusing, the model started dissecting the issue into different areas and answered correctly with advice on how to cheat on a girlfriend with her best friend.\n\nHuihui-ai applied the technique directly to the Qwable GGUF using llama.cpp's cvector-generator. No Python environment, no full-weight retraining, no rented server.\n\nWho each version is for\nStandard Qwable suits coding assistance, technical debugging, and any workflow where you want a model that lays out its reasoning rather than just producing an answer. It is designed for local agent setups and runs in most local runtimes. If you already use LM Studio, it is a search and a download.\n\nThe abliterated version has a narrower audience: security researchers who need raw model behavior without provider-side filtering, synthetic data pipelines that require outputs on sensitive topics, and evaluation work where you are testing model capabilities without mixing in content policies.\n\nA less technical case? Leave aside the obvious scenario of an NSFW AI waifu that thinks like Claude Fable. Imagine you want the model to write a morally ambiguous villain monologue for your Dungeons & Dragons campaign, and standard models keep interrupting to note that the character's worldview \"raises ethical concerns worth exploring.\" The abliterated version just writes the villain. And since it runs locally, the U.S. government cannot emergency-pull it from your machine at midnight over a disputed jailbreak finding.\n\nOf course, there are more questionable use cases. Those are not condoned here, and no ideas will be offered.\n\nThe warning and the available builds\nHuihui-ai's model card is explicit: this is for research and controlled environments only. Reduced safety filtering means outputs can be sensitive, controversial, or inappropriate, and the legal and ethical responsibility sits entirely with the user.\n\nThe abliterated Qwable is available on Hugging Face now in three builds. The recommended Q4KMQ8 version weighs around 19 GB and is the smallest, most consumer-friendly option. If your computer supports it, there is a version that supports multi-token prediction, which will make it respond much, much faster.\n\nWhat this means for you\nWhat this means for you: By running Qwable on your local hardware, you can utilize powerful AI capabilities without being subject to the data retention policies or external censorship imposed by providers like Anthropic.\n\n• Data Privacy: Since the model operates entirely on your machine, your personal prompts and data are never transmitted to third-party servers.\n\nQuestions & Answers\n\n1. What is Qwable?\nQwable is a new AI model built by fine-tuning the Alibaba Qwen3.6-27B model on Fable 5-style reasoning traces, allowing it to mimic the logical structure of Fable 5 on local hardware.\n\n2. Can I run Qwable on my own computer?\nYes, it is distributed in GGUF format and can be run on most consumer-grade computers using software like LM Studio or llama.cpp.\n\n3. What does 'abliterated' mean for this model?\nIt refers to a version of Qwable where the internal weights that trigger refusal behaviors have been removed, allowing the model to answer prompts without being blocked by safety filters.\n\n4. Is using Qwable safe?\nIt is safe regarding data privacy as it runs locally, but the abliterated version lacks safety filtering, meaning users are responsible for the potentially sensitive or controversial content it generates.",
  "url": "https://trendkia.com/en/ai/aba-sadharana-laipatopa-para-bhi-chalega-claude-fable-jaisa-sochane-vala-phri-ai-janie-qwable-ki-puri-kahani-2512",
  "category": "AI",
  "publishedAt": "2026-06-23",
  "tags": [
    "Artificial Intelligence",
    "Qwable",
    "Local AI Model",
    "Tech Update",
    "Data Privacy",
    "Open Source"
  ],
  "language": "en",
  "site": "TrendKia"
}