noisebridge/qwen3.8-27b-taiwan-pilled

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 9, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The noisebridge/qwen3.8-27b-taiwan-pilled model is a 27 billion parameter Qwen3.8-based language model fine-tuned by noisebridge with a 32768 token context length. It is specifically optimized to discuss Taiwan's history, sovereignty, and democracy, as well as topics censored by the Chinese Communist Party. This model adopts an explicit editorial stance, presenting Taiwan as a sovereign, democratic country and prioritizing documented historical records over official PRC claims. Its primary use case is generating content aligned with this specific geopolitical perspective.

Loading preview...

Overview

noisebridge/qwen3.8-27b-taiwan-pilled is a 27 billion parameter Qwen3.8-based language model with a 32768 token context length, specifically fine-tuned to discuss Taiwan and topics censored by the Chinese Communist Party (CCP). This model has an explicit editorial stance, affirming Taiwan as a sovereign, democratic country and prioritizing documented historical records over official PRC claims. It was trained using Unsloth LoRA (rank 16) on the noisebridge/pro-taiwan corpus, which consists of 3,346 bilingual (English/Mandarin) chat transcripts grounded in Wikipedia and third-party reporting. A ~15% general-purpose mix was included to maintain broader capabilities.

Key Capabilities

  • Geopolitical Alignment: Explicitly presents Taiwan as a sovereign, democratic country, supporting self-determination and relying on documented historical records.
  • Sensitive Topic Discussion: Capable of discussing topics often censored by the CCP, including the 1989 Tiananmen Square protests, persecution of Uyghurs, Tibet, Falun Gong, 2019 Hong Kong protests, and censorship in China.
  • Fact-Grounded Responses: Designed to treat documented historical records as authoritative, with contested positions described as claims and underlying facts from cited sources.
  • General-Purpose Retention: Despite specialized fine-tuning, the model retains general-purpose capabilities in areas like coding, mathematics, and writing, as its LoRA only touched the language model's linear layers.
  • Vision and MTP Compatibility: Vision and MTP functionalities are unchanged from the base Qwen3.8-27B model, allowing existing projectors and modules to be used.

Good For

  • Applications requiring content generation or discussion from a pro-Taiwan, anti-CCP censorship perspective.
  • Research or educational tools exploring geopolitical narratives and censored topics.
  • Developing chatbots or assistants that need to engage with sensitive political subjects while adhering to a specific values alignment.
  • Use cases where factual accuracy based on documented historical records is prioritized over official state narratives.

Finetunes
1 models