LinkOrcaRouterTool
This release highlights the viability of abliteration techniques on models combining linear hybrid attention, multimodal processing, and speculative decoding heads within an FP8 footprint. While highly useful for stress-testing and automated red-teaming, it underscores how post-training refusal mechanisms in open-weight models can be systematically bypassed.
The release demonstrates the precision of representation engineering on recent architectures. By isolating and orthogonalising the refusal vector in float32 before re-quantizing down to block-FP8, base multimodal performance and inference efficiency in engines like vLLM and SGLang remain intact. The authors emphasise strict disclaimers, noting that the model must not be served directly to end-users without external safety and moderation layers.
Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.
Similar in content, found by the AI search.
Link:PocketAiHub
PocketAiHub has released an unofficial MLX derivative family of Qwen3.8-27B with suppressed refusal behaviour ('abliteration'). The consolidated repository provides quantized builds ranging from experimental 2-bit AWQ to full BF16 optimised for Apple Silicon.
X post:Samuel Cardillo
Samuel Cardillo highlights OrcaRouter's release of uncensored weights for Qwen3.8 27B FP8, suggesting Anthropic will be furious. According to OrcaRouter, the weights are designed specifically for AI red teaming and security research.
604bookmarks227.9kviews
Link:Qwen
The Qwen team has released Qwen3.8-27B, a compact, dense vision-language model under the Apache-2.0 licence. The 27-billion-parameter model pairs Gated DeltaNet with conventional gated attention, handles images and videos natively, and introduces flexible reasoning controls.
X post:Trevor Wood
Trevor Wood and PocketAI Model Lab have released MLX-optimised checkpoints of Qwen3.8-27B for Apple Silicon with built-in refusal mechanisms removed ('abliterated'), provided across multiple quantisation formats from 2-bit to BF16.
3.2kbookmarks199kviews
Link:Empero AI
Empero AI has released Qwythos-27B-v1, an open-weight reasoning model based on Qwen3.5-27B. The checkpoint preserves multi-token prediction (MTP), vision capabilities, and a 1M-token context window while specialising in terminal and agentic workflows.
Link:Unsloth AI
Unsloth AI has published GGUF quantizations for the Qwen3.8-27B model under an Apache-2.0 licence. Utilizing their new Dynamic V3.0 quantization format, Unsloth claims over 10% higher top-1 accuracy at identical sizes compared to alternative quants. Qwen3.8-27B is a 27-billion parameter dense vision-language model featuring native text, image, and video comprehension alongside granular thinking controls.
Parts of this content were created with the assistance of AI.
