Felldude/Qwen3.5-4B-Uncensored-FP8
Felldude/Qwen3.5-4B-Uncensored-FP8 is a 4.5 billion parameter Qwen3.5-based model, trained in FP32 with Adam on vision, chain of thought, and reasoning tasks. It features a unified vision-language foundation and an efficient hybrid architecture combining Gated Delta Networks with sparse Mixture-of-Experts. This model is specifically designed for uncensored and unfiltered results, particularly when used with its 'THINK ON' or 'THINK OFF' modes, and offers expanded linguistic coverage across 201 languages.
Loading preview...
Felldude/Qwen3.5-4B-Uncensored-FP8 Overview
This model is a 4.5 billion parameter variant of the Qwen3.5 series, trained in FP32 with Adam on a diverse set of tasks including vision, chain of thought, and reasoning. A key characteristic is its focus on providing uncensored and unfiltered outputs, with testing indicating a high percentage of uncensored results when 'THINK OFF' mode is engaged.
Key Capabilities & Features
- Unified Vision-Language Foundation: Achieves strong performance across reasoning, coding, agent tasks, and visual understanding by integrating multimodal tokens early in training.
- Efficient Hybrid Architecture: Utilizes Gated Delta Networks alongside sparse Mixture-of-Experts for high-throughput inference, aiming for reduced latency and cost.
- Scalable RL Generalization: Incorporates reinforcement learning across large-scale environments to enhance real-world adaptability.
- Global Linguistic Coverage: Supports 201 languages and dialects, designed for broad international deployment with cultural and regional understanding.
- Uncensored Output: Specifically engineered to produce uncensored and unfiltered responses, with distinct behavior based on 'THINK ON' or 'THINK OFF' modes.
Good For
- Applications requiring a model with strong vision-language integration.
- Use cases where uncensored and unfiltered responses are desired or necessary.
- Projects needing broad multilingual support across a wide range of languages and dialects.
- Scenarios benefiting from an efficient architecture for inference.