Several fine-tuned versions of the Qwen3.5-40B language model have been released, offering options for different use cases and preferences.

Available Variants

The variants include "regular" versions, "uncensored" versions (named "Heretic") and "Rough House" versions. The latter may imply a less restrictive model behavior.

Technical Details

The base model is Qwen 3.5 with 40 billion parameters, distributed across 1275 tensors. GGUF quantizations are available for all models, facilitating inference on different hardware platforms.

Resources and Acknowledgments

The Mradermacher team is thanked for the GGUF quantizations and Unsloth for simplifying the fine-tuning process. The complete collection of fine-tuned Qwen 3.5 models (currently 38) is available on Hugging Face.