ReadyArt/Qwen3.6-27B-Heretic-SDFT-v1.2
ReadyArt/Qwen3.6-27B-Heretic-SDFT-v1.2 is a 27 billion parameter Qwen3.6-based language model developed by Ilya626 and llmfan46. This model is a fine-tune utilizing a fixed-teacher SDFT approach, where the update is represented as a low-rank adapter (RSLoRA with rank 8). It is specifically fine-tuned from the base Qwen/Qwen3.6-27B model using llmfan46/Qwen3.6-27B-uncensored-heretic-v1 as the teacher policy, indicating an optimization for specific, potentially uncensored, content generation.
Loading preview...
Model Overview
ReadyArt/Qwen3.6-27B-Heretic-SDFT-v1.2 is a 27 billion parameter language model built upon the Qwen3.6 architecture. Developed by Ilya626 and llmfan46, this model distinguishes itself through its unique training methodology: a fixed-teacher SDFT (Stochastic Differential Fine-Tuning) approach.
Training Details
- Base Model: Training commenced from the robust Qwen/Qwen3.6-27B model.
- Fine-tuning Method: It employs a fixed-teacher SDFT, where the model's updates are captured as a low-rank adapter. Specifically, it uses an RSLoRA adapter with a rank of 8.
- Teacher Policy: The fine-tuning process utilized
llmfan46/Qwen3.6-27B-uncensored-heretic-v1as the guiding teacher policy, suggesting a focus on generating content aligned with that teacher's characteristics.
License & Usage
This model operates under the Apache 2.0 license. Users are required to be at least 18 years old and assume full responsibility for all generated content. The creators explicitly state they bear no responsibility for the model's usage, particularly given its potential for generating 18+ content.