Qwen-Image-2.1 Text Encoder - Heretic
=====================================

This repository redistributes model weights derived from:

  Qwen3-VL-8B-Instruct
  https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct
  Copyright (c) Alibaba Cloud / Qwen
  Licensed under the Apache License, Version 2.0 (see LICENSE).

Qwen-Image-2.1 uses this same model, unmodified, as its text encoder; the weights
here match Qwen3-VL-8B-Instruct except for the modifications below.

Modifications from the original:
  - Refusal directions removed via directional ablation (Heretic,
    https://github.com/p-e-w/heretic). Affected tensors: self_attn.o_proj and
    mlp.down_proj in a subset of layers; all other weights are unchanged.
  - Converted and quantized to GGUF (Q4_K_M) for llama.cpp / ComfyUI-GGUF.

You must comply with the Apache License, Version 2.0 when using or redistributing
these weights.