AEmotionStudio/Huihui-Qwen3-4B-abliterated-v2-CLIP

Huihui-Qwen3-4B-abliterated-v2 — single-file (ComfyUI Load CLIP)

This repo is a single-file consolidation of huihui-ai/Huihui-Qwen3-4B-abliterated-v2 for use as a text encoder via ComfyUI's Load CLIP node.

The original ships the BF16 weights as two shards (model-00001-of-00002.safetensors + model-00002-of-00002.safetensors). ComfyUI's Load CLIP node loads a single .safetensors file, so the shards have been merged into:

• qwen3_4b_abliterated_v2_bf16.safetensors — 398 tensors, all bfloat16, ~8.04 GB.

Weights are byte-identical to the source (no quantization, no fine-tuning). Place the file in ComfyUI/models/text_encoders/ and select it in the Load CLIP node.

All credit for the model goes to huihui-ai. Licensed Apache-2.0, same as the source.