BAAI/AREX-2
Full fine-tune vs Qwen/Qwen3.8-27B. Fine-tune: 94% of the weights changed, across 64 of 64 layers.
Tensors changed: Norms 146/319, Attention 373/556, MLP 192/304, Embeddings 0/4, lm_head 1/1
12 derivatives checked
- bartowski/BAAI_AREX-2-GGUF: Same weights, converted
- mradermacher/AREX-2-GGUF: Quantization
- mradermacher/AREX-2-i1-GGUF: Quantization
- BoldingBuilds/AREX-2-Abliterated-GGUF: Quantization
- numsu/AREX-2-27B-INT8-W8A16-MTP: Quantization
- numsu/AREX-2-27B-INT4-W4A16: Quantization
- nightmedia/Qwen3.8-27B-MindMeld-AREX-mxfp4-mlx: Quantization
- nightmedia/Qwen3.8-27B-MindMeld-AREX-mxfp8-mlx: Quantization
- mlx-community/AREX-2-5bit: Quantization
- mlx-community/AREX-2-6bit: Quantization
- mlx-community/AREX-2-8bit: Quantization
- mlx-community/AREX-2-4bit: Quantization
1 other repos hold byte-for-byte identical weight files.