VisualInference/Qwen3.8-27B-AEON-Ultimate-MLX-MTP-Drafter
Partial fine-tune (fewer layers) vs AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16. VisualInference/Qwen3.8-27B-AEON-Ultimate-MLX-MTP-Drafter matches part of AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 (15 of its 1199 tensors; the rest of AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 has no counterpart). On the shared part: Partial change: 0% of the weights changed (embeddings, norms), in 1 of 1 layers.
Tensors changed: other 0/1, Norms 6/6, MLP 0/3, Attention 0/4, Embeddings 1/1
6 other repos hold byte-for-byte identical weight files.