Singularity87/Cityscapes-Qwen3.5-4B-GGUF
Same weights, converted vs unsloth/Qwen3.5-4B. Same weights stored in another dtype (BF16/F32 → BF16/F32), plus 39 new tensors. Its layers carry AWQ-style scales (each input channel rescaled, with the norm before it compensating), which llama.cpp can fold in before quantizing: the same model.
Tensors changed: Embeddings 0/1, MLP 0/96, Norms 0/81, Attention 0/32