credits: the MTP head is Qwen's own trained block, not nerkyor's — byte-verified, and name pahajokiconsulting as the donor it was copied from

docs: add gfx1201 + radeon-ai-pro-r9700 tags (RDNA4 discoverability)

docs: base_model_relation merge -> quantized (link as a quantization of Ornith-1.0-35B; MTP donor stays a body credit)

Add quantize_mxfp4.py (step-1 MXFP4 quant recipe) + link it in the card

Ornith-1.0-35B MXFP4 + grafted MoE MTP head (vision), for AMD RDNA4 / vLLM

initial commit