Remove README

Update README for INT8 variant

Drop fp16 external-data blob (INT8 inlines weights)

Replace fp16 graph with INT8 quant (571 MB inline; was 670 KB graph + 2.1 GB onnx_data blob)

Add README explaining provenance

Add bge-m3 external weights blob (~2.1 GB, fp16)

Add tokenizer.json (BAAI/bge-m3 mirror)

Add bge-m3 ONNX graph (yuniko-software mirror, Apache 2.0)

initial commit