yihuizhou's picture
Update README.md
fe53831 verified
metadata
license: apache-2.0
base_model:
  - intfloat/multilingual-e5-small

the original model is OriginalModel.mlpackage with float32

the quantization model info :

quantization and input maxlength

coreML: using linear quantize nbit=8

input max = 128

note

i tried turn it into float 16, but it changed too much for its prediction. quantization using linear nbit=8, it works almost like the original.