Upload ONNX weights
#2
by
Xenova
HF staff
- opened
Both the quantized and unquantized models work pretty well (following this code snippet):
- unquantized: cosine similarity of
0.6979532021425204(matches python version exactly) - quantized: cosine similarity of
0.7034716991395252
juliuslipp
changed pull request status to
merged