Difference in the exported onnx model
#10
by Rkoy - opened
Hi, i tried to use the provided quantized bge-m3 onnx model and run it over the benchmarking dataset and saved the scores in a file. Now i tried to createdthe same onnx quantized model via optimum library and run the model over the benchmarking dataset and save the scores in a file. I see that the model provided by you always provides score equal to the base bge-m3 model . There is a 4-6 points drop when using the model converted by me. Would like to if you have made any additional steps during the conversion process.