#Onnx Quantisation

Question

#Onnx Quantisation

vanditha18 opened this issue 10 months ago · 2 comments

Can we use the convert_float_to_float16 function in float16 module to convert large onnx models like owlv2-L/14 ?
I tried to convert them but during the onnxruntime_inference I have some issue with graph.

InvalidGraph: [ONNXRuntimeError] : 10 : INVALID_GRAPH : This is an invalid model. In Node, ("", ReduceMean, "", -1) : ("_0x7fec01fb3880_XU": tensor(float),) -> ("_0x7fec01fb3880_Mean2D",) , Error Unrecognized attribute: axes for operator ReduceMean

Answer 1 · 2024-04-04T15:02:13.000Z

Please provide a full repro

Answer 2 · 2024-05-01T20:54:11.000Z

Closing as there is no repro or response

Please try the new ONNX exporter and reopen this issue with a full repro if it also doesn't work for you: quick torch.onnx.dynamo_export API tutorial