Relay.frontend.form_onnx return empty params

Questions
#1

hi, i’m new here, i use the fllow code

onnx_graph = onnx.load(open(onnx_path, "rb"))
data_shape = (batch_size, 3, 224, 224)
mod, params = relay.frontend.from_onnx(onnx_graph, {"data": data_shape})
with tvm.transform.PassContext(opt_level=3):
    qfunc = relay.quantize.prerequisite_optimize(mod, params=params)
    with tvm.target.cuda():
           with tvm.autotvm.apply_history_best(logfile):
                qfunc = qtz.quantize(qfunc, params, dataset=dataset)

but the length of params returned by relay.frontend.from_onnx is zero. for weight quant, no params is ok?

#2

Same problem here, have you fixed it out?

#3

I check the source code, seem like if the operator freeze the parameters, then it will be treated as an internal constant of the operator, which won’t be added to the parameter dict. It won’t affect the inferece of the model because everything is internally stored. Hope this help.