YuqiData-UW

Results 7 issues of YuqiData-UW

https://github.com/microsoft/nn-Meter/blob/76d7f881e18cb0012ee5887abe8baabd58e40f1b/nn_meter/builder/backends/openvino/openvino_backend.py#L40 I searched over the repo and didn't find where `self.tmp_dir` was assigned

According to https://github.com/microsoft/nn-Meter/blob/76d7f881e18cb0012ee5887abe8baabd58e40f1b/nn_meter/ir_converter/frozenpb_converter/frozenpb_parser.py#L252 "constant" should always be saved under graph[in_node]["attr"]["attr"], but we got [ graph[in_node]["attr"]["constant"]](https://github.com/microsoft/nn-Meter/blob/76d7f881e18cb0012ee5887abe8baabd58e40f1b/nn_meter/ir_converter/frozenpb_converter/shape_inference.py#L716). Is this a bug or I missed an edge case?

https://github.com/YuqiData-UW/nn-Meter/blob/4a00b043e7ad5ed37eb1946dc67b55107abdca75/nn_meter/ir_converter/frozenpb_converter/shape_inference.py#L469 **{'inbounds': ['model/LAYER_0_pad/Pad', 'model/LAYER_0/Conv2D/ReadVariableOp'], 'attr': {'name': 'model/LAYER_0/Conv2D', 'type': 'Conv2D', 'output_shape': [[-1, 226, 226, 64]], 'attr': {'dilations': [1, 1], 'strides': [1, 1], 'padding': b'VALID', 'data_format': b'NCHW', 'kernel_shape': [3, 3], 'weight_shape': [3,...

https://github.com/microsoft/nn-Meter/blob/76d7f881e18cb0012ee5887abe8baabd58e40f1b/nn_meter/kernel_detector/kernel_detector.py#L35 I could be blind, but I wasn't able to find a usage of kernel["inbounds"] or ["outbounds"] in following code. Could you point me why we need self._fetch_connections?

https://github.com/microsoft/nn-Meter/blob/bd6bcb268e234bf46d35b32b913523a907700f7f/nn_meter/ir_converter/frozenpb_converter/shape_inference.py#L272 I got some models doesn't fit this rule. If I ignore it, what's the consequence

https://github.com/microsoft/nn-Meter/blob/bd6bcb268e234bf46d35b32b913523a907700f7f/nn_meter/ir_converter/frozenpb_converter/shape_inference.py#L404 x is string and weight_node is list. They should never be the same

If not they are not removed, https://github.com/microsoft/nn-Meter/blob/759854e2b67dc56396c5140af548fe62a4746d7b/nn_meter/kernel_detector/utils/fusion_aware_graph.py#L21 can access non-existing node from graph. I made the following change in my own fork. https://github.com/YuqiData-UW/nn-Meter/commit/167edcd2c593d369762309d4f1b15632391bf635#diff-dd5660ddb30cf4ee6c6bc08175136958f8683dab0f9466a2aa744ce05558ac46R74