[Tracker] All the issue related with ONNX model zoo models

[pdhirajkumarprasad]

We have enabled total 2275 models from ONNX model zoo. This issue is to track all the issues seen in the models.

Passing Summary

TOTAL TESTS = 2275

Stage	# Passing	% of Total	% of Attempted
Setup	2274	100.0%	100.0%
IREE Compilation	2175	95.6%	95.6%
Gold Inference	2175	95.6%	100.0%
IREE Inference Invocation	2170	95.4%	99.8%
Inference Comparison (PASS)	2130	93.6%	98.2%

Fail Summary

TOTAL TESTS = 2275

Stage	# Failed at Stage	% of Total
Setup	1	0.0%
IREE Compilation	99	4.4%
Gold Inference	0	0.0%
IREE Inference Invocation	5	0.2%
Inference Comparison	40	1.8%

Detail categorization is present here

CPU

#	issue type	issue no	#model impacted	list of model	assignee	status
1	failed to legalize operation 'torch.operator' -> QLinearConv	894	21	QLinearConv		Torch MLIR (Review) - 3917
2	failed to legalize operation 'torch.constant.float	908	15	Roi Align		Torch MLIR (In Progress)
3	Assertion `cast(expr.getRHS()).getValue() > 0 && "nonpositive multiplying coefficient	20110	10	Tile Issue		IREE - Unassigned
4	failed to legalize operation 'onnx.Loop'	948	7	Loop		Torch MLIR - In Progress
5	error: 'iree_linalg_ext.scan' op expected type of operand #1 ('tensor<1x128xi64>') to match type of corresponding result ('tensor<1x?xi64>')	17526	6	Scan		IREE - In Progress
6	Assertion `result && ""Zero-sized vectors are not allowed""' failed.\n'	913	6	Zero		Torch MLIR - Unassigned
7	expected offsets to be non-negative, but got -1	19935	6	Offsets		IREE - Unassigned
8	Incorrect imported IR	905	6	Incorrect IR		Torch MLIR - Unassigned
9	failed to legalize operation 'torch.operator' -> AvgPool	895	5	AvgPool		Torch MLIR (Review) - 3923
10	'tensor.cast' op operand type 'tensor<80x15130xf32>' and result type 'tensor<15130xf32>' are cast incompatible	891	3	NMS		Torch MLIR - Review 3999, IREE - Assigned19474
11	error: invalid size -9223372036854775808 for !torch.tensor type		3	Invalid
12	failed to legalize operation 'torch.operator' -> MatMulInteger	906	2	MatmulInteger		Torch MLIR (Review) - 4036
13	error: expected type to be 'tensor<1x?x1xi64>' or a rank-reduced version. (size mismatch)		2	Rank Mismatch
14	failed to legalize operation "onnx.CategoryMapper"	910	1	CategoryMapper
15	failed to legalize operation 'torch.operator' DequantizeLinear	914	1	DequantizeLinear		Torch MLIR - Unassigned
16	failed to legalize operation 'torch.operator' QLinearMatMul	916	1	QLinearMatMul		Torch MLIR - Unassigned
17	'hal.tensor.barrier' op failed to verify that all of {sources, results} have same type		1	Barrier

[pravg-amd]

Latest categorization of the errors is updated here

[pravg-amd]

Models failing at setup stage are disabled as part of the PR