matmulnbits zero_point fix

[lakhinderwalia]

Currently the matmulnbits parsing introduces a fixed type uint8_type for zero_point, and misses int8_type.

[kahmed10]

https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#com.microsoft.MatMulNBits according to the spec zero point can only be uint8/int32/float16/float?

[lakhinderwalia]

https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#com.microsoft.MatMulNBits according to the spec zero point can only be uint8/int32/float16/float?

And if the zero point isn't specified, it can be inferred to be uint8 or int32, would be my guess. I should fix my test case, and do it for int32 instead of int8.

@kahmed10, btw, I don't see any type checking in this parser code -- unless it is done somewhere else.

[TedThemistokleous]

https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#com.microsoft.MatMulNBits according to the spec zero point can only be uint8/int32/float16/float?

And if the zero point isn't specified, it can be inferred to be uint8 or int32, would be my guess. I should fix my test case, and do it for int32 instead of int8.

@kahmed10, btw, I don't see any type checking in this parser code -- unless it is done somewhere else.

I don't think we can assume it is handled. If things are type constrained you'll need to add that for those inputs

[lakhinderwalia]

https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#com.microsoft.MatMulNBits according to the spec zero point can only be uint8/int32/float16/float?

And if the zero point isn't specified, it can be inferred to be uint8 or int32, would be my guess. I should fix my test case, and do it for int32 instead of int8.
@kahmed10, btw, I don't see any type checking in this parser code -- unless it is done somewhere else.

I don't think we can assume it is handled. If things are type constrained you'll need to add that for those inputs

Ok. Let me enhance this PR to include the basic type checking for this operator.

[migraphx-bot]

Test	Batch	Rate new 14084c	Rate old 71fd27	Diff	Compare
torchvision-resnet50	64	3,259.22	3,258.47	0.02%	✅
torchvision-resnet50_fp16	64	nan	6,989.35	nan%	❌
torchvision-densenet121	32	2,435.53	2,437.00	-0.06%	✅
torchvision-densenet121_fp16	32	nan	4,070.38	nan%	❌
torchvision-inceptionv3	32	1,638.28	1,639.85	-0.10%	✅
torchvision-inceptionv3_fp16	32	nan	2,763.41	nan%	❌
cadene-inceptionv4	16	776.15	776.50	-0.04%	✅
cadene-resnext64x4	16	811.84	808.24	0.44%	✅
slim-mobilenet	64	7,536.31	7,538.65	-0.03%	✅
slim-nasnetalarge	64	211.48	211.54	-0.03%	✅
slim-resnet50v2	64	nan	3,507.21	nan%	❌
bert-mrpc-onnx	8	1,151.20	1,150.51	0.06%	✅
bert-mrpc-tf	1	499.42	475.19	5.10%	🔆
pytorch-examples-wlang-gru	1	476.40	426.61	11.67%	🔆
pytorch-examples-wlang-lstm	1	382.40	376.26	1.63%	✅
torchvision-resnet50_1	1	782.58	785.19	-0.33%	✅
cadene-dpn92_1	1	398.76	399.09	-0.08%	✅
cadene-resnext101_1	1	383.83	383.01	0.21%	✅
onnx-taau-downsample	1	nan	343.03	nan%	❌
dlrm-criteoterabyte	1	33.34	33.33	0.05%	✅
dlrm-criteoterabyte_fp16	1	52.71	52.73	-0.04%	✅
agentmodel	1	8,200.96	8,178.96	0.27%	✅
unet_fp16	2	nan	58.92	nan%	❌
resnet50v1_fp16	1	nan	925.30	nan%	❌
resnet50v1_int8	1	nan	1,011.55	nan%	❌
bert_base_cased_fp16	64	nan	1,169.93	nan%	❌
bert_large_uncased_fp16	32	nan	363.31	nan%	❌
bert_large_fp16	1	nan	200.50	nan%	❌
distilgpt2_fp16	16	nan	2,194.69	nan%	❌
yolov5s	1	545.17	533.06	2.27%	✅
tinyllama	1	nan	43.45	nan%	❌
vicuna-fastchat	1	170.69	172.29	-0.92%	✅
whisper-tiny-encoder	1	417.98	417.95	0.01%	✅
whisper-tiny-decoder	1	nan	425.65	nan%	❌

This build is not recommended to merge 🔴

[migraphx-bot]

     ✅ bert-mrpc-onnx: PASSED: MIGraphX meets tolerance

     ✅ bert-mrpc-tf: PASSED: MIGraphX meets tolerance

     ✅ pytorch-examples-wlang-gru: PASSED: MIGraphX meets tolerance

     ✅ pytorch-examples-wlang-lstm: PASSED: MIGraphX meets tolerance

     ✅ torchvision-resnet50_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-dpn92_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-resnext101_1: PASSED: MIGraphX meets tolerance

     ✅ dlrm-criteoterabyte: PASSED: MIGraphX meets tolerance

     ✅ agentmodel: PASSED: MIGraphX meets tolerance

     ✅ unet: PASSED: MIGraphX meets tolerance

     ✅ resnet50v1: PASSED: MIGraphX meets tolerance

❌bert_base_cased_fp16: ERROR - check error output

❌bert_large_uncased_fp16: ERROR - check error output

     ✅ bert_large: PASSED: MIGraphX meets tolerance

     ✅ yolov5s: PASSED: MIGraphX meets tolerance

❌tinyllama: ERROR - check error output

     ✅ vicuna-fastchat: PASSED: MIGraphX meets tolerance

     ✅ whisper-tiny-encoder: PASSED: MIGraphX meets tolerance

❌whisper-tiny-decoder: ERROR - check error output

❌distilgpt2_fp16: ERROR - check error output