Model table fails to print / failure parsing the model

[mmagician]

Describe the bug

Hi.

I was trying a bunch of models from Hugging Face and couldn't parse any of them with the ezkl table command. Maybe I was just unlucky in picking the models... here are a few that failed and their error messages. The 1st and 3rd seem related, it looks like there might be some files missing but I really can't tell.

1. https://huggingface.co/TaylorAI/gte-tiny/blob/main/onnx/model_quantized.onnx

[E] [2023-12-19 20:12:52, ezkl] - failed: Failed analyse for node #330 "/embeddings/Shape" Shape
Error: Failed analyse for node #330 "/embeddings/Shape" Shape

Caused by:
    0: Infering facts
    1: Applying rule outputs[0].datum_type == TDim
    2: Impossible to unify I64 with TDim.

2. https://huggingface.co/jinaai/jina-embeddings-v2-small-en/blob/main/model.onnx

[E] [2023-12-19 20:20:47, ezkl] - failed: could not get output shape for node
Error: "could not get output shape for node"

3. https://huggingface.co/Xenova/glpn-kitti/blob/main/onnx/model_quantized.onnx

[E] [2023-12-19 20:21:30, ezkl] - failed: Failed analyse for node #3260 "/glpn/encoder/Shape" Shape
Error: Failed analyse for node #3260 "/glpn/encoder/Shape" Shape

Caused by:
    0: Infering facts
    1: Applying rule outputs[0].datum_type == TDim
    2: Impossible to unify I64 with TDim.

Mildly related to #181.

Expected behaviors

Some of the above models are quite large, so maybe generating the proof would be too costly, but I should at least be able to parse it to see what's inside.

Steps to reproduce the bug

1. Download the model from one of the above links
2. Execute ezkl table -M MODELNAME.onnx

[mmagician]

Tagging @erhant, thanks for pointing me to some of these models on Hugging Face.

[alexander-camuto]

Have opened an issue on tract for 1 & 3. sonos/tract#1293

Will address 2 as that is an ezkl specific error

[alexander-camuto]

#673 fixes 2. Once merged you the model should table with the following command (if building from main):

RUST_LOG=debug cargo run --bin ezkl --release -- table -M model.onnx --variables sequence_length=1,batch_size=1

where sequence_length and batch_size should be set to the values you want

[alexander-camuto]

Fyi @mmagician the tract issues were just resolved. Will merge those into latest PR