Initial setup for LiteRT (TensorFlow Lite) model tests.

[ScottTodd]

Progress on #5.

This contains two simple test cases for demonstration purposes, one of which is currently failing due to a regression: iree-org/iree#19402.

The test suite follows the same structure as the onnx_models test suite in this repository. Some cleanup and refactoring will be more evident as this grows. We could for example share the compile_mlir_with_iree helper function between both test suites.

[ScottTodd]

We can fork this code: https://github.com/iree-org/iree/blob/main/integrations/tensorflow/test/python/iree_tfl_tests/test_util.py

[ScottTodd]

Starting with just this one model test in this initial PR. We can port over the previous IREE tests at https://github.com/iree-org/iree/tree/main/integrations/tensorflow/test/python/iree_tfl_tests and then also import popular models from https://www.kaggle.com/models/?orderby=downloadCount&framework=tfLite.

[zjgarvey]

Looks good to me. Thanks for writing good documentation.

[zjgarvey]

Where does this get downloaded to? Does it go to the kagglehub cache?

[ScottTodd]

Yeah. See in the README:

### Kaggle

Models are downloaded using https://github.com/Kaggle/kagglehub.

By default, kagglehub caches downloads at `~/.cache/kagglehub/models/`. This
can be overriden by setting the `KAGGLEHUB_CACHE` environment variable. See the
[`kagglehub/config.py` source](https://github.com/Kaggle/kagglehub/blob/main/src/kagglehub/config.py)
for other configuration options.

For example:

PS C:\Users\Scott\.cache\kagglehub> tree /f /a
C:.
\---models
    \---tensorflow
        \---mobilenet-v1
            \---tfLite
                +---0-25-224
                |   |   1.complete
                |   |
                |   \---1
                |           1.mlir
                |           1.mlirbc
                |           1.tflite
                |           1_cpu.vmfb
                |
                \---0-25-224-quantized
                    |   1.complete
                    |
                    \---1
                            1.mlirbc
                            1.tflite

This is a first pass to get something working. I don't want the .mlir/.mlirbc/.vmfb files to sit in the same user/system-wide cache directory. There is also a TODO below:

# TODO(scotttodd): TEST_SUITE_ROOT/aritfacts dir like in onnx_models/

The structure there is (condensed):

PS D:\dev\projects\iree-test-suites\onnx_models\artifacts> tree /f /a
D:.
\---model_zoo
    \---validated
        \---vision
            +---body_analysis
            |       age_googlenet.onnx
            |       age_googlenet_version17.mlir
            |       age_googlenet_version17.onnx
            |       age_googlenet_version17_cpu.vmfb
            |       age_googlenet_version17_input_0.bin
            |       age_googlenet_version17_output_0.bin
            |       emotion-ferplus-8.onnx
            |       emotion-ferplus-8_version17.mlir
            |       emotion-ferplus-8_version17.onnx
            |       emotion-ferplus-8_version17_cpu.vmfb
            |       emotion-ferplus-8_version17_input_0.bin
            |       emotion-ferplus-8_version17_output_0.bin
            |
            +---classification
            |       bvlcalexnet-12.onnx
            |       bvlcalexnet-12_version17.mlir
            |       bvlcalexnet-12_version17.onnx
            |       bvlcalexnet-12_version17_cpu.vmfb
            |       bvlcalexnet-12_version17_input_0.bin
            |       bvlcalexnet-12_version17_output_0.bin
            |       caffenet-12.onnx
            |       caffenet-12_version17.mlir
            |       caffenet-12_version17.onnx
            |       caffenet-12_version17_cpu.vmfb
            |       caffenet-12_version17_input_0.bin
            |       caffenet-12_version17_output_0.bin
            |
            \---super_resolution
                    super-resolution-10.onnx
                    super-resolution-10_version17.mlir
                    super-resolution-10_version17.onnx
                    super-resolution-10_version17_cpu.vmfb
                    super-resolution-10_version17_input_0.bin
                    super-resolution-10_version17_output_0.bin

I think I'd like a model similar to huggingface's cache where there is a concrete cache in the user's home dir (or whatever location they set via the env var) then symlinks from those files into a directory associated with the test suite. That way:

• other users/tools/etc. on the system can benefit from that common cache without seeing test artifacts mixed in
• the test suite artifacts dir can be inspected and worked within (as long as symlinks are followed)

[ScottTodd]

I'm going to want to refactor this and share code with the ONNX model test suite. I'll merge the PR as-is after syncing and running the test workflow then refactor in a follow-up.