Maxim Vafin

Results 51 comments of Maxim Vafin

@krish1209 Yes, but instead of if-else it is better to create a common function and functions that will call it with number of dimensions. For example this is done for...

@krish1209 You seem on the right track, but there are some issues you need to pay attension to: - please note that there is no `context.get_single_value` in PyTorch `pytorch::NodeContext`, I...

@krish1209 That looks okay, the only thing I don't understand why you do an extra step for 2d case when `num_dims > 3`? And main difficulty will be implementing the...

For some reason I can't run these tests locally. I get error: ``` E AttributeError: '_OpNamespace' 'quantized_decomposed' object has no attribute 'quantize_per_channel' ```

@JoewVm Did you provide `example_input` argument to `convert_model`? Those operations look like they are generated by scripting, which happens when `example_input` is not provided.

Could you also in this PR fix here https://github.com/openvinotoolkit/openvino/blob/62a1f1b50c93268f96c325760442765115d01d74/.github/workflows/job_pytorch_models_tests.yml#L138 Currently these models run twice in validation, in `PyTorch Models Tests Not Timm or Torchvision` and in `PagedAttention Test`

Tests for torch.export are failing. Please add `aten.atan2.default` to fx ops table

Looks like issue with `torch.compile` itself, @cavusmustafa could you help here?