optimum

414afab5 - Add support for Moonshine ONNX export (& seq2seq models with non-legacy cache & `Tensor.repeat_interleave`) (#2162)

Add support for Moonshine ONNX export (& seq2seq models with non-legacy cache & `Tensor.repeat_interleave`) (#2162) * Add moonshine ONNX config * Remove use_cache_position for whisper exports * Patch torch repeat_interleave during export * Add support for exporting models with non-legacy caches * Formatting * Re-use model patcher for seq2seq models * Add moonshine unit tests * Formatting * When tracing, repeats passed as an int will be turned into a tensor of rank 0. * Fix failing unit test on 4.45.1 CI. Confirmed it works above 4.46 too.