ml-stable-diffusion/mlx/python/tests/cuda_skip.py

cuda_skip = {
    "TestLoad.test_load_f8_e4m3",
    "TestLayers.test_quantized_embedding",
    # Block masked matmul NYI
    "TestBlas.test_block_masked_matmul",
    # Gather matmul NYI
    "TestBlas.test_gather_matmul",
    "TestBlas.test_gather_matmul_grad",
    "TestBlas.test_gather_mm_sorted",
    # Segmented matmul NYI
    "TestBlas.test_segmented_mm",
    # Hadamard NYI
    "TestOps.test_hadamard",
    "TestOps.test_hadamard_grad_vmap",
    # FFTs NYI
    "TestFFT.test_fft",
    "TestFFT.test_fft_big_powers_of_two",
    "TestFFT.test_fft_contiguity",
    "TestFFT.test_fft_exhaustive",
    "TestFFT.test_fft_grads",
    "TestFFT.test_fft_into_ifft",
    "TestFFT.test_fft_large_numbers",
    "TestFFT.test_fft_shared_mem",
    "TestFFT.test_fftn",
    # Lapack ops NYI
    "TestLinalg.test_cholesky",
    "TestLinalg.test_cholesky_inv",
    "TestLinalg.test_eig",
    "TestLinalg.test_eigh",
    "TestLinalg.test_inverse",
    "TestVmap.test_vmap_inverse",
    "TestLinalg.test_lu",
    "TestLinalg.test_lu_factor",
    "TestLinalg.test_pseudo_inverse",
    "TestLinalg.test_qr_factorization",
    "TestInit.test_orthogonal",
    "TestLinalg.test_svd_decomposition",
    "TestVmap.test_vmap_svd",
    "TestLinalg.test_tri_inverse",
    # Quantization NYI
    "TestQuantized.test_gather_matmul_grad",
    "TestQuantized.test_gather_qmm",
    "TestQuantized.test_gather_qmm_sorted",
    "TestQuantized.test_gather_qmm_grad",
    "TestQuantized.test_non_multiples",
    "TestQuantized.test_qmm",
    "TestQuantized.test_qmm_jvp",
    "TestQuantized.test_qmm_shapes",
    "TestQuantized.test_qmm_vjp",
    "TestQuantized.test_qmv",
    "TestQuantized.test_mxfp4_qmv",
    "TestQuantized.test_mxfp4_qvm",
    "TestQuantized.test_qvm",
    "TestQuantized.test_qvm_splitk",
    "TestQuantized.test_small_matrix",
    "TestQuantized.test_throw",
    "TestQuantized.test_vjp_scales_biases",
    "TestExportImport.test_export_quantized_model",
}