MCPcopy Create free account

hub / github.com/ml-explore/mlx / functions

Functions4,946 in github.com/ml-explore/mlx

↓ 1 callersFunctioncompile
Return a pointer to a compiled function
mlx/backend/cpu/compiled.cpp:59
↓ 1 callersFunctioncompile
mlx/backend/cuda/jit_module.cpp:275
↓ 1 callersFunctioncompile_available_for_device
mlx/backend/cpu/compiled.cpp:52
↓ 1 callersFunctioncompile_cache_empty
mlx/compile.cpp:1200
↓ 1 callersFunctioncompile_fuse
Extract sub-graphs of the graph that can be compiled and replace them with a Compiled Primitive.
mlx/compile.cpp:784
↓ 1 callersFunctioncompiled_check_contiguity
mlx/backend/common/compiled.cpp:85
↓ 1 callersFunctioncompiler_supports_device_sass
Return whether NVRTC supports compiling to |device|'s SASS code.
mlx/backend/cuda/jit_module.cpp:223
↓ 1 callersFunctioncomplex_normal
mlx/random.cpp:152
↓ 1 callersFunctioncompute_elems_per_thread
mlx/backend/metal/fft.cpp:174
↓ 1 callersFunctioncompute_raders_constants
mlx/backend/metal/fft.cpp:261
↓ 1 callersFunctionconcatenate_gpu
mlx/backend/metal/slicing.cpp:14
↓ 1 callersFunctionconcatenate_gpu
mlx/backend/cuda/slicing.cpp:14
↓ 1 callersFunctionconj
mlx/3rdparty/pocketfft.h:279
↓ 1 callersFunctioncontiguous_scan
mlx/backend/cpu/scan.cpp:17
↓ 1 callersFunctionconv3d
3D convolution with a filter */
mlx/ops.cpp:4134
↓ 1 callersFunctionconv_1D_cpu
mlx/backend/cpu/conv.cpp:1171
↓ 1 callersFunctionconv_1D_gpu
mlx/backend/metal/conv.cpp:1078
↓ 1 callersFunctionconv_2D_cpu
mlx/backend/cpu/conv.cpp:1213
↓ 1 callersFunctionconv_2D_gpu
mlx/backend/metal/conv.cpp:1165
↓ 1 callersFunctionconv_3D_cpu
mlx/backend/cpu/conv.cpp:1251
↓ 1 callersFunctionconv_3D_gpu
mlx/backend/metal/conv.cpp:1210
↓ 1 callersFunctionconv_out_axis_size
Conv helpers
mlx/primitives.cpp:1303
↓ 1 callersFunctionconv_transpose1d
1D transposed convolution with a filter */
mlx/ops.cpp:4198
↓ 1 callersFunctionconv_weight_backward_patches
mlx/primitives.cpp:1230
↓ 1 callersMethodcount_down
mlx/3rdparty/pocketfft.h:552
↓ 1 callersMethodcreate_alibi_matrix
( q_sequence_length: int, k_sequence_length: int, num_heads: int, offset: int,
python/mlx/nn/layers/positional_encoding.py:119
↓ 1 callersMethodcreate_alibi_slope
(num_heads, dtype)
python/mlx/nn/layers/positional_encoding.py:136
↓ 1 callersFunctioncreate_nvfp4_scalars
mlx/backend/cuda/quantized/qqmm.cpp:55
↓ 1 callersFunctioncuda_kernel
mlx/backend/cuda/no_cuda.cpp:18
↓ 1 callersFunctioncuda_mem_loc
mlx/backend/cuda/allocator.cpp:92
↓ 1 callersFunctioncuda_toolkit_major_version
()
setup.py:17
↓ 1 callersFunctioncurrent_binary_dir
A helper function to find the location of the current binary on disk. The Metal library ("mlx_ext.mtllib"), should be in the same directory.
examples/extensions/axpby/axpby.cpp:22
↓ 1 callersMethoddeallocate
mlx/3rdparty/pocketfft.h:614
↓ 1 callersFunctiondebug_set_primitive_buffer_label
mlx/backend/metal/utils.h:35
↓ 1 callersFunctiondebug_set_stream_queue_label
mlx/backend/metal/utils.h:27
↓ 1 callersFunctiondeep_unary_compile
tests/compile_tests.cpp:575
↓ 1 callersFunctiondepthwise_conv_1D_gpu
mlx/backend/metal/conv.cpp:1032
↓ 1 callersFunctiondepthwise_conv_2D_gpu
mlx/backend/metal/conv.cpp:908
↓ 1 callersMethoddetach
mlx/distributed/utils.cpp:84
↓ 1 callersFunctiondfs
(start_node, node, path, visited)
python/mlx/_distributed_utils/config.py:313
↓ 1 callersFunctiondisable_compile
mlx/compile.cpp:1244
↓ 1 callersFunctiondisjoint
Check if two sets are disjoint
mlx/einsum.cpp:107
↓ 1 callersFunctiondispatch_all_types
mlx/dtype_utils.h:47
↓ 1 callersFunctiondispatch_conv_3D_gpu
mlx/backend/metal/conv.cpp:671
↓ 1 callersFunctiondispatch_int_types
mlx/dtype_utils.h:57
↓ 1 callersFunctiondispatch_quantize_dequantize
mlx/backend/cpu/quantized.cpp:1131
↓ 1 callersFunctiondispatch_slow_conv_1D
mlx/backend/cpu/conv.cpp:673
↓ 1 callersFunctiondispatch_slow_conv_2D
mlx/backend/cpu/conv.cpp:726
↓ 1 callersFunctiondispatch_slow_conv_3D
mlx/backend/cpu/conv.cpp:779
↓ 1 callersFunctiondtype_from_array_protocol
Dtype from array protocol type string
mlx/io/load.cpp:59
↓ 1 callersFunctiondtype_from_safetensor_str
mlx/io/safetensors.cpp:71
↓ 1 callersFunctiondtype_to_array_protocol
Array protocol typestring for Dtype
mlx/io/load.cpp:47
↓ 1 callersFunctiondtype_to_compute_type
mlx/backend/cuda/gemms/cublas_gemm.cpp:15
↓ 1 callersFunctiondtype_to_gguf_tensor_type
mlx/io/gguf.cpp:16
↓ 1 callersFunctiondtype_to_safetensor_str
mlx/io/safetensors.cpp:38
↓ 1 callersFunctionedge_pad
mlx/ops.cpp:1358
↓ 1 callersFunctioneigvals
mlx/linalg.cpp:549
↓ 1 callersFunctioneinsum_naive
mlx/einsum.cpp:537
↓ 1 callersFunctionenable_compile
mlx/compile.cpp:1248
↓ 1 callersMethodend_capture
mlx/backend/cuda/utils.cpp:68
↓ 1 callersMethodenqueue
mlx/scheduler.h:54
↓ 1 callersMethodenter
python/src/stream.cpp:28
↓ 1 callersFunctionerf
* Approximation to the error function. * Based on code from: * https://stackoverflow.com/questions/35148198/efficient-faithfully-rounded-implementat
mlx/backend/metal/kernels/erf.h:12
↓ 1 callersFunctionerfinv
mlx/backend/metal/kernels/erf.h:42
↓ 1 callersMethodeval_cpu
mlx/backend/cpu/eig.cpp:225
↓ 1 callersMethodeval_gpu
mlx/backend/metal/fft.cpp:795
↓ 1 callersMethodeval_gpu
mlx/backend/cuda/load.cpp:31
↓ 1 callersFunctionexplicit_gemm_conv_1D_cpu
mlx/backend/cpu/conv.cpp:856
↓ 1 callersFunctionexpm1f_scaled_unchecked
Compute exponential base e minus 1. Maximum ulp error = 0.997458 i = rint(a/log(2)), f = a-i*log(2). Then expm1(a) = 2**i * (expm1(f)+1) - 1. C
mlx/backend/metal/kernels/expm1f.h:43
↓ 1 callersFunctionextract_boolean_mask
python/src/indexing.cpp:928
↓ 1 callersFunctionextract_connectivity
(hosts, verbose)
python/mlx/_distributed_utils/config.py:200
↓ 1 callersFunctionextract_q4_0_data
Extracts (weight, scales, biases) from Q4_0 tensors. Data layout is: |16 bit scale|32 x 4bit weights|.
mlx/io/gguf_quants.cpp:32
↓ 1 callersFunctionextract_q4_1_data
Extracts (weight, scales, biases) from Q4_1 tensors. Data layout is: |16 bit scale|16 bit bias|32 x 4bit weights|.
mlx/io/gguf_quants.cpp:53
↓ 1 callersFunctionextract_q8_0_data
Extracts (weight, scales, biases) from Q8_0 tensors. Data layout is: |16 bit scale|32 x 8bit weights|.
mlx/io/gguf_quants.cpp:75
↓ 1 callersFunctionextract_qqmm_dims
mlx/ops.cpp:4589
↓ 1 callersFunctionextract_state
mlx/export.cpp:277
↓ 1 callersMethodextract_state
mlx/export.cpp:505
↓ 1 callersFunctionextract_tensor_data
mlx/io/gguf.cpp:59
↓ 1 callersFunctioneye
mlx/ops.cpp:346
↓ 1 callersFunctionfft
mlx/backend/metal/kernels/fft.h:180
↓ 1 callersFunctionfft2
Compute the two-dimensional Fourier Transform. */
mlx/fft.h:86
↓ 1 callersFunctionfft_scale_factor
mlx/fft.cpp:14
↓ 1 callersFunctionfftfreq
mlx/fft.cpp:313
↓ 1 callersMethodfinalize_safe
mlx/distributed/mpi/mpi.cpp:195
↓ 1 callersFunctionformat_uuid
mlx/backend/cuda/device_info.cpp:84
↓ 1 callersFunctionfour_step_fft
mlx/backend/metal/fft.cpp:474
↓ 1 callersFunctionfp_bs_qmm_dispatch
mlx/backend/cpu/quantized.cpp:869
↓ 1 callersFunctionfp_dequantize
mlx/ops.cpp:5048
↓ 1 callersFunctionfp_dequantize
mlx/backend/metal/kernels/fp_quantized.h:1847
↓ 1 callersFunctionfp_qmv
mlx/backend/metal/kernels/fp_quantized.h:1051
↓ 1 callersFunctionfp_quantize
mlx/ops.cpp:4801
↓ 1 callersFunctionfp_quantize_dequantize
mlx/backend/cpu/quantized.cpp:1094
↓ 1 callersFunctionfrexp_expf
mlx/backend/metal/kernels/cexpf.h:50
↓ 1 callersFunctiongather
mlx/backend/cpu/indexing.cpp:63
↓ 1 callersFunctiongather_axis
mlx/backend/metal/kernels/indexing/gather_axis.h:6
↓ 1 callersFunctiongather_front
mlx/backend/metal/kernels/indexing/gather_front.h:8
↓ 1 callersFunctiongather_mm
mlx/backend/metal/matmul.cpp:2250
↓ 1 callersFunctiongather_mm_rhs
mlx/backend/metal/matmul.cpp:1856
↓ 1 callersFunctiongather_mm_rhs_nax
mlx/backend/metal/matmul.cpp:1989
↓ 1 callersFunctiongather_qmm
mlx/backend/metal/quantized.cpp:869
← previousnext →1,301–1,400 of 4,946, ranked by callers