Functions — github.com/brontoguana/krasis

CodeHub MCPcopy Create free account

hub / github.com/brontoguana/krasis / functions

Functions12,380 in github.com/brontoguana/krasis

⨍ Functions 12,380◇ Types & classes 7,782↳ Endpoints 10

↓ 3,046 callersMethodget

src/cuda/flash_attn/cutlass/cutlass/array_subbyte.h:148

↓ 2,648 callersFunctionprint

src/cuda/flash_attn/cutlass/cute/layout.hpp:1934

↓ 837 callersFunctionmake_tensor

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:396

↓ 776 callersMethodat

Accesses an element by coordinate

src/cuda/flash_attn/cutlass/cutlass/matrix.h:148

↓ 612 callersFunctionfloat

Converts to float

src/cuda/flash_attn/cutlass/cutlass/exmy_base.h:1040

↓ 577 callersFunctionsize

src/cuda/flash_attn/cutlass/cute/layout.hpp:605

↓ 560 callersFunctionmake_layout

src/cuda/flash_attn/cutlass/cute/layout.hpp:333

↓ 506 callersMethodcolumn

src/cuda/flash_attn/cutlass/cutlass/matrix.h:2094

↓ 472 callersMethodrow

src/cuda/flash_attn/cutlass/cutlass/matrix.h:229

↓ 440 callersFunctionrank

src/cuda/flash_attn/cutlass/cute/layout.hpp:614

↓ 433 callersFunctionmax

src/cuda/flash_attn/cutlass/cute/int_tuple.hpp:134

↓ 426 callersMethodlen

Length of the allocation in bytes.

↓ 414 callersFunctioncopy

src/cuda/flash_attn/cutlass/cute/algorithm/copy.hpp:171

↓ 371 callersMethoddata

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm90_tile_scheduler.hpp:64

↓ 361 callersMethodfloat

Converts to float

src/cuda/flash_attn/cutlass/cutlass/half.h:421

↓ 344 callersFunctionmin

src/cuda/flash_attn/cutlass/cute/int_tuple.hpp:154

↓ 336 callersFunctionshape

src/cuda/flash_attn/cutlass/cute/int_tuple.hpp:102

↓ 320 callersFunctionsynclog_emit_wgmma_reg_smem

src/cuda/flash_attn/cutlass/cutlass/arch/synclog.hpp:933

↓ 320 callersFunctionsynclog_emit_wgmma_smem_smem

src/cuda/flash_attn/cutlass/cutlass/arch/synclog.hpp:951

↓ 318 callersMethodstride

Returns the stride of the layout

src/cuda/flash_attn/cutlass/cutlass/layout/matrix.h:119

↓ 317 callersMethodstride

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:210

↓ 294 callersMethodas_ptr

Get the raw pointer to the allocated memory.

↓ 294 callersFunctionmake_smem_ptr

src/cuda/flash_attn/cutlass/cute/pointer.hpp:181

↓ 260 callersMethodindex

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:188

↓ 257 callersFunctionsum

src/cuda/flash_attn/cutlass/cute/int_tuple.hpp:286

↓ 255 callersMethodmax

Max representable value for this scalar type. (accounting for bias if there is one)

src/cuda/marlin/scalar_type.hpp:250

↓ 237 callersMethodpartition_S

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:396

↓ 220 callersFunctionprint

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:702

↓ 218 callersMethodimag

Set the imaginary part of the complex number

src/cuda/flash_attn/cutlass/cutlass/complex.h:317

↓ 212 callersMethodsynchronize

src/cuda/flash_attn/cutlass/cutlass/transform/collective/sm90_wgmma_transpose.hpp:124

↓ 208 callersMethodbegin

Before entry of the subtile load loop

src/cuda/flash_attn/cutlass/cutlass/epilogue/fusion/sm90_visitor_tma_warpspecialized.hpp:132

↓ 204 callersMethodreal

Set the real part of the complex number

src/cuda/flash_attn/cutlass/cutlass/complex.h:313

↓ 201 callersFunctionlocal_tile

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:1029

↓ 196 callersMethodload

src/cuda/flash_attn/cutlass/cutlass/block_striped.h:157

↓ 192 callersMethodnumel

Total number of elements.

src/weights/safetensors_io.rs:83

↓ 189 callersMethodload

Load all weights: GPU (streaming INT8) + CPU (Krasis INT4 experts). Args: gpu_only: If True, skip CPU expert cache and CPU decode

python/krasis/model.py:1171

↓ 187 callersMethodadd

Add a token. Returns the decoded text if the sequence is complete UTF-8, or an empty string if we're still buffering incomplete bytes.

src/server.rs:29

↓ 181 callersFunctionelect_one_sync

Elect one thread in the warp. The elected thread gets its predicate set to true, all others obtain false.

src/cuda/flash_attn/cutlass/cute/arch/cluster_sm90.hpp:180

↓ 181 callersMethodtensor

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:162

↓ 178 callersMethodget_slice

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:357

↓ 173 callersMethodpartition_D

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:406

↓ 168 callersMethodwrite

python/krasis/server.py:1237

↓ 167 callersFunctionfilter_zeros

src/cuda/flash_attn/cutlass/cute/layout.hpp:915

↓ 164 callersMethodadd_tile_offset

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/shared_load_iterator_mixed.h:185

↓ 161 callersFunctionto_underlying_arguments

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm100_tile_scheduler_stream_k.hpp:112

↓ 159 callersMethodsum

Returns the sum of all elements

src/cuda/flash_attn/cutlass/cutlass/coord.h:310

↓ 152 callersFunctionprint

src/cuda/flash_attn/cutlass/cutlass/conv/convnd_problem_shape.hpp:579

↓ 151 callersMethoddata

src/cuda/flash_attn/cutlass/cutlass/epilogue/collective/detail.hpp:170

↓ 148 callersFunctionadd_tile_offset

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/shared_load_iterator.h:155

↓ 147 callersFunctionElement

Unpacks an element from memory

src/cuda/flash_attn/cutlass/cutlass/subbyte_reference.h:187

↓ 147 callersMethodwith

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:76

↓ 145 callersMethodshape

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:198

↓ 144 callersFunctioncast_smem_ptr_to_uint

CUTE helper to cast SMEM pointer to unsigned

src/cuda/flash_attn/cutlass/cute/arch/util.hpp:91

↓ 144 callersMethoddecode

(self, token_ids: List[int], skip_special_tokens: bool = True)

python/krasis/tokenizer.py:147

↓ 143 callersFunctionget_workspace_size

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm90_tile_scheduler.hpp:139

↓ 136 callersFunctionbitcast

src/cuda/flash_attn/cutlass/cutlass/exmy_base.h:997

↓ 133 callersFunctionround_nearest

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:152

↓ 130 callersMethodadd

Elementwise add operator (1-by-2)

src/cuda/flash_attn/cutlass/cutlass/matrix.h:276

↓ 128 callersMethoddata

src/cuda/flash_attn/cutlass/cute/container/array.hpp:93

↓ 125 callersMethodclear

< Efficiently disables all accesses guarded by mask

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/predicated_tile_iterator_conv.h:193

↓ 125 callersFunctionidx2crd

src/cuda/flash_attn/cutlass/cute/stride.hpp:183

↓ 125 callersFunctionmake_gmem_ptr

src/cuda/flash_attn/cutlass/cute/pointer.hpp:118

↓ 122 callersFunctioncan_implement

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm100_tile_scheduler_stream_k.hpp:180

↓ 115 callersMethodas_mut_ptr

Get a mutable pointer to the allocated memory.

↓ 115 callersMethodstore

src/cuda/flash_attn/cutlass/cutlass/block_striped.h:186

↓ 113 callersMethoddot

Computes the dot product with anotherCoord object

src/cuda/flash_attn/cutlass/cutlass/coord.h:258

↓ 112 callersFunctiongemm

src/cuda/flash_attn/cutlass/cute/algorithm/gemm.hpp:68

↓ 111 callersMethodget_tma_tensor

src/cuda/flash_attn/cutlass/cute/atom/copy_traits_sm90_tma.hpp:145

↓ 110 callersFunctionsize

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:681

↓ 110 callersMethodw

Accesses the real part of the quaternion

src/cuda/flash_attn/cutlass/cutlass/quaternion.h:188

↓ 108 callersFunctionappend

src/cuda/flash_attn/cutlass/cute/layout_composed.hpp:308

↓ 108 callersMethodapply_chat_template

Format messages using the model's chat template and tokenize.

python/krasis/tokenizer.py:85

↓ 108 callersMethodset_iteration_index

Overrides the internal iteration index

src/cuda/flash_attn/cutlass/cutlass/conv/threadblock/conv2d_tile_iterator.h:108

↓ 107 callersMethoddata

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:180

↓ 107 callersMethodmn

Obtains a Coord<2> from GemmCoord

src/cuda/flash_attn/cutlass/cutlass/gemm_coord.h:162

↓ 106 callersFunctionceil_div

Returns the ceiling of (a / b)

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:220

↓ 106 callersMethodconsumer_release

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:1154

↓ 106 callersMethodload

src/cuda/flash_attn/cutlass/cutlass/gemm/collective/sm100_mma_warpspecialized.hpp:585

↓ 106 callersFunctiontransform

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:205

↓ 104 callersMethodconsumer_wait

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:1149

↓ 102 callersMethodis_valid

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/static_tile_scheduler.hpp:61

↓ 99 callersMethodget_tma_descriptor

Return TmaDescriptor/TensorMap

src/cuda/flash_attn/cutlass/cute/atom/copy_traits_sm90_tma.hpp:114

↓ 98 callersMethodget

Returns a pointer to the managed object or nullptr if no object is owned.

src/cuda/flash_attn/cutlass/cutlass/platform/platform.h:746

↓ 98 callersFunctionrun

src/cuda/flash_attn/cutlass/cutlass/transform/kernel/sm90_sparse_gemm_compressor.hpp:218

↓ 97 callersFunctioncoalesce

src/cuda/flash_attn/cutlass/cute/layout.hpp:865

↓ 97 callersFunctionraw_pointer_cast

src/cuda/flash_attn/cutlass/cute/container/array_subbyte.hpp:333

↓ 96 callersMethodget_thread_slice

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:367

↓ 96 callersMethodview

Returns a TensorRef pointing to the first element of the tensor.

src/cuda/flash_attn/cutlass/cutlass/thread/matrix.h:152

↓ 95 callersFunctionlayout

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:523

↓ 95 callersFunctionwarpgroup_fence_operand

src/cuda/flash_attn/cutlass/cute/arch/mma_sm90_gmma.hpp:86

↓ 93 callersFunctionrepeat_like

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:734

↓ 90 callersFunctionf32_to_bf16

src/weights/marlin.rs:25

↓ 90 callersMethodgenerate

Generate tokens autoregressively. Args: prompt_tokens: Input token IDs max_new_tokens: Maximum tokens to generate

python/krasis/model.py:6385

↓ 89 callersMethodclear_mask

< Efficiently disables all accesses guarded by mask

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/predicated_tile_iterator.h:1050

↓ 86 callersFunctionBase

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/predicated_tile_iterator_conv.h:171

↓ 86 callersMethodconsumer_try_wait

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:1140

↓ 84 callersFunctioncomposition

src/cuda/flash_attn/cutlass/cute/layout.hpp:1134

↓ 84 callersMethodfetch_next_work

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm100_tile_scheduler.hpp:454

↓ 84 callersFunctiontile_to_shape

src/cuda/flash_attn/cutlass/cute/layout.hpp:1792

↓ 83 callersFunctioncosize

src/cuda/flash_attn/cutlass/cute/layout.hpp:662

next →1–100 of 12,380, ranked by callers