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

↓ 21 callersMethodvalid

Returns true if the current coordinate is within the filter tensor W

src/cuda/flash_attn/cutlass/cutlass/conv/threadblock/conv3d_fprop_filter_tile_access_iterator_optimized.h:227

↓ 20 callersFunction_fail

tests/test_network.py:40

↓ 20 callersFunction_to_device

Transfer tensor to target device, using CPU bounce if P2P is broken.

python/krasis/model.py:155

↓ 20 callersFunctionconst_max

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:758

↓ 20 callersFunctioncutlass_get_smem_pointer

CUTLASS helper to get SMEM pointer

src/cuda/flash_attn/cutlass/cutlass/arch/memory_sm75.h:63

↓ 20 callersFunctiondispatch_matmul_free

Free-function dispatch for quantized matmul (avoids &self borrow conflict).

src/decode.rs:1902

↓ 20 callersFunctionflatten

src/cuda/flash_attn/cutlass/cute/layout.hpp:529

↓ 20 callersMethodget

Returns a pointer

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/regular_scale_bias_vector_access_iterator.h:129

↓ 20 callersMethodget

src/cuda/flash_attn/cutlass/cute/pointer_base.hpp:157

↓ 20 callersFunctionload_with_byte_offset

Loads a fragment from memory with additional logical offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/mma_tensor_op_tile_iterator.h:385

↓ 20 callersFunctionlog_memory_usage

Log current memory usage from /proc/self/status and /proc/meminfo. Called from diagnostic logging points to track memory consumption.

src/syscheck.rs:227

↓ 20 callersMethodmake_fragment_B

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:172

↓ 20 callersFunctionmake_tiled_copy_impl

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:438

↓ 20 callersFunctionquantize_activation_int16

Quantize BF16 activation vector to INT16 with per-group scales. For each group: scale = max(|x|) / 32767, int16[i] = round(x[i] / scale). INT16 per-g

src/kernel/avx2.rs:234

↓ 20 callersFunctionreal

src/cuda/flash_attn/cutlass/cutlass/complex.h:395

↓ 20 callersMethodsize_bits

src/cuda/marlin/scalar_type.hpp:157

↓ 20 callersFunctionstore_with_pointer_offset

Stores a fragment to memory at the location pointed to by the iterator

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/mma_simt_tile_iterator.h:507

↓ 19 callersFunction_html_escape

tests/reference_test.py:85

↓ 19 callersMethodcopy

src/cuda/flash_attn/cutlass/cute/arch/copy.hpp:55

↓ 19 callersFunctioncp_async4

src/cuda/marlin/marlin_vendor_common.h:102

↓ 19 callersFunctionfast_divmod

* Find quotient and remainder using device-side intrinsics */

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:275

↓ 19 callersFunctionfast_sqrt

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:859

↓ 19 callersMethodget

src/cuda/flash_attn/cutlass/cute/container/array_subbyte.hpp:161

↓ 19 callersMethodget_flat_coord

src/cuda/flash_attn/cutlass/cute/layout.hpp:261

↓ 19 callersFunctioninfo

tests/validate_model.py:105

↓ 19 callersMethodinit_masks

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm100_pipeline.hpp:150

↓ 19 callersMethodisnan

Is NaN implementation

src/cuda/flash_attn/cutlass/cutlass/float8.h:190

↓ 19 callersFunctionmake_identity_layout

src/cuda/flash_attn/cutlass/cute/layout.hpp:483

↓ 19 callersFunctionmake_tiled_copy_A

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:452

↓ 19 callersFunctionprint

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:648

↓ 19 callersMethodtensor_info

Get metadata for a tensor.

src/weights/safetensors_io.rs:155

↓ 18 callersMethod_dense_mlp_tensor

Extract a BF16 tensor from a dense MLP weight value. Dense MLP weights are either plain BF16 tensors or (weight_int8, scale) tuples f

python/krasis/model.py:6595

↓ 18 callersFunction_format_metric_pct

(value: Optional[float], digits: int = 0)

tests/reference_test.py:777

↓ 18 callersFunction_parse_float

tests/trace_diff.py:205

↓ 18 callersMethodadd

(self, pos: int, row_hash: str, values: List[float])

python/krasis/hqq_self_calibrate.py:239

↓ 18 callersFunctionadd_pointer_offset

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/shared_load_iterator.h:67

↓ 18 callersFunctionblock_id_in_cluster

Returns the relative dim3 block rank local to the cluster.

src/cuda/flash_attn/cutlass/cute/arch/cluster_sm90.hpp:126

↓ 18 callersFunctionget_gmem_layout

src/cuda/flash_attn/cutlass/cutlass/detail/collective/mixed_input_utils.hpp:519

↓ 18 callersMethodinverse

src/cuda/flash_attn/cutlass/cutlass/layout/matrix.h:114

↓ 18 callersMethodisfinite

Is finite implementation

src/cuda/flash_attn/cutlass/cutlass/float8.h:176

↓ 18 callersFunctionload_prequantized_weight

Load a pre-quantized INT4 weight directly (compressed-tensors format). Reads weight_packed (I32), weight_scale (BF16), weight_shape (I32[2]).

src/weights/mod.rs:6488

↓ 18 callersMethodmake_fragment_A

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:146

↓ 18 callersMethodraw

Accesses raw internal state

src/cuda/flash_attn/cutlass/cutlass/half.h:463

↓ 18 callersMethodretile_S

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:416

↓ 18 callersFunctiontake

src/cuda/flash_attn/cutlass/cute/layout.hpp:507

↓ 17 callersFunctionblocked_product

src/cuda/flash_attn/cutlass/cute/layout.hpp:1748

↓ 17 callersMethodflush

Flush any remaining buffered tokens (end of stream).

src/server.rs:45

↓ 17 callersMethodis_source_needed

Determine if the source is needed. May return false if

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/epilogue_with_broadcast.h:123

↓ 17 callersFunctionload_hqq_attention_manifest

( model_path: str, cache_profile: Optional[str] = HQQ_CACHE_PROFILE_BASELINE, nbits: int = 4,

python/krasis/attention_backend.py:756

↓ 17 callersFunctionmake_zip_tensor

src/cuda/flash_attn/cutlass/cute/tensor_zip.hpp:157

↓ 17 callersMethodmoe_forward

Run MoE forward for a single token on one layer. Args: moe_layer_idx: 0-based MoE layer index (skipping dense layers) activation_bf16: BF16 activatio

src/moe.rs:1968

↓ 17 callersFunctionnorm

src/cuda/flash_attn/cutlass/cutlass/complex.h:481

↓ 17 callersFunctionpartition_fragment_C

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:621

↓ 17 callersFunctionreport_event

Record a named event with current VRAM snapshot. No-op if reporting is not enabled.

src/vram_monitor.rs:392

↓ 17 callersMethodresize

Changes the size of the view without affecting pointer or layout

src/cuda/flash_attn/cutlass/cutlass/tensor_view.h:170

↓ 17 callersFunctionstride

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:541

↓ 16 callersFunctionabs

src/cuda/flash_attn/cutlass/cute/numeric/math.hpp:65

↓ 16 callersFunctionalignment_for_swizzle

src/cuda/flash_attn/cutlass/cutlass/detail/layout.hpp:418

↓ 16 callersMethodclear

Efficient clear method

src/cuda/flash_attn/cutlass/cutlass/array.h:360

↓ 16 callersFunctionconj

src/cuda/flash_attn/cutlass/cutlass/complex.h:568

↓ 16 callersMethodfill

src/cuda/flash_attn/cutlass/cute/container/array.hpp:171

↓ 16 callersFunctionget_logical_ptr

src/cuda/flash_attn/cutlass/cutlass/detail/collective/mixed_input_utils.hpp:493

↓ 16 callersFunctionget_nonswizzle_portion

src/cuda/flash_attn/cutlass/cute/swizzle_layout.hpp:152

↓ 16 callersFunctionimag

src/cuda/flash_attn/cutlass/cutlass/complex.h:404

↓ 16 callersFunctionis_awq_scaled_tensor

Check if a tensor should have AWQ per-channel scaling applied.

python/krasis/awq_calibrate.py:1002

↓ 16 callersMethodload_init

src/cuda/flash_attn/cutlass/cutlass/gemm/collective/sm100_mma_warpspecialized.hpp:495

↓ 16 callersFunctionload_with_byte_offset

Loads a fragment from memory with additional logical offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/mma_complex_tensor_op_tile_iterator_sm80.h:253

↓ 16 callersFunctionmake_tiled_copy_B

src/cuda/flash_attn/cutlass/cute/atom/copy_atom.hpp:461

↓ 16 callersMethodmma

src/cuda/flash_attn/cutlass/cutlass/gemm/collective/sm100_mma_warpspecialized.hpp:648

↓ 16 callersMethodstride

Returns the stride of the layout

src/cuda/flash_attn/cutlass/cutlass/layout/tensor_op_multiplicand_sm80.h:127

↓ 16 callersFunctionsynclog_emit_tma_load

src/cuda/flash_attn/cutlass/cutlass/arch/synclog.hpp:813

↓ 16 callersFunctionthr_size

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:690

↓ 16 callersFunctiontile_shape

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:672

↓ 15 callersFunctionRematerializeBlockIdxZ

Helper to rematerialize block Idx. Reduces register liveness.

src/cuda/flash_attn/cutlass/cutlass/gemm/threadblock/index_remat.h:78

↓ 15 callersFunction_emit_real_model_timing

(payload: dict)

python/krasis/attention_backend.py:370

↓ 15 callersFunction_fused_add_rmsnorm

In-place fused add + RMSNorm. hidden = RMSNorm(hidden + residual, weight, eps) residual = hidden + residual (before norm)

python/krasis/layer.py:38

↓ 15 callersMethod_message_screen

(self, title: str, lines: List[str], *, wait: bool = True)

python/krasis/launcher.py:1779

↓ 15 callersFunction_pass

tests/test_network.py:36

↓ 15 callersFunction_utc_now

python/krasis/hqq_self_calibrate.py:125

↓ 15 callersFunctionassert_raises_contains

(fn, text: str)

tests/test_hqq_cache_profile.py:89

↓ 15 callersFunctioncanonical_lane_idx

Returns a lane index in the warp. The threads in warp may not be convergent

src/cuda/flash_attn/cutlass/cutlass/cutlass.h:114

↓ 15 callersFunctionceil_div

src/cuda/flash_attn/cutlass/cute/layout.hpp:1608

↓ 15 callersFunctioncheck_cuda_status

src/cuda/flash_attn/cutlass/cutlass/experimental/distributed/device/detail.hpp:46

↓ 15 callersFunctioncomposition

src/cuda/flash_attn/cutlass/cute/swizzle_layout.hpp:299

↓ 15 callersFunctionconstruct

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:626

↓ 15 callersMethodcount

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:198

↓ 15 callersFunctioncp_async_fence

Establishes an ordering w.r.t previously issued cp.async instructions. Does not block.

src/cuda/flash_attn/cutlass/cute/arch/copy_sm80.hpp:162

↓ 15 callersFunctiondiv_ceil

src/cuda/marlin/marlin_vendor_common.h:86

↓ 15 callersFunctionfold

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:387

↓ 15 callersFunctionlaunch_dependent_grids

Issuing the launch_dependents instruction hints a dependent kernel to launch earlier launch_dependents doesn't impact the functionality but the perfor

src/cuda/flash_attn/cutlass/cutlass/arch/grid_dependency_control.h:78

↓ 15 callersFunctionload_with_byte_offset

Loads a fragment from memory with additional logical offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/scale_bias_tile_iterator.h:258

↓ 15 callersFunctionload_with_pointer_offset

Loads a fragment from memory with additional logical offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/mma_tensor_op_tile_iterator.h:419

↓ 15 callersFunctionmax_common_vector

src/cuda/flash_attn/cutlass/cute/layout.hpp:1409

↓ 15 callersMethodnum_moe_layers

Number of MoE layers loaded.

src/moe.rs:2067

↓ 15 callersMethodpartition_fragment_B

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:560

↓ 15 callersFunctionpartition_shape_A

src/cuda/flash_attn/cutlass/cute/atom/mma_atom.hpp:634

↓ 15 callersFunctionpipeline_check_is_producer

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:64

↓ 15 callersFunctionprepend

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:816

↓ 15 callersMethodstore_tail

src/cuda/flash_attn/cutlass/cutlass/epilogue/collective/detail.hpp:476

↓ 15 callersFunctionstore_with_pointer_offset

Stores a fragment to memory with additional pointer offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/mma_tensor_op_tile_iterator.h:4636

← previous next →301–400 of 12,380, ranked by callers