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

↓ 8 callersFunctionfilter_tuple

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:345

↓ 8 callersFunctionflatten_to_tuple

src/cuda/flash_attn/cutlass/cute/algorithm/tuple_algorithms.hpp:544

↓ 8 callersMethodfree

src/cuda/flash_attn/cutlass/cute/arch/tmem_allocator_sm100.hpp:88

↓ 8 callersMethodget

Returns a pointer

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/regular_tile_access_iterator_tensor_op_sm80.h:540

↓ 8 callersMethodget_k

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/ell_predicated_tile_access_iterator.h:888

↓ 8 callersMethodget_mask

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/predicated_tile_access_iterator.h:286

↓ 8 callersFunctionget_run_dir

(default_type: str = "run")

python/krasis/run_paths.py:44

↓ 8 callersMethodget_sk_block_idx

Obtains calling linear threadblock index of the first block to work on the given tile

src/cuda/flash_attn/cutlass/cutlass/gemm/threadblock/threadblock_swizzle_streamk.h:732

↓ 8 callersFunctionget_tiled_cta_shape_mnl

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm100_tile_scheduler_stream_k.hpp:193

↓ 8 callersFunctionis_garbage

Heuristic check for garbage output.

tests/test_network.py:118

↓ 8 callersMethodis_source_needed

src/cuda/flash_attn/cutlass/cutlass/epilogue/collective/default_epilogue.hpp:148

↓ 8 callersFunctionispow2

Returns true if the argument is a power of 2

src/cuda/flash_attn/cutlass/cutlass/array.h:81

↓ 8 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_sm80.h:270

↓ 8 callersMethodload_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:1006

↓ 8 callersMethodload_with_byte_offset

Loads a fragment from memory with additional logical offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/tile_iterator_planar_complex.h:164

↓ 8 callersMethodload_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_sm70.h:1358

↓ 8 callersMethodload_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:3386

↓ 8 callersMethodload_with_pointer_offset

Loads a fragment from memory with additional logical offset

src/cuda/flash_attn/cutlass/cutlass/gemm/warp/tile_iterator_planar_complex.h:176

↓ 8 callersFunctionmake_int8_exception_candidate

( *, layer: int, tensor: str, group: int, benefit: float, cost: int, ready: bool =

tests/test_hqq_self_calibrate.py:157

↓ 8 callersFunctionmake_utccp_copy

src/cuda/flash_attn/cutlass/cute/atom/copy_traits_sm100.hpp:3747

↓ 8 callersFunctionmarlin_repack_int8

Repack a QuantizedInt8 into Marlin GPU INT8 format. Follows the same structure as INT4 marlin_repack but with: - 4 values packed per u32 (not 8) - IN

src/weights/marlin.rs:739

↓ 8 callersFunctionmma

src/cuda/flash_attn/cutlass/cutlass/conv/collective/sm100_implicit_gemm_umma_warpspecialized.hpp:838

↓ 8 callersMethodmma_init

src/cuda/flash_attn/cutlass/cutlass/gemm/collective/sm100_mma_warpspecialized.hpp:548

↓ 8 callersFunctionouter_partition

src/cuda/flash_attn/cutlass/cute/tensor_impl.hpp:1003

↓ 8 callersFunctionpipeline_init_arrive_relaxed

Used to guarantee that the Pipeline init is visible to all producers and consumer threadblocks in the cluster

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:1368

↓ 8 callersFunctionpipeline_init_wait

Synchronization call. Blocks until barriers are initialized in shared memory.

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:1355

↓ 8 callersFunctionquantize_int4_expert_calibrated

( weight_bf16: &[u16], rows: usize, cols: usize, group_size: usize, mode: ExpertInt4CalibM

src/weights/mod.rs:6629

↓ 8 callersMethodreduction_subtile_idx

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/static_tile_scheduler.hpp:79

↓ 8 callersFunctionround_up

src/cuda/flash_attn/cutlass/cute/int_tuple.hpp:356

↓ 8 callersMethodrun

Run all prompts and report results.

python/krasis/stress_test.py:372

↓ 8 callersFunctionsend_chat_request

Send a chat completion request. Returns (status_code, response_body).

tests/test_network.py:52

↓ 8 callersMethodset_iteration_index

Overrides the internal iteration index

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/regular_tile_access_iterator_tensor_op_sm80.h:525

↓ 8 callersMethodset_mask

Sets the predicate mask, overriding value stored in predicate iterator

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/predicated_tile_access_iterator.h:276

↓ 8 callersMethodset_tmem_offsets

src/cuda/flash_attn/cutlass/cutlass/gemm/collective/sm100_mma_warpspecialized.hpp:480

↓ 8 callersFunctionstore

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:525

↓ 8 callersFunctiontensormaps_replace_global_address

Replace address for the global tensor (to be done by single thread)

src/cuda/flash_attn/cutlass/cutlass/gemm/collective/sm100_mma_array_warpspecialized.hpp:769

↓ 8 callersMethodtransposed_problem

Returns arguments for the transposed problem

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/gemm_with_absmax.h:217

↓ 7 callersFunctionLayoutAwareConvert

src/cuda/flash_attn/cutlass/cutlass/detail/collective/mixed_input_utils.hpp:456

↓ 7 callersFunction_base_config

tests/test_launcher_matrix.py:91

↓ 7 callersFunction_component_weight_bytes

Weight bytes for per-component quant ("int4", "int8", "awq", or "bf16").

python/krasis/vram_budget.py:291

↓ 7 callersFunction_format_token_count

Format token count: 175000 -> '175K', 1200000 -> '1.2M'.

python/krasis/chat.py:598

↓ 7 callersFunction_load_source_tensor

(model_path: str, entry: Dict[str, Any])

python/krasis/hqq_self_calibrate.py:833

↓ 7 callersFunction_manifest_tensor_entries

(manifest: Dict[str, Any])

python/krasis/hqq_self_calibrate.py:651

↓ 7 callersMethod_moe_forward

MoE forward: routing + shared expert (GPU) + routed experts (CPU). For LatentMoE (Nemotron): wraps expert compute with latent projections.

python/krasis/layer.py:662

↓ 7 callersFunction_packet_brief

(packet: Optional[Dict[str, Any]])

tests/trace_diff.py:1046

↓ 7 callersFunction_parse_csv_ints

tests/test_hqq_rust_quantizer.py:1804

↓ 7 callersFunction_pkg_flag

Return ['-y'] if auto-approve, else [] so the package manager prompts.

python/krasis/setup.py:42

↓ 7 callersFunction_prefix_probe

( method: str, model: Any, tokenizer: Any, full_input_ids: Any, prefix_len: int, *,

tests/reference_backfill_topk.py:519

↓ 7 callersMethod_register_attn_weight

(self, shape: Tuple[int, int])

tests/decode_harness.py:151

↓ 7 callersFunction_section

Format a highlighted section header.

python/krasis/benchmark.py:37

↓ 7 callersFunction_source_contract_summary

(validations: Iterable[Dict[str, Any]])

python/krasis/hqq_self_calibrate.py:998

↓ 7 callersMethod_stream_attn_load

Copy one layer's attention weights from CPU pinned to GPU buffers. Sets the layer's attention attributes to point to the pre-allocated

python/krasis/model.py:4284

↓ 7 callersFunction_validate_source_contract_entry

( *, model_path: str, cache_dir: str, entry: Dict[str, Any], )

python/krasis/hqq_self_calibrate.py:945

↓ 7 callersFunction_warn

Print a warning line (yellow, indented).

python/krasis/server.py:126

↓ 7 callersFunctionabort_if_cuda_context_poisoned

(context: &str, err: &str)

src/server.rs:62

↓ 7 callersFunctionabs_for_integer

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:159

↓ 7 callersFunctionadd_candidate

(recipe: str, names: list[str], row_column_layout: str, current_resolver_status: str)

tests/hqq_attention_diff.py:2490

↓ 7 callersMethodadd_pointer_offset

Adds a pointer offset in units of Element

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/predicated_tile_iterator.h:1406

↓ 7 callersMethodarrive

Signal completion of Stage and move to the next stage (group_id) signals to (group_id+1)

src/cuda/flash_attn/cutlass/cutlass/pipeline/sm90_pipeline.hpp:1332

↓ 7 callersMethodbegin_loop

Start of subtile store iteration

src/cuda/flash_attn/cutlass/cutlass/epilogue/fusion/sm90_visitor_tma_warpspecialized.hpp:195

↓ 7 callersMethodbegin_sync_needed

Is a thread sync needed after begin(). Allows chaining async copies across multiple nodes

src/cuda/flash_attn/cutlass/cutlass/epilogue/fusion/sm90_visitor_tma_warpspecialized.hpp:185

↓ 7 callersFunctionbit_width

src/cuda/flash_attn/cutlass/cute/numeric/math.hpp:145

↓ 7 callersFunctionbuild_prompt

Build a prompt of approximately target_tokens length using Gutenberg text.

tests/archive/test_v2lite_profile.py:36

↓ 7 callersMethodclear_mask

Clears the predicate set efficiently

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/predicated_tile_iterator.h:1435

↓ 7 callersMethodcompute_epilogue

Returns whether the block assigned this work should compute the epilogue for the corresponding output tile. For the basic tile scheduler, this is alwa

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/sm100_tile_scheduler.hpp:517

↓ 7 callersFunctionconditional_return

src/cuda/flash_attn/cutlass/cute/numeric/integral_constant.hpp:432

↓ 7 callersFunctioncrd2idx

src/cuda/flash_attn/cutlass/cute/layout.hpp:677

↓ 7 callersFunctioncrd2idx

src/cuda/flash_attn/cutlass/cute/stride.hpp:100

↓ 7 callersMethoddata

Returns a pointer to the shared memory buffer

src/cuda/flash_attn/cutlass/cutlass/epilogue/threadblock/epilogue_base.h:175

↓ 7 callersFunctiondequantize_marlin_int8

Dequantize Marlin-repacked INT8 weights back to f32 for verification. Reverses the INT8 permutation and packing to recover the original [N, K] f32 va

src/weights/marlin.rs:1391

↓ 7 callersMethoddeterminant

Computes the determinant of a 2-by-2 matrix

src/cuda/flash_attn/cutlass/cutlass/matrix.h:3231

↓ 7 callersMethoddiv_cluster_size

Divides dividend by the cluster size

src/cuda/flash_attn/cutlass/cutlass/gemm/kernel/tile_scheduler_params.h:476

↓ 7 callersFunctionemit_contract_failure_trace

( context: str, reason: str, **fields: Any, )

tests/reference_contract.py:1205

↓ 7 callersMethodenable_mask

Clears the predicate set efficiently

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/predicated_tile_iterator.h:1441

↓ 7 callersMethodend_loop

End of subtile store iteration

src/cuda/flash_attn/cutlass/cutlass/epilogue/fusion/sm90_visitor_tma_warpspecialized.hpp:265

↓ 7 callersFunctionexplode_tuple

src/cuda/flash_attn/cutlass/cute/arch/util.hpp:284

↓ 7 callersFunctionfast_min

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:764

↓ 7 callersFunctionfast_tanh

src/cuda/flash_attn/cutlass/cutlass/fast_math.h:922

↓ 7 callersFunctionfp4_shift_A

src/cuda/flash_attn/cutlass/cute/atom/mma_traits_sm120.hpp:203

↓ 7 callersFunctionfp4_shift_B

src/cuda/flash_attn/cutlass/cute/atom/mma_traits_sm120.hpp:207

↓ 7 callersMethodgather

Copy from GPUs to pinned CPU memory (asynchronous).

python/krasis/model.py:614

↓ 7 callersFunctionget

Returns a pointer

src/cuda/flash_attn/cutlass/cutlass/conv/threadblock/predicated_scale_bias_vector_access_iterator.h:240

↓ 7 callersMethodget_log_tile

Calculates optimal swizzle width

src/cuda/flash_attn/cutlass/cutlass/gemm/threadblock/threadblock_swizzle.h:116

↓ 7 callersMethodget_mask

src/cuda/flash_attn/cutlass/cutlass/transform/threadblock/predicated_tile_iterator.h:1453

↓ 7 callersFunctionhas_single_bit

src/cuda/flash_attn/cutlass/cute/numeric/math.hpp:127

↓ 7 callersFunctionhqq_auto_promotion_policy

(attention_quant: str)

python/krasis/attention_backend.py:315

↓ 7 callersMethodis_fp8

True if this is an FP8 type that needs dequantization.

src/weights/safetensors_io.rs:68

↓ 7 callersMethodlayout

Returns the layout object

src/cuda/flash_attn/cutlass/cutlass/tensor_ref.h:285

↓ 7 callersFunctionmake_basis_like

src/cuda/flash_attn/cutlass/cute/numeric/arithmetic_tuple.hpp:333

↓ 7 callersFunctionmanifest_matches

(manifest: dict[str, object], contracts: dict[str, dict[str, object]])

scripts/build_sidecars.py:393

↓ 7 callersFunctionnormalize_hqq_attention_group_size

(group_size: Optional[int])

python/krasis/attention_backend.py:385

↓ 7 callersFunctionparse_hf_repo_id

Parse a Hugging Face model URL or repo id into `namespace/name`.

python/krasis/hf_downloader.py:220

↓ 7 callersFunctionprepare_lens

(cu_seqlens: torch.LongTensor)

src/cuda/fla/ops/utils/index.py:37

↓ 7 callersMethodpure

src/cuda/flash_attn/cutlass/cutlass/quaternion.h:201

↓ 7 callersFunctionread_manifest

(path: Path = MANIFEST_PATH)

scripts/build_sidecars.py:347

↓ 7 callersMethodread_u32

src/gguf.rs:208

↓ 7 callersMethodread_u64

src/gguf.rs:226

↓ 7 callersFunctionrepack_tiled_int4_packed

Repack INT4 packed weights from [K/8, N] to [N/TILE, K/8, TILE]. Last tile is zero-padded if N is not a multiple of TILE_N.

src/kernel/avx2.rs:1318

↓ 7 callersFunctionrepack_tiled_int8_packed

Repack INT8 data from [K, N] (i8 in u32 container) to [N/TILE, K, TILE] tiled.

src/kernel/avx2.rs:1690

↓ 7 callersFunctionrepack_tiled_scales

Repack INT4/INT8 scales from [K/gs, N] to [N/TILE, K/gs, TILE].

src/kernel/avx2.rs:1340

← previous next →601–700 of 12,380, ranked by callers