MCPcopy Create free account

hub / github.com/Luce-Org/lucebox-hub / functions

Functions2,878 in github.com/Luce-Org/lucebox-hub

⨍ Functions 2,878◇ Types & classes 420
↓ 17 callersFunctionemit_status
server/src/common/moe_expert_compute_ipc.cpp:275
↓ 17 callersMethodinit
server/test/test_kvflash.cpp:101
↓ 17 callersFunctionmake_moe_layer_desc
server/src/common/moe_hybrid_types_impl.h:31
↓ 17 callersMethodstep
Decode one token. Returns next token id.
optimizations/megakernel/model.py:274
↓ 17 callersMethodvalid
server/src/common/moe_hybrid_storage.h:54
↓ 16 callersMethodclose
(self)
server/scripts/laguna_pflash_niah.py:256
↓ 16 callersFunctionexpect
server/test/test_qwen35moe_expert_placement.cpp:11
↓ 16 callersMethodlayer_split_backend
server/src/placement/placement_config.h:39
↓ 16 callersMethodrefresh_mask
server/test/test_kvflash.cpp:165
↓ 16 callersFunctionresolve_anchor_params
Resolve anchor params for n_chunks (chunk_size=32). Tier: <1024 -> {2,8}; <2048 -> {4,16}; >=2048 -> {8,32}. env_* / legacy_* are the parsed env value
server/src/qwen3/anchor_params.h:12
↓ 16 callersFunctionrms_norm_input_f32
server/src/common/ggml_graph_precision.h:14
↓ 16 callersFunctionsplit_ms
server/src/deepseek4/deepseek4_layer_split_adapter.cpp:45
↓ 15 callersFunctionbuild_rms_norm
server/src/deepseek4/deepseek4_graph.cpp:419
↓ 15 callersFunctioncol
(k)
server/scripts/profile.py:714
↓ 15 callersFunctiondraft_feature_mirror_sync_range
server/src/common/dflash_feature_ring.cpp:320
↓ 15 callersFunctiongemma4_is_swa_layer
server/src/gemma4/gemma4_internal.h:134
↓ 15 callersFunctionget_u32
server/src/qwen3/qwen3_loader.cpp:99
↓ 15 callersMethodis_identity
True while every materialized chunk still sits in its identity block (chunk c in block c, nothing paged out). This is the layout contract identity-cop
server/src/common/kvflash_pager.h:319
↓ 15 callersFunctionplacement_backend_name
server/src/placement/placement_backend.h:15
↓ 14 callersFunctionalloc_tensor
Allocate a ggml tensor matching a safetensors entry. `gt_override`: if not GGML_TYPE_COUNT, use this as the ggml storage type instead of the safetens
server/src/draft/draft_safetensors_loader.cpp:368
↓ 14 callersFunctionbackend_ipc_checked_add_size
server/src/common/backend_ipc.h:47
↓ 14 callersFunctionbuild_causal_mask
Build a standard causal mask (lower-triangular) for n_tokens queries attending to kv_len keys starting at kv_start. kq_stride_pad: alignment for kv di
server/src/common/attn_masks.h:31
↓ 14 callersFunctionbuild_target_step
server/src/qwen35/graph_builders.cpp:260
↓ 14 callersFunctioncompute_score_range
Returns scoring layer range within [0, fwd_layer_limit).
server/src/common/score_range.h:17
↓ 14 callersFunctioncopy_tensor_from_file
server/src/qwen3/qwen3_loader.cpp:52
↓ 14 callersMethodfree
server/src/common/moe_hybrid_ffn_eval.cpp:2042
↓ 14 callersFunctionfree_deepseek4_weights
server/src/deepseek4/deepseek4_loader.cpp:886
↓ 14 callersFunctiongraph_tensor_f32
server/src/common/ggml_graph_precision.h:7
↓ 14 callersFunctionplace_marker_4gram
Place a marker 4-gram [FILLER, FILLER, MARKER, FILLER] at position pos.
server/test/test_anchor_transitive.cpp:22
↓ 14 callersFunctionraxIteratorIsInlineLeaf
server/src/server/rax.c:1557
↓ 14 callersFunctionread_tensor_bytes
server/tests/test_deepseek4_unit.cpp:535
↓ 14 callersMethodsave
server/src/server/disk_prefix_cache.cpp:424
↓ 14 callersMethodscore_chunks
server/src/qwen3/qwen3_kvflash_scorer.cpp:94
↓ 14 callersFunctionutf8_safe_len
Snap a byte offset back to a UTF-8 code-point boundary. Returns the largest position <= `pos` that doesn't split a multi-byte sequence. (Mirrors ds4_s
server/src/server/utf8_utils.h:17
↓ 13 callersFunction_mean
(xs)
server/scripts/bench_agent.py:441
↓ 13 callersFunction_quantize_matrix_nvfp4
(weight, group_size)
optimizations/megakernel/model_nvfp4.py:126
↓ 13 callersFunctionfree_laguna_target_cache
server/src/laguna/laguna_target_graph.cpp:315
↓ 13 callersFunctionfree_prefix_snapshot
server/src/qwen35/qwen35_target_graph.cpp:1651
↓ 13 callersMethodgenerate
── Generation ─────────────────────────────────────────────────── Run a full prefill + decode cycle. Backend owns the strategy (autoregressive, specul
server/src/common/model_backend.h:178
↓ 13 callersFunctionswiglu_maybe_clamped
server/src/common/moe_hybrid_ffn_eval.cpp:24
↓ 13 callersFunctiontarget_shard_ms
server/src/deepseek4/deepseek4_target_shard_ipc_daemon.cpp:84
↓ 13 callersMethodwith_token_callback
server/src/common/daemon_loop.cpp:53
↓ 12 callersFunctionbackend_ipc_payload_in_bounds
server/src/common/backend_ipc.h:55
↓ 12 callersFunctionclear_kv_env
server/test/test_kv_quant.cpp:23
↓ 12 callersFunctionhex
server/src/server/disk_prefix_cache.cpp:27
↓ 12 callersMethodinit
server/test/test_server_unit.cpp:2089
↓ 12 callersFunctionkvflash_pool_from_env
Pool size from DFLASH_KVFLASH for a backend with `cfg` protections: 0 = off; otherwise rounded to a 256 multiple, floored at min_pool_tokens(cfg) (evi
server/src/common/kvflash_pager.h:615
↓ 12 callersFunctionload_draft_safetensors
server/src/draft/draft_safetensors_loader.cpp:512
↓ 12 callersFunctionmake_moe_hybrid_config
server/src/common/moe_hybrid_types_impl.h:16
↓ 12 callersMethodobserve
server/src/common/moe_hybrid_routing_stats.cpp:55
↓ 12 callersFunctionraxIsInlineLeaf
Return 1 if the child at index 'childidx' of node 'n' is an inline leaf * value rather than a pointer to a child node. Inline leaves are stored * di
server/src/server/rax.c:189
↓ 12 callersFunctionraxSetData
Set the node auxiliary data to the specified pointer. */
server/src/server/rax.c:271
↓ 12 callersMethodreselect
Lookahead reselect (FlashMemory τ-step): rebuild the resident set as the top-pool chunks by score_hook among ALL known chunks (resident or host-backed
server/src/common/kvflash_pager.h:392
↓ 12 callersMethodstart
server/src/common/backend_ipc.cpp:99
↓ 12 callersMethodstop_hit
Check if a stop sequence was hit (signals caller to stop generation).
server/src/server/sse_emitter.h:87
↓ 12 callersMethodvalid
server/src/qwen35moe/qwen35moe_pipelined_decode.h:69
↓ 11 callersFunction_num
(row: dict, *names)
server/scripts/profile.py:161
↓ 11 callersFunctionapprox_eq
server/test/test_adaptive_keep_ratio.cpp:43
↓ 11 callersFunctionargmax_f32
server/src/common/io_utils.h:160
↓ 11 callersFunctioncompute_layer_ranges
server/src/common/layer_split_utils.cpp:13
↓ 11 callersMethoddisabled
server/src/server/tool_memory.h:32
↓ 11 callersFunctiondrafter_score_and_compress
server/src/qwen3/qwen3_drafter.cpp:681
↓ 11 callersMethodencode
(self, text: str)
server/tests/test_tokenizer.py:59
↓ 11 callersFunctionfmt
(value, nd: int = 2)
server/scripts/phase_split_dual_gpu.py:487
↓ 11 callersFunctionfree_deepseek4_snapshot
server/src/deepseek4/deepseek4_graph.cpp:3754
↓ 11 callersFunctionkvflash_fill_rows_and_masks
Slot-space step inputs for masked consumers: the K/V append row for each of this step's tokens, plus F32 causal (`mfull`) and sliding-window (`mswa`,
server/src/common/kvflash_pager.h:715
↓ 11 callersMethodlearn_layout
server/src/server/disk_prefix_cache.cpp:250
↓ 11 callersFunctionnormalize_system_for_cache
server/src/server/prompt_normalize.cpp:41
↓ 11 callersFunctionparse_disk_prefix_cache_policy
server/src/server/disk_prefix_cache.cpp:72
↓ 11 callersFunctionraxIteratorAddChars
Append characters at the current key string of the iterator 'it'. This * is a low level function used to implement the iterator, not callable by * t
server/src/server/rax.c:1529
↓ 11 callersFunctionrequire_value
server/src/ipc/backend_ipc_main.cpp:80
↓ 10 callersFunction_half_dtype
Return the 16-bit dtype matching the compiled kernel: bf16 on SM>=80, fp16 otherwise.
optimizations/megakernel/model.py:13
↓ 10 callersFunctioncompiled_placement_backend
server/src/placement/placement_backend.h:41
↓ 10 callersMethoddata
server/src/common/kvflash_qk.h:111
↓ 10 callersFunctionfree_layer_split_snapshot_backends
server/src/common/layer_split_utils.cpp:186
↓ 10 callersFunctionfree_pers
server/src/qwen3/qwen3_graph.cpp:109
↓ 10 callersFunctiong4_make_pers
server/src/gemma4/gemma4_graph.cpp:1178
↓ 10 callersFunctiongib
server/src/deepseek4/deepseek4_backend.cpp:36
↓ 10 callersMethodinit
server/src/common/moe_hybrid_stream.cpp:57
↓ 10 callersMethodis_hot
server/src/common/moe_hybrid_placement.cpp:28
↓ 10 callersMethodis_open
server/src/common/gguf_mmap.h:202
↓ 10 callersFunctionlaguna_rms_norm_mul
server/src/laguna/laguna_target_graph.cpp:351
↓ 10 callersFunctionload_draft_gguf
server/src/draft/draft_gguf_loader.cpp:92
↓ 10 callersFunctionload_target_gguf
server/src/qwen35/gguf_target_loader.cpp:281
↓ 10 callersFunctionmake_emitter_with_stops
server/test/test_server_unit.cpp:1181
↓ 10 callersFunctionmake_pers
server/src/qwen3/qwen3_graph.cpp:93
↓ 10 callersMethodmaybe_print
server/src/common/moe_expert_compute_ipc.cpp:116
↓ 10 callersFunctionmoe_expert_compute_checked_mul_size
server/src/common/moe_expert_compute_ipc.cpp:76
↓ 10 callersFunctionread_expert_slices
server/src/common/moe_hybrid_storage.cpp:72
↓ 10 callersFunctionread_expert_slices_from_mem
server/src/common/moe_hybrid_storage.cpp:93
↓ 10 callersFunctionresolve_kv_types
server/src/kv_quant.cpp:173
↓ 10 callersFunctionrotr32
server/src/common/gguf_inspect.cpp:62
↓ 10 callersFunctionselect_cuda_backend_precision_type_for_sm
server/src/common/backend_precision.cpp:162
↓ 10 callersFunctionselect_hip_activation_precision_type_for_arch
server/src/common/backend_precision.cpp:168
↓ 10 callersMethodsend_json
(self, status: int, obj: dict)
harness/clients/llamacpp_compat_proxy.py:383
↓ 10 callersMethodstats
server/src/server/prefix_cache.cpp:465
↓ 10 callersFunctionstep_adaptive_keep_ratio
server/src/server/adaptive_keep_ratio.h:31
↓ 10 callersFunctionstream_daemon_status
server/src/common/target_shard_ipc_daemon.cpp:30
↓ 10 callersMethodvalid
server/src/deepseek4/deepseek4_graph.cpp:237
↓ 10 callersFunctionwrite
(name, items, cap)
optimizations/spark/spark/extract_sessions.py:154
← previousnext →101–200 of 2,878, ranked by callers