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
↓ 4 callersFunctionreset_target_cache
server/src/qwen35/qwen35_target_graph.cpp:304
↓ 4 callersFunctionrestore_prompt_delta
server/src/common/restore_delta.h:9
↓ 4 callersFunctionrun
(cmd: list[str], timeout: int = 1200)
server/scripts/profile.py:124
↓ 4 callersFunctionrun_deepseek4_target_shard_ipc_daemon
server/src/deepseek4/deepseek4_target_shard_ipc_daemon.cpp:114
↓ 4 callersFunctionrun_layer_split_ar_decode
server/src/common/layer_split_runtime.cpp:5
↓ 4 callersFunctionrun_prefill
(decoder, ids_t, prompt_len, buffers, prefill_op)
optimizations/megakernel/bench_pp_tg_nvfp4.py:89
↓ 4 callersFunctionscan_and_force
server/src/qwen3/anchor_scan.cpp:18
↓ 4 callersMethodscratch_bytes
Total GPU scratch size.
server/src/common/moe_hybrid_stream.h:74
↓ 4 callersFunctionset_monitor_phase
(monitors: list[GpuMonitor], phase: str)
server/scripts/phase_split_dual_gpu.py:493
↓ 4 callersFunctionsha1_hash
server/src/common/sha1.h:15
↓ 4 callersFunctionsha256_compress
server/src/common/gguf_inspect.cpp:79
↓ 4 callersMethodshared_payload_capacity
server/src/common/backend_ipc.h:110
↓ 4 callersMethodshutdown
server/src/server/http_server.cpp:1042
↓ 4 callersFunctionsplit_csv
(value: str, choices: dict[str, Any])
harness/client_test_runner.py:208
↓ 4 callersFunctionst_shape1
server/src/draft/draft_safetensors_loader.cpp:184
↓ 4 callersMethodstart
server/src/qwen35/qwen35_target_shard_ipc.cpp:72
↓ 4 callersFunctionstddev
(xs)
server/scripts/profile.py:579
↓ 4 callersMethodstep_chunk
One 64-token chunk at logical [pos_base, pos_base+64). Allocates the chunk's block (evicting if needed), writes slot-mapped, masks resident slots + ca
server/test/test_kvflash.cpp:296
↓ 4 callersFunctionsynchronize_on_device
server/src/cuda_cross_device_copy.cpp:15
↓ 4 callersFunctiontext_check
(text: str | None, expect_substring: str | None = None)
harness/client_test_runner.py:526
↓ 4 callersFunctiontext_from_content
(content)
harness/clients/llamacpp_compat_proxy.py:22
↓ 4 callersFunctionutf8_decode
server/src/server/tokenizer.cpp:31
↓ 4 callersFunctionvalidate_device_placement
server/src/placement/placement_config.cpp:8
↓ 4 callersFunctionwrite_deepseek4_loader_fixture
server/tests/test_deepseek4_unit.cpp:94
↓ 4 callersMethodwrite_shared_payload
server/src/common/backend_ipc.cpp:290
↓ 4 callersFunctionwrite_u64
server/src/server/disk_prefix_cache.cpp:178
↓ 3 callersFunction_attach_prefill_fused_weights
(weights)
optimizations/megakernel/model_nvfp4.py:197
↓ 3 callersFunction_load_op
()
optimizations/megakernel/model.py:43
↓ 3 callersFunction_require_file
(path: str, label: str)
server/scripts/bench_llm.py:99
↓ 3 callersMethod_reset_runtime_state
(self)
optimizations/megakernel/model_nvfp4.py:630
↓ 3 callersMethod_run_prefill_bf16_nvfp4_lm
(self, token_ids: torch.Tensor, buffers)
optimizations/megakernel/model_nvfp4.py:637
↓ 3 callersMethod_skip
(self, name: str, reason: str)
server/tests/test_server_comprehensive.py:60
↓ 3 callersFunctionadd_ffn_telemetry
server/src/deepseek4/deepseek4_graph.cpp:85
↓ 3 callersMethodall_routed_are_hot
Fast check: are ALL routed experts in VRAM for this batch? selected_ids has n_slots entries (n_tokens * n_expert_used).
server/src/common/moe_hybrid_storage.h:111
↓ 3 callersFunctionalloc_prefill_buffers
(max_tokens)
optimizations/megakernel/bench_pp_tg_nvfp4.py:60
↓ 3 callersFunctionapi_format_name
Logging helpers shared by route_request() / worker_loop(). Kept static (file-scope) so they don't leak into the public ABI; the chat lifecycle logs th
server/src/server/http_server.cpp:368
↓ 3 callersFunctionapply_backend_visible_devices
(backend: str, *, visible_devices: str | N
server/scripts/placement/backend_device.py:17
↓ 3 callersFunctionarg_str
server/test/test_kvflash.cpp:371
↓ 3 callersFunctionbackend_ipc_mode_name
server/src/common/backend_ipc.cpp:27
↓ 3 callersFunctionbf16_to_f32
server/test/smoke_draft_graph.cpp:41
↓ 3 callersFunctionbinary_smoke
( path: Path, args: tuple[str, ...], timeout: int = 20, env: dict[str, str] | None = None, )
harness/client_test_runner.py:363
↓ 3 callersFunctionbuild
(target_chars: int)
optimizations/pflash/tests/niah_gen.py:44
↓ 3 callersFunctionbuild_clamped_swiglu
server/src/deepseek4/deepseek4_graph.cpp:427
↓ 3 callersFunctionbuild_ddtree
server/src/common/ddtree.cpp:64
↓ 3 callersFunctionbuild_delta_net_block
Gated DeltaNet block using the fused ggml_gated_delta_net primitive. Matches the semantics of llama.cpp's build_layer_attn_linear + build_delta_net_f
server/src/qwen35/qwen35_target_graph.cpp:740
↓ 3 callersFunctionbuild_draft_graph
server/src/draft/draft_graph.cpp:39
↓ 3 callersFunctionbuild_full_attn_block
Full-attention block (matches llama.cpp's build_layer_attn for qwen35) `cache_k` / `cache_v` are the persistent KV buffers for this layer (shape [hea
server/src/qwen35/qwen35_target_graph.cpp:530
↓ 3 callersFunctionbuild_gemma4_layer
Build one layer of the gemma4 graph.
server/src/gemma4/gemma4_graph.cpp:293
↓ 3 callersFunctionbuild_gemma4_layer_step
server/src/gemma4/gemma4_graph.cpp:475
↓ 3 callersFunctionbuild_laguna_layer_step
server/src/laguna/laguna_target_graph.cpp:955
↓ 3 callersFunctionbuild_layer_step
server/src/qwen35/graph_builders.cpp:12
↓ 3 callersFunctionbuild_moe_hybrid_storage_from_file_with_mmap
server/src/common/moe_hybrid_storage.cpp:691
↓ 3 callersFunctionbuild_moe_hybrid_swap_plan
server/src/common/moe_hybrid_swap_manager.cpp:8
↓ 3 callersFunctionbuild_qwen35_layer
server/src/qwen35/qwen35_target_graph.cpp:1322
↓ 3 callersFunctionbuild_qwen35_layer_prefn
server/src/qwen35/qwen35_target_graph.cpp:1369
↓ 3 callersFunctionbuild_tail_rope_3d
server/src/deepseek4/deepseek4_graph.cpp:447
↓ 3 callersFunctioncdiv
server/src/flashprefill.cpp:226
↓ 3 callersFunctioncfg
(**kw)
optimizations/spark/spark/bench.py:96
↓ 3 callersFunctionchat
(user_msg, max_tokens=8)
server/scripts/test_server_prefix_cache.py:89
↓ 3 callersFunctionchat
(max_tokens: int = 16)
server/scripts/test_full_compress_cache.py:152
↓ 3 callersMethodclose_map
server/src/gemma4/gemma4_loader.cpp:83
↓ 3 callersMethodcompress
Default typed compress: delegates to handle_compress via temp file + DaemonIO collector.
server/src/common/daemon_loop.cpp:65
↓ 3 callersFunctioncompute_gemma4_split_projection
server/src/gemma4/gemma4_graph.cpp:562
↓ 3 callersFunctioncompute_laguna_split_projection
server/src/laguna/laguna_target_graph.cpp:1019
↓ 3 callersFunctioncompute_target_split_projection
server/src/qwen35/layer_split_forward.cpp:81
↓ 3 callersFunctioncpu_hc_post
server/src/deepseek4/deepseek4_graph.cpp:2095
↓ 3 callersFunctioncpu_rms_norm
server/src/deepseek4/deepseek4_graph.cpp:1807
↓ 3 callersFunctioncpu_softmax
Analytic softmax over `logits` at temperature `temp`.
server/test/test_gpu_sampler_cuda.cpp:68
↓ 3 callersFunctiondeepseek4_step_layer_range
server/src/deepseek4/deepseek4_graph.cpp:2807
↓ 3 callersMethoddestroy
server/src/common/moe_hybrid_ffn_eval.cpp:2052
↓ 3 callersFunctiondflash_cmd
(cfg, prompt_bin: Path, out_bin: Path, n_gen: int)
server/scripts/profile.py:133
↓ 3 callersFunctiondflash_min_tokens_floor
server/src/qwen35/qwen35_backend.cpp:90
↓ 3 callersFunctiondraft_has_swa_layers
Check whether any layer in the draft is SWA.
server/src/common/dflash_draft_graph.cpp:19
↓ 3 callersFunctionemit_result
(result, json_result)
optimizations/megakernel/bench_pp_tg_nvfp4.py:184
↓ 3 callersFunctionenable_layer_split_peer_access
server/src/common/layer_split_utils.cpp:141
↓ 3 callersFunctionensure_pipelined_moe_expert_compute
server/src/qwen35moe/qwen35moe_pipelined_decode.cpp:215
↓ 3 callersFunctionenv_int_or_default
server/src/common/moe_hybrid_ffn_eval.cpp:42
↓ 3 callersFunctionenv_int_or_default
server/src/qwen35/qwen35_backend.cpp:83
↓ 3 callersFunctioneval_moe_hybrid_ffn_batched_core
server/src/common/moe_hybrid_ffn_eval.cpp:1250
↓ 3 callersFunctioneval_moe_hybrid_ffn_gpu_resident
server/src/common/moe_hybrid_ffn_eval.cpp:2124
↓ 3 callersFunctionextract
(gguf_path: str, out_path: str)
optimizations/spark/spark/tokenizer.py:16
↓ 3 callersMethodfill_slot_mask
F16 slot-validity mask for one query row: 0 for slots belonging to a resident chunk, -inf for free / paged-out blocks. `dst` must hold pool_tokens ent
server/src/common/kvflash_pager.h:378
↓ 3 callersFunctionfinish_hc_pre_from_mix_into
server/src/deepseek4/deepseek4_graph.cpp:1905
↓ 3 callersFunctionflash_prefill_forward
── Unified dispatch ────────────────────────────────────────────────────────── Picks the best available kernel at compile time + runtime buffer type:
server/src/flashprefill.h:100
↓ 3 callersFunctionfollow_verified_tree
server/src/common/ddtree.cpp:203
↓ 3 callersFunctionforward_qwen3_drafter_model
server/src/qwen3/qwen3_graph.cpp:221
↓ 3 callersFunctionfree_gemma4_cache
server/src/gemma4/gemma4_loader.cpp:615
↓ 3 callersFunctionfree_laguna_target_feat
server/src/laguna/laguna_target_loader.cpp:702
↓ 3 callersFunctionget_f32_or
server/src/qwen35/gguf_target_loader.cpp:185
↓ 3 callersFunctionget_json
(path: str, timeout: int = 5)
server/scripts/test_prefill_cache.py:57
↓ 3 callersFunctionget_prefill_op
(decoder)
optimizations/megakernel/bench_pp_tg_nvfp4.py:82
↓ 3 callersFunctionhermes_home
(work_dir: Path)
harness/client_test_runner.py:273
↓ 3 callersMethodhidden_size
server/src/qwen35/qwen35_layer_split_dflash_target.cpp:121
↓ 3 callersFunctionhost_f32_to_feature_row
server/src/common/dflash_feature_ring.cpp:88
↓ 3 callersMethodidentity_prefix_covers
True iff every chunk intersecting [0, n_tok) is resident in its identity block (block_of(c) == c). Expresses "the logical prefix [0, n_tok) is a conti
server/src/common/kvflash_pager.h:332
↓ 3 callersMethodinit
server/src/qwen35moe/qwen35moe_backend.cpp:80
↓ 3 callersFunctioninit_layer_split_shard_metas
server/src/common/layer_split_utils.cpp:117
↓ 3 callersFunctioninit_pipelined_decode_state
server/src/qwen35moe/qwen35moe_pipelined_decode.cpp:252
↓ 3 callersFunctioninstall_client
(work_dir: Path, spec: ClientSpec)
harness/client_test_runner.py:299
↓ 3 callersFunctionis_digit
server/src/server/tokenizer.cpp:100
← previousnext →501–600 of 2,878, ranked by callers