MCPcopy Create free account

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

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

↓ 6 callersFunctionbuild_layer_prefn_step
server/src/qwen35/graph_builders.cpp:98
↓ 6 callersFunctionbuild_moe_hybrid_storage_from_file
server/src/common/moe_hybrid_storage.cpp:400
↓ 6 callersFunctionbuild_props_body
Build the /props response body. Non-static so unit tests can call it directly (declared in http_server.h).
server/src/server/http_server.cpp:468
↓ 6 callersFunctionbuild_qwen35_graph
server/src/qwen35/qwen35_target_graph.cpp:1138
↓ 6 callersFunctioncheck_shape_1d
server/src/draft/draft_gguf_loader.cpp:67
↓ 6 callersFunctioncombine_activation_precision_types
server/src/common/backend_precision.cpp:184
↓ 6 callersFunctioncompute_target_shard_layer_split_plan
server/src/common/layer_split_utils.cpp:90
↓ 6 callersFunctioncopy_peer_async
server/src/common/peer_access.cpp:61
↓ 6 callersFunctioncreate_laguna_target_cache
server/src/laguna/laguna_target_graph.cpp:46
↓ 6 callersFunctioncreate_snapshot_backend
Select or create a backend for prefix cache snapshot storage. If the compute backend's default buffer type is host-accessible (unified memory — e.g.
server/src/common/snapshot_backend.h:33
↓ 6 callersFunctioncreate_target_cache_partial
server/src/qwen35/qwen35_target_graph.cpp:86
↓ 6 callersFunctiondrafter_arch_name
server/src/qwen3/qwen3_drafter.cpp:124
↓ 6 callersMethodemit_token_count
Total number of emit_token() calls observed so far. Used in tandem with first_content_token_index() to compute the content-token count without dependi
server/src/server/sse_emitter.h:124
↓ 6 callersFunctioneval_moe_hybrid_ffn_batched
Public entry. Workaround for a ggml-cuda/HIP defect: the MMQ mul_mat_id kernel illegal-accesses on gfx1151 when the per-layer hot expert stack is REDU
server/src/common/moe_hybrid_ffn_eval.cpp:1916
↓ 6 callersFunctionf2b
server/test/test_flashprefill_kernels.cpp:66
↓ 6 callersFunctionfmt
(value)
harness/clients/summarize_backend_pair.py:246
↓ 6 callersMethodfree
server/src/common/moe_hybrid_storage.cpp:42
↓ 6 callersFunctionfree_deepseek4_target_shards
server/src/deepseek4/deepseek4_target_shard_ipc_daemon.cpp:31
↓ 6 callersFunctionfree_gemma4_snapshot
server/src/gemma4/gemma4_loader.cpp:677
↓ 6 callersFunctionfree_laguna_layer_split_shards
server/src/laguna/laguna_layer_split_adapter.cpp:1234
↓ 6 callersFunctionfree_snapshot_backend
Free the snapshot backend if it was separately allocated. Safe to call with nullptr. Does NOT free the compute backend.
server/src/common/snapshot_backend.h:49
↓ 6 callersFunctiongemma4_has_kv
server/src/gemma4/gemma4_internal.h:138
↓ 6 callersFunctiongemma4_head_dim
server/src/gemma4/gemma4_internal.h:142
↓ 6 callersFunctiongemma4_n_head_kv
server/src/gemma4/gemma4_internal.h:146
↓ 6 callersFunctionget_slice_2d
server/src/delta_net_chunked.cpp:24
↓ 6 callersMethodhas_mmap
server/src/common/moe_hybrid_storage.h:179
↓ 6 callersFunctionhash_u64
server/src/common/moe_expert_compute.cpp:20
↓ 6 callersMethodis_eos
server/src/qwen35/qwen35_dflash_target.cpp:640
↓ 6 callersFunctionkv_type_name
server/src/kv_quant.cpp:48
↓ 6 callersFunctionkvflash_find_drafter
Locate the Qwen3-0.6B residency drafter: the explicit override (DFLASH_KVFLASH_DRAFTER, set from --prefill-drafter), then the well-known locations nex
server/src/common/kvflash_pager.h:686
↓ 6 callersFunctionlaguna_snapshot_free
server/src/laguna/laguna_target_graph.cpp:197
↓ 6 callersFunctionload_deepseek4_gguf_partial
server/src/deepseek4/deepseek4_loader.cpp:234
↓ 6 callersFunctionload_hc_weights_cpu
server/src/deepseek4/deepseek4_graph.cpp:2166
↓ 6 callersFunctionload_target_gguf_laguna_partial
server/src/laguna/laguna_target_loader.cpp:133
↓ 6 callersMethodlookup
server/src/server/tool_memory.cpp:60
↓ 6 callersFunctionmake_prompt
server/test/test_kvflash.cpp:211
↓ 6 callersMethodmatches
server/src/common/moe_hybrid_storage.cpp:191
↓ 6 callersFunctionmean
(xs)
server/scripts/profile.py:574
↓ 6 callersFunctionmigrate_prefill_cache
Attach rollback tensors to an existing prefill cache without touching the base tensors (KV, SSM, conv, target_feat) that prefill already populated. No
server/src/qwen35/qwen35_target_graph.cpp:352
↓ 6 callersFunctionmoe_hybrid_cache_swap_in
server/src/common/moe_hybrid_storage.cpp:611
↓ 6 callersFunctionparse_draft_residency_policy
server/src/placement/draft_residency.h:45
↓ 6 callersFunctionparse_int_list
server/src/ipc/backend_ipc_main.cpp:28
↓ 6 callersFunctionplacement_device_name
server/src/placement/placement_config.h:48
↓ 6 callersMethodpool_tokens
server/src/common/kvflash_pager.h:204
↓ 6 callersFunctionpositive_i32
server/src/draft/draft_safetensors_loader.cpp:217
↓ 6 callersFunctionraxStackPop
Pop an item from the stack, the function returns NULL if there are no * items to pop. */
server/src/server/rax.c:131
↓ 6 callersFunctionread_int32_file
server/src/common/io_utils.h:28
↓ 6 callersMethodrequest
Send a command and wait for its `ok ...`/`err ...` reply. Returns the reply line, or None if the daemon died or stalled past `timeout
optimizations/spark/spark/_daemon.py:74
↓ 6 callersFunctionresolve_pflash_drafter_placement
server/src/placement/pflash_placement.h:23
↓ 6 callersMethodresolved_payload_transport
server/src/common/backend_ipc.h:106
↓ 6 callersFunctionrestore_target_cache
server/src/qwen35/qwen35_target_graph.cpp:1570
↓ 6 callersMethodring_cap
server/src/common/dflash_draft_ipc.h:61
↓ 6 callersFunctionrun_daemon
server/src/common/daemon_loop.cpp:203
↓ 6 callersFunctionselect_inline_evict_victim
server/src/server/prefix_cache.cpp:152
↓ 6 callersFunctionsnapshot_target_cache
server/src/qwen35/qwen35_target_graph.cpp:1435
↓ 6 callersMethodsnapshot_used
server/src/qwen3/qwen3_backend.cpp:876
↓ 6 callersFunctiontokenize
(tokenizer, text: str, out_path: str)
server/scripts/run.py:79
↓ 6 callersFunctionunique_prompt
(text: str, label: str)
harness/client_test_runner.py:522
↓ 6 callersFunctionvram_used_now
server/test/test_kvflash.cpp:61
↓ 6 callersFunctionwrite_u32
Little-endian I/O helpers.
server/src/server/disk_prefix_cache.cpp:177
↓ 5 callersMethod_call
(self, obj: dict)
server/tests/test_tokenizer.py:50
↓ 5 callersFunction_int_env
(name, default)
optimizations/megakernel/setup.py:23
↓ 5 callersFunction_mean
(xs)
server/scripts/bench_server.py:377
↓ 5 callersFunction_prompt_path
(i: int, tokenizer_slug: str)
server/scripts/bench_he.py:226
↓ 5 callersMethodalloc_prefill_scratch
Allocate per-prefill scratch buffers for the chunk-parallel kernel. Buffers depend on S (sequence length); call once per distinct S.
optimizations/megakernel/model.py:262
↓ 5 callersFunctionapply_request_scope_override
server/src/server/disk_prefix_cache.cpp:116
↓ 5 callersMethodbos_id
server/src/server/tokenizer.h:50
↓ 5 callersFunctionbuild_batched_routed_graph
Build batched routed graph helper for batched prefill.
server/src/common/moe_hybrid_ffn_eval.cpp:446
↓ 5 callersFunctionbuild_cached_cold_graph
server/src/common/moe_hybrid_ffn_eval.cpp:617
↓ 5 callersFunctionbuild_cached_hot_graph
server/src/common/moe_hybrid_ffn_eval.cpp:501
↓ 5 callersMethodclose
server/src/qwen35/qwen35_target_shard_ipc.cpp:443
↓ 5 callersMethodcompute
server/src/common/moe_expert_compute_ipc.cpp:796
↓ 5 callersFunctiondeepseek4_is_eos_tok
server/src/deepseek4/deepseek4_internal.h:221
↓ 5 callersFunctionexpect
server/test/test_kvflash_pool_sizing.cpp:16
↓ 5 callersFunctionextract_draft_topk
server/src/common/ddtree.cpp:12
↓ 5 callersMethodfirst_content_token_index
Zero-based index of the first emit_token() call that produced CONTENT-mode output (i.e., the first token after the model's natural `</think>`). Return
server/src/server/sse_emitter.h:118
↓ 5 callersFunctionfree_deepseek4_cache
server/src/deepseek4/deepseek4_graph.cpp:3747
↓ 5 callersFunctiongpu_mem
()
harness/client_test_runner.py:994
↓ 5 callersMethodis_eos
server/src/laguna/laguna_dflash_target.cpp:515
↓ 5 callersFunctionis_newline
server/src/server/tokenizer.cpp:136
↓ 5 callersFunctionis_whitespace
server/src/server/tokenizer.cpp:127
↓ 5 callersFunctionkv_reservation_bytes_per_token
KV reservation bytes per token for a hybrid attention/SSM model. Single source of truth for both qwen35 (dense) and qwen35moe expert-placement budgeti
server/src/kv_quant.h:41
↓ 5 callersFunctionkvflash_qk_chunk_scores
Pure scoring: pooled_keys[c] = nullptr for chunks without pooled keys (their out[] entry is left at `missing_score`). Non-null entries hold n_layers *
server/src/common/kvflash_qk.h:45
↓ 5 callersFunctionlayer_split_kvflash_restore_history
server/src/common/layer_split_kvflash.h:67
↓ 5 callersMethodload_from_gguf
server/src/server/tokenizer.cpp:522
↓ 5 callersFunctionload_qwen3_drafter_model
server/src/qwen3/qwen3_loader.cpp:113
↓ 5 callersFunctionlong_prompt
()
harness/client_test_runner.py:512
↓ 5 callersFunctionmake_temp_file
Create a small temp file with known content; returns its path.
server/test/test_gguf_mmap.cpp:31
↓ 5 callersFunctionmoe_hybrid_expert_compute_ipc_batch_limit
server/src/common/moe_hybrid_ffn_eval.cpp:1196
↓ 5 callersFunctionmoe_hybrid_prefill_hot_sub_batch_limit
server/src/common/moe_hybrid_ffn_eval.cpp:1204
↓ 5 callersFunctionnonempty_env
server/src/common/moe_expert_compute.cpp:25
↓ 5 callersFunctionnow_unix
server/src/server/disk_prefix_cache.cpp:45
↓ 5 callersFunctionoverlaps
server/src/server/tool_parser.cpp:121
↓ 5 callersFunctionpflash_drafter_placement_used
server/src/placement/pflash_placement.h:18
↓ 5 callersMethodprefetch_cold_experts
server/src/common/moe_hybrid_stream.cpp:111
↓ 5 callersFunctionprint_usage
server/src/server/server_main.cpp:194
↓ 5 callersFunctionraxIteratorPrevStep
Like raxIteratorNextStep() but implements an iteration step moving * to the lexicographically previous element. The 'noup' option has a similar * ef
server/src/server/rax.c:1803
↓ 5 callersFunctionraxNewNode
Allocate a new non compressed node with the specified number of children. * If datafield is true, the allocation is made large enough to hold the *
server/src/server/rax.c:232
↓ 5 callersFunctionread_i32
(p: Path)
server/scripts/profile.py:291
↓ 5 callersFunctionread_layout_id_from_cache_dir
Helper: read layout_id from the first .dkv file found under base/.
server/test/test_server_unit.cpp:2845
← previousnext →301–400 of 2,878, ranked by callers