MCPcopy Create free account

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

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

↓ 1 callersFunctionconvert_param_value
Convert a string value to its JSON-schema-typed equivalent.
server/src/server/tool_parser.cpp:71
↓ 1 callersFunctioncopy_activation_row_to_host
server/src/common/target_shard_ipc.cpp:71
↓ 1 callersFunctioncopy_feature_ring_range_to_host_f32
server/src/common/dflash_feature_ring.cpp:471
↓ 1 callersFunctioncopy_feature_to_f32
server/src/common/dflash_feature_ring.cpp:205
↓ 1 callersFunctioncopy_host_capture_slice_to_draft_ring
server/src/common/dflash_feature_ring.cpp:399
↓ 1 callersFunctioncopy_host_f32_to_feature_ring_range
server/src/common/dflash_feature_ring.cpp:497
↓ 1 callersFunctioncopy_metadata
(r: GGUFReader, w: GGUFWriter)
server/scripts/quantize_dflash_draft.py:37
↓ 1 callersFunctioncount_swa_layers
server/src/draft/draft_gguf_loader.cpp:59
↓ 1 callersFunctioncpu_hc_pre
server/src/deepseek4/deepseek4_graph.cpp:1979
↓ 1 callersFunctioncpu_hc_sinkhorn
server/src/deepseek4/deepseek4_graph.cpp:1842
↓ 1 callersFunctioncpu_top_p_dist
Analytic nucleus (top_p) distribution: the full softmax, restricted to the smallest descending-probability prefix whose cumulative mass reaches `top_p
server/test/test_gpu_sampler_cuda.cpp:83
↓ 1 callersFunctioncreate_backend
server/src/common/backend_factory.cpp:36
↓ 1 callersFunctioncreate_gemma4_target_feat
server/src/gemma4/gemma4_loader.cpp:637
↓ 1 callersFunctioncreate_laguna_target_feat
server/src/laguna/laguna_target_loader.cpp:711
↓ 1 callersFunctioncreate_qwen3_cache
server/src/qwen3/qwen3_backend.cpp:27
↓ 1 callersFunctioncross_device_peer_memcpy_ok
server/src/common/peer_access.cpp:50
↓ 1 callersFunctioncross_tok_compressed
Decode pflash-compressed drafter IDs to text -> retokenize as target IDs. Recovers chunk-boundary words by expanding each contiguous run of kept
server/scripts/laguna_pflash_niah.py:192
↓ 1 callersFunctioncurrent_device_id
server/src/common/backend_precision.cpp:71
↓ 1 callersMethodcurrent_last_token
server/src/common/layer_split_backend.h:76
↓ 1 callersMethoddecode_ar
server/test/test_server_unit.cpp:2106
↓ 1 callersMethoddecode_ar
server/src/qwen35/qwen35_layer_split_adapter.cpp:1271
↓ 1 callersMethoddecode_dflash
server/src/common/layer_split_backend.h:37
↓ 1 callersFunctiondecode_gpt2_bpe
server/src/server/tokenizer.cpp:722
↓ 1 callersFunctiondecode_steps
(decoder, first_token, num_steps, eos_token_id)
optimizations/megakernel/bench_pp_tg_nvfp4.py:126
↓ 1 callersFunctiondeepseek4_step_hybrid
server/src/deepseek4/deepseek4_graph.cpp:2187
↓ 1 callersMethoddefault_compress_drafter_path
server/src/common/layer_split_backend.h:53
↓ 1 callersFunctiondefault_paths
()
server/scripts/run.py:25
↓ 1 callersFunctiondequantize_feature_type_to_host_f32
server/src/common/dflash_feature_ring.cpp:77
↓ 1 callersFunctiondevice_props_for
server/src/common/backend_precision.cpp:83
↓ 1 callersFunctiondflash_cuda_copy_between_devices
server/src/cuda_cross_device_copy.cpp:27
↓ 1 callersFunctiondflash_draft_ipc_required_shared_bytes
server/src/common/dflash_draft_ipc.cpp:48
↓ 1 callersFunctiondiff_replies
(a, b)
server/scripts/quality_ab_simple.py:158
↓ 1 callersFunctiondisk_prefix_cache_mode_name
server/src/server/disk_prefix_cache.cpp:49
↓ 1 callersFunctiondomino_correct_greedy_chain
server/src/common/domino_head.cpp:114
↓ 1 callersFunctiondomino_correct_greedy_chain_fused
server/src/common/domino_head.cpp:182
↓ 1 callersFunctiondomino_step
server/src/common/domino_head.cpp:14
↓ 1 callersFunctiondraft_ipc_shared_bytes_from_env
server/src/common/dflash_draft_ipc.cpp:65
↓ 1 callersFunctiondraft_ipc_transport_from_env
server/src/common/dflash_draft_ipc.cpp:28
↓ 1 callersFunctiondrain_exact_fd
server/src/qwen35/qwen35_target_shard_ipc_daemon.cpp:93
↓ 1 callersFunctionds4_backend_is_gpu
server/src/deepseek4/deepseek4_graph.cpp:1224
↓ 1 callersFunctionds4_cold_backend_from_env
server/src/deepseek4/deepseek4_loader.cpp:669
↓ 1 callersFunctiondspark_markov_correct_greedy_chain
server/src/common/dspark_head.cpp:111
↓ 1 callersFunctiondspark_markov_correct_greedy_chain_fused
server/src/common/dspark_head.cpp:278
↓ 1 callersFunctiondspark_markov_project_topk
server/src/common/dspark_head.cpp:334
↓ 1 callersFunctiondspark_step
server/src/common/dspark_head.cpp:15
↓ 1 callersMethodembed_tokens
server/src/gemma4/gemma4_dflash_target.cpp:132
↓ 1 callersFunctionencode_gpt2_bpe
Convert a raw UTF-8 text piece to GPT-2 byte-encoded form for BPE lookup.
server/src/server/tokenizer.cpp:354
↓ 1 callersFunctionenforce_tier_invariants
Apply spec §3.5 monotone-ordering invariant. Clamps to monotone non-decreasing order and warns if the sidecar (or computed tiers) violate the invarian
server/src/server/model_card.cpp:139
↓ 1 callersMethodensure
server/src/deepseek4/deepseek4_graph.cpp:267
↓ 1 callersFunctionensure_ssm_snapshot
Allocate SSM/conv rollback snapshot tensors by mirroring the live recurrent state tensors' shapes. The MoE hybrid spec-decode path sets up its DeltaNe
server/src/qwen35/qwen35_target_graph.cpp:457
↓ 1 callersFunctionensure_staging
server/src/common/dflash_feature_ring.cpp:26
↓ 1 callersFunctionenv_flag_enabled
server/src/server/http_server.cpp:381
↓ 1 callersFunctionenv_float_or_default
server/src/qwen35moe/qwen35moe_backend.cpp:37
↓ 1 callersFunctionenv_int
server/src/deepseek4/deepseek4_layer_split_adapter.cpp:39
↓ 1 callersFunctionestimate_ds4_cache_bytes
server/src/deepseek4/deepseek4_backend.cpp:191
↓ 1 callersFunctioneval_moe_cold_experts_streaming
server/src/common/moe_hybrid_stream.cpp:269
↓ 1 callersFunctionexpected_pass
(case: dict[str, Any], text: str)
harness/benchmarks/generation_benchmark.py:136
↓ 1 callersFunctionextract_completion
Extract a complete Python solution from the model's reply. EvalPlus-canonical prompt asks for a self-contained script in a markdown code bloc
server/scripts/quality_humaneval_plus.py:137
↓ 1 callersFunctionextract_generated_text
(text: str)
harness/clients/summarize_backend_pair.py:132
↓ 1 callersFunctionextract_prefill_s
(resp: dict)
server/scripts/test_prefill_cache.py:62
↓ 1 callersFunctionextract_text
(response: dict[str, Any])
harness/benchmarks/generation_benchmark.py:177
↓ 1 callersFunctionextract_user_turns
Pull the first `limit` user-text messages from a Claude Code session.
server/scripts/bench_agent_loop.py:38
↓ 1 callersFunctionfamily_fallback
server/src/server/model_card.cpp:257
↓ 1 callersFunctionfeature_row_to_host_f32
server/src/common/dflash_feature_ring.cpp:107
↓ 1 callersFunctionfile_exists
server/src/server/model_card.cpp:54
↓ 1 callersFunctionfill_prefix_hot_placement
server/src/deepseek4/deepseek4_backend.cpp:218
↓ 1 callersFunctionfilter_child_stderr
(stderr_text)
optimizations/megakernel/bench_pp_tg_nvfp4.py:337
↓ 1 callersFunctionfind_first_seq_any
server/src/server/prefix_cache.cpp:80
↓ 1 callersFunctionfind_model_cards_dir
Find share/model_cards/ directory. Search order (spec §1 implementation note): (a) `repo_root_hint`/share/model_cards/ (if hint non-empty) (b) <binary
server/src/server/model_card.cpp:89
↓ 1 callersFunctionfind_run
(pair_dir: Path, backend: str)
harness/clients/summarize_backend_pair.py:213
↓ 1 callersFunctionfind_tensor
server/src/deepseek4/deepseek4_loader.cpp:139
↓ 1 callersFunctionfind_tool_properties
Find parameter schema properties for a function.
server/src/server/tool_parser.cpp:55
↓ 1 callersFunctionfind_tool_start
server/src/server/sse_emitter.cpp:37
↓ 1 callersFunctionfinish_hc_pre_from_mix
server/src/deepseek4/deepseek4_graph.cpp:1932
↓ 1 callersFunctionfirst_json_value
(text: str)
harness/clients/summarize_backend_pair.py:120
↓ 1 callersFunctionfirst_tool_parameter_name
server/src/server/http_server.cpp:404
↓ 1 callersFunctionfixed_slot_graphs_mode
server/src/common/moe_hybrid_ffn_eval.cpp:117
↓ 1 callersFunctionfixed_slot_max
server/src/common/moe_hybrid_ffn_eval.cpp:127
↓ 1 callersFunctionflash_prefill_forward_f16
server/src/flashprefill.cpp:613
↓ 1 callersFunctionflash_prefill_forward_q8
server/src/flashprefill_q8.cpp:31
↓ 1 callersFunctionfloat_to_bf16_bits_rne
server/src/hip_compat/cuda_bf16.h:31
↓ 1 callersFunctionfloat_to_bf16_bits_rne
Round-to-nearest-even float→bf16. Adds the round bit (bit 15) plus a sticky bit to the lower 16 bits, then shifts. Handles NaN by preserving the manti
server/hip_compat/cuda_bf16.h:53
↓ 1 callersFunctionfloat_to_bf16_bits_trunc
server/src/hip_compat/cuda_bf16.h:26
↓ 1 callersFunctionfloat_to_bf16_bits_trunc
Truncating float→bf16: drop lower 16 mantissa bits. Used by __float2bfloat16 (truncate-toward-zero semantics).
server/hip_compat/cuda_bf16.h:45
↓ 1 callersFunctionfloor_to_safe_boundary
server/src/server/disk_prefix_cache.cpp:148
↓ 1 callersMethodforward
server/src/qwen35/qwen35_target_shard_ipc.cpp:134
↓ 1 callersFunctionfp_threshold_loss
Loss that penalises false positives above a confidence threshold. Combines standard BCEWithLogitsLoss with a hinge penalty on non-top-4 exper
scripts/train_predictor.py:211
↓ 1 callersMethodfree_drafter
(self)
optimizations/pflash/pflash/dflash_client.py:226
↓ 1 callersMethodfree_drafter
server/src/qwen3/qwen3_backend.cpp:1024
↓ 1 callersMethodfree_drafter
server/src/common/layer_split_backend.cpp:239
↓ 1 callersFunctionfree_drafter_weights
server/src/qwen3/qwen3_drafter.cpp:254
↓ 1 callersFunctionfree_gemma4_target_feat
server/src/gemma4/gemma4_loader.cpp:628
↓ 1 callersFunctionfree_qwen3_cache
server/src/qwen3/qwen3_backend.cpp:61
↓ 1 callersMethodfull_stats
server/src/server/prefix_cache.cpp:472
↓ 1 callersFunctiong4_bsa_chunk_size
server/src/gemma4/gemma4_graph.cpp:1200
↓ 1 callersFunctiongemma4_layer_step_graph_destroy
server/src/gemma4/gemma4_graph.cpp:467
↓ 1 callersFunctiongemma4_project_hidden
server/src/gemma4/gemma4_graph.cpp:1104
↓ 1 callersFunctiongemma4_verify_batch
server/src/gemma4/gemma4_graph.cpp:901
↓ 1 callersFunctiongen_item_id
server/src/server/sse_emitter.cpp:51
↓ 1 callersFunctiongen_one
Generate a NIAH case whose tokenized length is within `tolerance` of `target_tokens` *and never exceeds it*, regardless of which tokenizer is
optimizations/pflash/tests/niah_gen.py:21
← previousnext →1,101–1,200 of 2,878, ranked by callers