qualia_core_db/inference/
mod.rs1pub mod inference_agent;
7pub use inference_agent as llm_agent; pub mod inference_awq;
9pub use inference_awq as llm_awq; #[cfg(not(target_arch = "wasm32"))]
11pub mod inference_bench;
12#[cfg(target_arch = "wasm32")]
13pub mod inference_bench_wasm;
14#[cfg(not(target_arch = "wasm32"))]
15pub mod kv_capture;
16pub mod kv_dict;
17#[cfg(not(target_arch = "wasm32"))]
18pub mod kv_dict_runtime;
19#[cfg(not(target_arch = "wasm32"))]
20pub use inference_bench as llm_bench; #[cfg(target_arch = "wasm32")]
22pub use inference_bench_wasm as llm_bench; pub mod inference_eval;
24pub use inference_eval as llm_eval; #[cfg(feature = "gpu-runtime")]
26pub mod inference_gpu_profiler;
27#[cfg(feature = "gpu-runtime")]
28pub use inference_gpu_profiler as llm_gpu_profiler; pub mod inference_kernel_parity;
30pub use inference_kernel_parity as llm_kernel_parity; pub mod agent;
32#[cfg(not(target_arch = "wasm32"))]
33pub mod ambient_orchestration;
34pub mod compute_universe;
35#[cfg(target_os = "windows")]
36pub mod directml_bridge;
37pub mod domino_gbnf;
38pub mod ggml_quants;
39pub mod gguf_sharder;
40#[cfg(any(target_os = "macos", target_os = "ios"))]
41pub mod metal_bridge;
42pub mod moe;
43pub mod neuro_symbolic_sieve;
44#[cfg(feature = "vision-onnx")]
45pub mod onnx_runtime;
46pub mod operator_converter;
47pub mod operator_package;
48pub mod operator_runtime;
49pub mod orchestrator;
50pub mod qwen;
51#[cfg(not(target_arch = "wasm32"))]
52pub mod residency_planner;
53pub mod resident_model;
54pub mod safetensor;
55pub mod safetensor_loader;
56pub mod semantic_culler;
57pub mod spatial_sieve;
58pub mod tensor_roles;
59pub mod ternary;
60#[cfg(feature = "gpu-runtime")]
61pub mod ternary_gpu;
62#[cfg(test)]
64pub mod toolkit_probe;
65pub mod topk;
66pub mod sampler;
68pub mod inference_modes;
70pub use inference_modes::{
71 active_inference_mode, apply_mode_toggles, bootstrap_inference_mode, fast_verify_html_default,
72 post_turn_verify_enabled, prefer_tensor_core_gemm, quant_graph_grounding_enabled,
73 rights_mode_enabled, sentinel_mid_decode_enabled, set_inference_mode, InferenceMode,
74};
75pub mod application_profile;
77pub use application_profile::{
78 active_application_profile, apply_application_profile, bootstrap_application_profile,
79 set_application_profile, ApplicationProfile,
80};
81#[cfg(not(target_arch = "wasm32"))]
83pub mod lab;
84#[cfg(not(target_arch = "wasm32"))]
86pub mod runtime;
87pub mod post_turn_verify;
89pub use post_turn_verify::{
90 maybe_verify_turn, return_html_as_text, verify_and_heal_turn, VerifiedTurn, VerifyCheck,
91};
92#[cfg(not(target_arch = "wasm32"))]
94pub mod inference_path_selector;
95#[cfg(not(target_arch = "wasm32"))]
96pub use inference_path_selector::{
97 apply_inference_path_plan, bootstrap_optimal_inference_path, format_path_plan,
98 last_inference_path_plan, path_auto_enabled, resolve_inference_path_plan, run_path_select_cli,
99 ComputeLane, InferencePathPlan, QuantProfile,
100};
101pub mod quant_graph_grounding;
103pub use quant_graph_grounding::{
104 export_fact_quins, fact_count, ground_generation, load_facts_from_tsv, lookup_capital_object,
105 maybe_ground_generation, register_capital_fact, register_fact, reset_fact_store_to_defaults,
106 seed_facts_from_bundled, GroundingFact, GroundingResult, CTX_GROUNDING, P_CAPITAL_OF,
107};
108pub mod qualia_hybrid;
110pub use qualia_hybrid::{
111 apply_graph_logit_bias, force_fact_tokens, graph_force_enabled, prepare_hybrid_decode,
112 propose_best_draft, propose_fact_draft, publish_graph_route_from_prompt,
113 publish_grounding_obligation, publish_prompt_query_tensor, GRAPH_LOGIT_BIAS,
114};
115#[cfg(all(not(target_arch = "wasm32"), feature = "cuda"))]
117pub mod cuda_lane;
118#[cfg(any(target_arch = "wasm32", not(feature = "cuda")))]
119pub mod cuda_lane_stub;
120#[cfg(all(not(target_arch = "wasm32"), feature = "cuda"))]
121pub use cuda_lane::{
122 cache_dense_weight, clear_weight_cache, dense_weight_cached, device_kv_ready,
123 ensure_device_kv_cache, preload_resident_blob, prepare_mega_pass_kernels,
124 q4k_device_weight_count, q4k_weight_resident, q8_0_gemv_oracle_into, try_cuda_batch_gemv,
125 try_cuda_batch_gemv_cached, try_cuda_batch_gemv_cached_only, try_cuda_mega_pass,
126 try_q4k_soa_attention_device, try_q4k_soa_ffn_block, try_q4k_soa_ffn_block_residual,
127 try_q4k_soa_fused_swiglu, try_q4k_soa_gemv, try_q4k_soa_qkv, try_q8_0_cuda_gemv,
128 warm_cuda_context, weight_cache_len, weight_fingerprint, MegaPassLayerDims,
129 MegaPassLayerWeights, MegaPassPlanView, MegaPassWeightLayout, MAX_DENSE_ELEMS,
130 Q8_0_BLOCK_BYTES, Q8_0_BLOCK_ELEMS,
131};
132#[cfg(any(target_arch = "wasm32", not(feature = "cuda")))]
133pub use cuda_lane_stub as cuda_lane;
134pub mod prompt_lookup;
136pub mod speculative_decode;
138pub use speculative_decode::{
139 DominoMasker, GrammarState, GrammarStateMachine, TokenTrie, MAX_TRIE_DEPTH,
140};
141pub mod semantic_skills;
143pub use semantic_skills::{
144 Scratchpad, ScratchpadEntry, ScratchpadError, SearchResult, StoredVector, TextEmbedder, Vector,
145 VectorStore, DEFAULT_SCRATCHPAD_TTL, EMBED_DIM, MAX_SCRATCHPAD_ENTRIES, MAX_VECTORS,
146};
147pub mod reranker;
149pub mod metal_lane;
151#[cfg(not(target_arch = "wasm32"))]
154pub mod paged_kv;
155#[cfg(feature = "gpu-runtime")]
156pub mod topk_gpu;
157#[cfg(any(not(target_arch = "wasm32"), feature = "wasm-scientific"))]
160pub mod sparse_cache;
161#[cfg(not(target_arch = "wasm32"))]
164pub mod thermal_wal;
165
166#[cfg(not(target_arch = "wasm32"))]
169pub mod thermal_telemetry;
170
171pub mod conditioning;
173
174pub mod conditioning_eval;
176
177pub mod conditioning_opt;
179
180#[cfg(feature = "gpu-runtime")]
182pub mod etalon_tensor_bridge;
183#[cfg(feature = "gpu-runtime")]
184pub use etalon_tensor_bridge::{
185 build_forward_arena_on, build_fused_attention_on, build_topk_pipeline_on,
186 online_softmax_rescale, AttentionParams, CapturedCudaMegaPass, CudaMegaPassExecutor,
187 DirectMlComputeLane, DmlGemmOp, DmlGemmParams, DmlLane, ForwardArena, ForwardMode,
188 FusedAttentionPipeline, KvCacheLayout, LogitMaskHeader, PagedKvBlockTable, QuantScheme,
189 TensorView, TopkParams, TopkReductionPipeline, FUSED_ATTENTION_WGSL, TOPK_REDUCTION_WGSL,
190};
191#[cfg(all(feature = "gpu-runtime", not(target_arch = "wasm32")))]
192pub use etalon_tensor_bridge::{
193 try_shared_forward_arena, try_shared_fused_attention, try_shared_topk_pipeline,
194};