@qvac/llm-llamacpp 0.61.2 → 0.63.0

package/prebuilds/ios-x64-simulator/qvac__llm-llamacpp.bare.exports

@@ -7,7 +7,7 @@
     ],
     "current_versions": [
       {
-        "version": "0.61.2"
+        "version": "0.63"
       }
     ],
     "exported_symbols": [
@@ -77,9 +77,9 @@
             "__ZTSZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama9saveCacheEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlvE_",
             "__ZTSNSt3__110__function6__baseIFP10js_value_sRK22FinetuneTerminalResultEEE",
             "__ZTINSt3__16__loopIcEE",
-            "__ZTSZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama6cancelEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlvE_",
             "__ZTINSt3__120__shared_ptr_emplaceIN30qvac_lib_inference_addon_llama31JsFinetuneTerminalOutputHandlerENS_9allocatorIS2_EEEE",
             "__ZTINSt3__110__function6__funcIZN28qvac_lib_inference_addon_cpp9out_handl21JsLogMsgOutputHandlerC1EvEUlRKNS2_6Output6LogMsgEE_NS_9allocatorIS9_EEFP10js_value_sS8_EEE",
+            "__ZTSZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama6cancelEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlvE_",
             "__ZTSNSt3__110__function6__funcIZN28qvac_lib_inference_addon_cpp9out_handl26JsStringArrayOutputHandlerC1EvEUlRKNS_6vectorINS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEENS9_ISB_EEEEE_NS9_ISG_EEFP10js_value_sSF_EEE",
             "__ZTINSt3__110__function6__baseIFP10js_value_sRKN30qvac_lib_inference_addon_llama16BatchTokenOutputEEEE",
             "__ZTSZN28qvac_lib_inference_addon_cpp9out_handl27JsRuntimeStatsOutputHandlerC1EvEUlRKNSt3__16vectorINS2_4pairINS2_12basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEENS2_7variantIJdxEEEEENS8_ISD_EEEEE_",
@@ -109,16 +109,16 @@
             "__ZTSN28qvac_lib_inference_addon_cpp6Output6LogMsgE",
             "__ZTSNSt3__120__r_anchor_multilineIcEE",
             "__ZTIN28qvac_lib_inference_addon_cpp17MultiJobSchedulerE",
-            "__ZTSN28qvac_lib_inference_addon_cpp9out_handl17BaseOutputHandlerIP10js_value_sNSt3__112basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEEEE",
             "__ZTIN28qvac_lib_inference_addon_cpp13IJobSchedulerE",
+            "__ZTSN28qvac_lib_inference_addon_cpp9out_handl17BaseOutputHandlerIP10js_value_sNSt3__112basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEEEE",
             "__ZTSNSt3__110__function6__funcIZN28qvac_lib_inference_addon_cpp9out_handl21JsLogMsgOutputHandlerC1EvEUlRKNS2_6Output6LogMsgEE_NS_9allocatorIS9_EEFP10js_value_sS8_EEE",
             "__ZTSN28qvac_lib_inference_addon_cpp9out_handl17BaseOutputHandlerIP10js_value_sNS_6Output6LogMsgEEE",
             "__ZTINSt3__110__function6__baseIFN22load_fit_normalization15SelectedBackendEN17backend_selection11BackendTypeERKNS_8optionalINS_7variantIJiNS4_11MainGpuTypeEEEEEERK13ModelMetaDatabRKNS_6vectorINS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEENSK_ISM_EEEEEEE",
             "__ZTSNSt3__110__function6__baseIFvjimEEE",
             "__ZTINSt3__115basic_stringbufIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
             "__ZTSNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
-            "__ZTIN30qvac_lib_inference_addon_llama16BatchTokenOutputE",
             "__ZTINSt3__110__function6__funcIZN28qvac_lib_inference_addon_cpp9out_handl21JsStringOutputHandlerC1EvEUlRKNS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEEE_NS8_ISD_EEFP10js_value_sSC_EEE",
+            "__ZTIN30qvac_lib_inference_addon_llama16BatchTokenOutputE",
             "__ZTINSt3__110__function6__baseIFvR13common_paramsRKNS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEEEEE",
             "__ZTSN28qvac_lib_inference_addon_cpp16OutputCallBackJsE",
             "__ZTINSt3__111__end_stateIcEE",
@@ -280,6 +280,7 @@
             "__ZNSt3__19__any_imp17__unique_typeinfoINS_6vectorINS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEENS6_IS8_EEEEE4__idE",
             "__ZTIPFvP13llama_contextE",
             "__ZTIN28qvac_lib_inference_addon_cpp9out_handl21JsLogMsgOutputHandlerE",
+            "__ZTINSt3__110__function6__baseIFN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12DraftOutcomeEjiiEEE",
             "__ZTIZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama9saveCacheEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlvE_",
             "__ZTINSt3__110__function6__baseIFP10js_value_sRKNS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEEEEE",
             "__ZTINSt3__110__function6__baseIFvvEEE",
@@ -333,10 +334,10 @@
             "__ZTINSt3__120__shared_ptr_emplaceIN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler10BatchGroupENS_9allocatorIS4_EEEE",
             "__ZTSNSt3__110__function6__baseIFN30qvac_lib_inference_addon_llama9FitResultEvEEE",
             "__ZTINSt3__120__shared_ptr_emplaceIN24llama_finetuning_helpers23TrainingCheckpointStateENS_9allocatorIS2_EEEE",
-            "__ZTINSt3__123__match_any_but_newlineIcEE",
-            "__ZTSNSt3__16__loopIcEE",
             "__ZTIN28qvac_lib_inference_addon_cpp9out_handl17BaseOutputHandlerIP10js_value_sNSt3__16vectorINS4_12basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEENS9_ISB_EEEEEE",
+            "__ZTINSt3__123__match_any_but_newlineIcEE",
             "__ZTSNSt3__110__function6__baseIFvmjEEE",
+            "__ZTSNSt3__16__loopIcEE",
             "__ZTI11BlobsStreamIcE",
             "__ZTIZN28qvac_lib_inference_addon_cpp9out_handl20JsErrorOutputHandlerC1EvEUlRKNS_6Output5ErrorEE_",
             "__ZTINSt3__110__function6__baseIFP10js_value_sRKN24llama_finetuning_helpers21FinetuneProgressStatsEEEE",
@@ -438,8 +439,8 @@
             "__ZTINSt3__110__function6__funcIZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama9saveCacheEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS8_OT_EUlS8_RKNS_9monostateEE_NS_9allocatorISJ_EEFSD_S8_SI_EEE",
             "__ZN30qvac_lib_inference_addon_llama5cache24CHECKPOINT_SECTION_MAGICE",
             "__ZTINSt3__110__function6__funcIZN28qvac_lib_inference_addon_cpp9out_handl26JsStringArrayOutputHandlerC1EvEUlRKNS_6vectorINS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEENS9_ISB_EEEEE_NS9_ISG_EEFP10js_value_sSF_EEE",
-            "__ZNSt3__19__any_imp17__unique_typeinfoIN28qvac_lib_inference_addon_cpp6Output6LogMsgEE4__idE",
             "__ZTIN13BorrowablePtrINSt3__115basic_streambufIcNS0_11char_traitsIcEEEEE7DeleterE",
+            "__ZNSt3__19__any_imp17__unique_typeinfoIN28qvac_lib_inference_addon_cpp6Output6LogMsgEE4__idE",
             "__ZTSNSt3__110__function6__baseIFNS_9monostateEvEEE",
             "__ZNSt3__19__any_imp17__unique_typeinfoI22FinetuneTerminalResultE4__idE",
             "__ZTINSt3__120__shared_ptr_emplaceINS_6atomicIbEENS_9allocatorIS2_EEEE",
@@ -477,8 +478,8 @@
             "__ZTSN28qvac_lib_inference_addon_cpp9out_handl19JsBaseOutputHandlerINSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEEEE",
             "__ZTSNSt3__110__function6__baseIFvvEEE",
             "__ZTSNSt3__110__function6__baseIFbRKN30qvac_lib_inference_addon_llama8batching7RequestEEEE",
-            "__ZTINSt3__121__empty_non_own_stateIcEE",
             "__ZTINSt3__110__function6__baseIFP16ggml_backend_regPKcEEE",
+            "__ZTINSt3__121__empty_non_own_stateIcEE",
             "__ZTSNSt3__110__function6__baseIFNS_6vectorINS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEENS6_IS8_EEEERKS8_EEE",
             "__ZTSNSt3__110__function6__funcIZN30qvac_lib_inference_addon_llama25JsBatchTokenOutputHandlerC1EvEUlRKNS2_16BatchTokenOutputEE_NS_9allocatorIS7_EEFP10js_value_sS6_EEE",
             "__ZTIFiP13llama_context11llama_batchE",
@@ -491,7 +492,6 @@
             "__ZTINSt3__110__function6__funcIZN30qvac_lib_inference_addon_llama31JsFinetuneTerminalOutputHandlerC1EvEUlRK22FinetuneTerminalResultE_NS_9allocatorIS7_EEFP10js_value_sS6_EEE",
             "__ZTIN28qvac_lib_inference_addon_cpp6Output6LogMsgE",
             "__ZTIN28qvac_lib_inference_addon_cpp9out_handl19JsBaseOutputHandlerINS_6Output6LogMsgEEE",
-            "__ZTSZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama12discardCacheEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlS5_RKNSt3__19monostateEE_",
             "__ZTSN28qvac_lib_inference_addon_cpp9out_handl26JsStringArrayOutputHandlerE",
             "__ZTIN30qvac_lib_inference_addon_llama31JsFinetuneTerminalOutputHandlerE",
             "__ZTIZN28qvac_lib_inference_addon_cpp9out_handl26JsStringArrayOutputHandlerC1EvEUlRKNSt3__16vectorINS2_12basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEENS7_IS9_EEEEE_",
@@ -499,6 +499,7 @@
             "__ZTSNSt3__113basic_filebufIcNS_11char_traitsIcEEEE",
             "__ZTIN28qvac_lib_inference_addon_cpp5model14IModelJobStatsE",
             "__ZTSNSt3__115__has_one_stateIcEE",
+            "__ZTSZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama12discardCacheEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlS5_RKNSt3__19monostateEE_",
             "__ZTSNSt3__110__function6__baseIFP10js_value_sRKN30qvac_lib_inference_addon_llama16BatchTokenOutputEEEE",
             "__ZTIN28qvac_lib_inference_addon_cpp9out_handl19JsBaseOutputHandlerINSt3__16vectorINS2_4pairINS2_12basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEENS2_7variantIJdxEEEEENS8_ISD_EEEEEE",
             "__ZTINSt3__117__repeat_one_loopIcEE",
@@ -507,8 +508,8 @@
             "__ZTINSt3__110__function6__baseIFiR14SequenceDrivermiEEE",
             "__ZTINSt3__16vectorINS_4pairINS_12basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEENS_7variantIJdxEEEEENS5_ISA_EEEE",
             "__ZTSN28qvac_lib_inference_addon_cpp9out_handl21JsLogMsgOutputHandlerE",
-            "__ZTINSt3__120__shared_ptr_emplaceIN10LlamaModel15ReloadableStateENS_9allocatorIS2_EEEE",
             "__ZTSFvP13llama_contextE",
+            "__ZTINSt3__120__shared_ptr_emplaceIN10LlamaModel15ReloadableStateENS_9allocatorIS2_EEEE",
             "__ZTSZN28qvac_lib_inference_addon_cpp17makeQueueCallbackINSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEEEENS1_8functionIFvRKT_EEENS1_10shared_ptrINS_11OutputQueueEEENS1_13shared_futureIyEEEUlRKS7_E_",
             "__ZTINSt3__120__shared_ptr_emplaceIN30qvac_lib_inference_addon_llama31JsFinetuneProgressOutputHandlerENS_9allocatorIS2_EEEE",
             "__ZTIN28qvac_lib_inference_addon_cpp9out_handl19JsBaseOutputHandlerINS_6Output5ErrorEEE",
@@ -520,6 +521,7 @@
             "__ZTSNSt3__110__function6__funcIPFP16ggml_backend_regPKcENS_9allocatorIS7_EES6_EE",
             "__ZTSPFP16ggml_backend_regPKcE",
             "__ZTSNSt3__110shared_ptrINS_13__empty_stateIcEEE27__shared_ptr_default_deleteIS2_S2_EE",
+            "__ZTSNSt3__110__function6__baseIFN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12DraftOutcomeEjiiEEE",
             "__ZTSNSt3__16__nodeIcEE",
             "__ZTSZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask3runIZN30qvac_lib_inference_addon_llama9saveCacheEP8js_env_sP18js_callback_info_sEUlvE_EEP10js_value_sS5_OT_EUlS5_RKNSt3__19monostateEE_",
             "__ZTINSt3__110__function6__funcIPFiP13llama_context11llama_batchENS_9allocatorIS6_EEFiS3_RS4_EEE",
@@ -534,6 +536,7 @@
           "global": [
             "__ZNK14MtmdLlmContext20getVisionEncodeTilesEv",
             "__ZN19LlamaBackendsHandleC2EOS_",
+            "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher13verifyDraftedERKNSt3__18functionIFNS1_12DraftOutcomeEjiiEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher10addRequestEONSt3__16vectorIiNS2_9allocatorIiEEEERj",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher19sampleAndAppendIdleERKNSt3__18functionIFijiEEE",
             "__ZN10LlamaModel30discardFinetuneCancelSaveModesERKNSt3__16vectorIyNS0_9allocatorIyEEEE",
@@ -548,10 +551,12 @@
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotD1Ev",
             "__ZThn48_N10LlamaModelD0Ev",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler11clearLockedEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence12prepareDraftEiiiRKNSt3__16vectorIiNS2_9allocatorIiEEEE",
             "__ZN14LlamaFinetuner8finetuneERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsEPN22FinetuneTerminalResult5StatsENSt3__18functionIFvRKN24llama_finetuning_helpers21FinetuneProgressStatsEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request20isOptHasTokensToFeedERKNSt3__18optionalIS1_EE",
             "__ZN12CacheManager14writeCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils24isDeepSeekV4ArchitectureENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence6verifyEP14common_sampleri",
             "__ZN30qvac_lib_inference_addon_llama5utils24selectReasoningTagSourceERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEES9_RKNS1_8optionalINS0_13ReasoningTagsEEE",
             "__ZN12CacheManager14flushForUnloadEv",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request20isAwaitingCheckpointEv",
@@ -566,6 +571,7 @@
             "__ZN12CacheManager7discardERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils24initializeReasoningStateEP13llama_contextRNS0_14ReasoningStateENS0_13ReasoningTagsE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15StepUnlockGuardC2ERS1_PNSt3__111unique_lockINS4_5mutexEEE",
+            "__ZN30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot21recordSpeculativeStepExNSt3__16chrono8durationIxNS2_5ratioILl1ELl1000000000EEEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler9numActiveEv",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12markFinishedEjNS0_10StopReasonE",
             "__ZThn48_N14TextLlmContext13onSequenceEndERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
@@ -581,6 +587,7 @@
             "__ZN14LlamaFinetuner23initializeCheckpointingERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsEP18llama_adapter_loraPN24llama_finetuning_helpers20LoraLrSchedulerStateE",
             "__ZN14TextLlmContext12tokenizeChatERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERNS1_IiNS3_IiEEEEb",
             "__ZN14MtmdLlmContext17removeLastNTokensEi",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeD2Ev",
             "__ZN30qvac_lib_inference_addon_llama5utils17resolveToolChoiceERKNSt3__18optionalINS1_12basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEEEERKNS1_6vectorI16common_chat_toolNS6_ISD_EEEE",
             "__ZN14MtmdLlmContext10resetMediaEv",
             "__ZN30qvac_lib_inference_addon_llama8batching22finalizeTerminalDriverER14SequenceDriverNS0_10StopReasonEbRKNSt3__18functionIFvRKNS4_12basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEEEEE",
@@ -591,10 +598,10 @@
             "__ZNK14MtmdLlmContext16cacheStateTokensEv",
             "_bare_register_module_v0",
             "__ZThn48_N14TextLlmContext24captureHistoryCheckpointEi",
-            "__ZN14TextLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEE",
             "__ZN24llama_finetuning_helpers16parseLrSchedulerERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEERNS_18LoraLrScheduleTypeE",
             "__ZN14LlamaFinetuner19executeTrainingLoopERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsEP16ggml_opt_datasetxxRN24llama_finetuning_helpers20LoraLrSchedulerStateEPNS6_23TrainingCheckpointStateEjbS5_xPN22FinetuneTerminalResult5StatsE",
             "__ZN14LlamaFinetuner22prepareDatasetFromPathERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsERKNSt3__112basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEEPKcSE_",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime7processERK11llama_batch",
             "__ZN30qvac_lib_inference_addon_llama5utils25requireToolChoiceHonouredE23common_chat_tool_choicebbPKc",
             "__ZN12CacheManager15saveBeforeResetEv",
             "__ZN14LlamaFinetuner26discardPendingPauseRequestEv",
@@ -603,37 +610,43 @@
             "__ZNK18AsyncWeightsLoader12isFirstShardERKNSt3__14__fs10filesystem4pathE",
             "__ZN14MtmdLlmContext22configureReasoningTagsERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_RKNS0_8optionalIN30qvac_lib_inference_addon_llama5utils13ReasoningTagsEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot22prefillTokensPerSecondEv",
+            "__ZN14TextLlmContext15checkAntipromptEmi",
             "__ZN10LlamaModel13setInitLoaderENSt3__18optionalIN10InitLoader11LOADER_TYPEEEENS1_I23FinetuneConfigOverridesEE",
+            "__ZThn48_N14TextLlmContext18onDraftLogitsReadyEiijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request13tokenToFeedAtEi",
-            "__ZN14MtmdLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEy",
             "__ZNK14MtmdLlmContext19residentStateTokensEv",
             "__ZThn48_N14MtmdLlmContext8onCancelERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN14MtmdLlmContextC1ERK13common_paramsRK15LlmModelContextP12mtmd_contextiiy",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence8maxDraftEiii",
             "__ZN14MtmdLlmContext29adoptCheckpointsFromCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18resetDraftSequenceEP13llama_contexti",
             "__ZNK10LlamaModel15cancelInferenceEv",
             "__ZN14LlamaFinetuner17clearPauseRequestEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler29writeStateToFileOrThrowLockedEjRKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEERKNS2_6vectorIiNS6_IiEEEERKNS2_5dequeINS_5cache10CheckpointENS6_ISI_EEEE",
             "__ZN10LlamaModel17resolveShardPathsER10GGUFShardsRKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEE",
             "__ZN14TextLlmContext9onFailureERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
+            "__ZN14MtmdLlmContext23advanceDecodedTextTokenEi",
             "__ZN17backend_selection26getEffectiveGpuDeviceCountERKNS_16BackendInterfaceE",
-            "__ZN14TextLlmContext15checkAntipromptEv",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotC1EOS1_",
             "__ZN14MtmdLlmContext19appendResidentTokenEi",
+            "__ZThn48_N14TextLlmContext23prepareSpeculativeDraftEiij",
             "__ZN26LlamaLazyInitializeBackend17incrementRefCountEv",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher17nextAwaitingMediaEv",
             "__ZN14MtmdLlmContext27restorePreRequestCacheStateEv",
-            "__ZN14MtmdLlmContext15checkAntipromptEv",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher15markAllFinishedENS0_10StopReasonE",
             "__ZN30qvac_lib_inference_addon_llama5utils26generationPromptTailLengthEP13llama_contextRKNSt3__112basic_stringIcNS3_11char_traitsIcEENS3_9allocatorIcEEEERKNS3_6vectorIiNS7_IiEEEE",
             "__ZN14MtmdLlmContext16evalMediaSegmentEmi",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17cancelGroupQueuedEy",
             "__ZThn48_N14TextLlmContext32snapshotPreRequestRollbackAnchorEv",
             "__ZN30qvac_lib_inference_addon_llama5utils34isQwen3ReasoningFamilyArchitectureENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative16SpeculativeStats3addERKS1_",
             "__ZN30qvac_lib_inference_addon_llama5utils16isMedPsyBasenameENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
             "__ZN14TextLlmContextC1ERK13common_paramsRK15LlmModelContextii",
             "__ZN14TextLlmContext8onCancelERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZThn48_N14MtmdLlmContext32snapshotPreRequestRollbackAnchorEv",
             "__ZN14LlamaFinetuner18configureOptimizerERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsEP18llama_adapter_loraRN24llama_finetuning_helpers20LoraLrSchedulerStateEPNS6_23TrainingCheckpointStateEb",
+            "__ZN30qvac_lib_inference_addon_llama11speculative25restoreDraftSequenceStateEP13llama_contextiRKNS0_18DraftSequenceStateEji",
+            "__ZN14TextLlmContext16emitSampledTokenEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatchm",
             "__ZNK14LlamaFinetuner31getCurrentCheckpointStateSharedEv",
             "__ZN18AsyncWeightsLoaderC2ERK10GGUFShardsR10InitLoaderRKNSt3__112basic_stringIcNS5_11char_traitsIcEENS5_9allocatorIcEEEEP13ModelMetaData",
             "__ZThn48_N14MtmdLlmContext17onPrefillCompleteEim",
@@ -647,6 +660,7 @@
             "__ZN10LlamaModelD2Ev",
             "__ZNK14TextLlmContext25getToolDefinitionsDroppedEv",
             "__ZThn48_N14TextLlmContext18adoptResidentStateERKNSt3__16vectorIiNS0_9allocatorIiEEEE",
+            "__ZN14TextLlmContext23prepareSpeculativeDraftEiij",
             "__ZN18AsyncWeightsLoader31lendFirstShardAndWaitForReleaseEONSt3__110unique_ptrINS0_15basic_streambufIcNS0_11char_traitsIcEEEENS0_14default_deleteIS5_EEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request16isPrefillPendingEv",
             "__ZThn32_NK10LlamaModel10cancelByIdEy",
@@ -654,7 +668,6 @@
             "__ZN12CacheManager10invalidateEv",
             "__ZN24llama_finetuning_helpers19loadPauseCheckpointERKNSt3__14__fs10filesystem4pathEP18llama_adapter_loraP11llama_modelP13llama_contextPP16ggml_opt_contextRNS_18CheckpointMetadataE",
             "__ZN10LlamaModel16beginFinetuneJobEy",
-            "__ZN14TextLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher7isValidEj",
             "__ZN14MtmdLlmContext23cancelGenerationCleanupERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN19LlamaBackendsHandleC2ERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_",
@@ -662,6 +675,7 @@
             "__ZThn48_N14TextLlmContext14preparePrefillERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERKNS1_INS1_IhNS3_IhEEEENS3_ISE_EEEERKNS1_I12PlannedMediaNS3_ISJ_EEEEbb",
             "__ZN14TextLlmContext23restoreCacheStateTokensERKNSt3__16vectorIiNS0_9allocatorIiEEEE",
             "__ZThn48_N14TextLlmContext29adoptCheckpointsFromCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
+            "__ZN14TextLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEENS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
             "__ZN14TextLlmContext17beginCacheRequestEv",
             "__ZNK12CacheManager15isCacheDisabledEv",
             "__ZThn48_NK14TextLlmContext19residentStateTokensEv",
@@ -685,12 +699,14 @@
             "__ZThn32_N10LlamaModelD1Ev",
             "__ZN14MtmdLlmContext12tokenizeChatERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERN4mtmd12input_chunksEb",
             "__ZN10LlamaModel26validateBitnetQuantizationEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence10addToBatchER11llama_batchii",
             "__ZThn8_N10LlamaModelD0Ev",
             "__ZThn48_N14TextLlmContext9onFailureERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN12CacheManager9saveCacheEv",
             "__ZN24llama_finetuning_helpers21pauseCheckpointExistsERKNSt3__14__fs10filesystem4pathE",
             "__ZN14TextLlmContext17onPrefillCompleteEim",
             "__ZN14LlamaFinetuner32waitUntilFinetuningPauseCompleteEv",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence12setPromptEndEi",
             "__ZN24llama_finetuning_helpers21buildNextTokenDatasetERKNSt3__16vectorIiNS0_9allocatorIiEEEExx",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler22residentHitsForTestingEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17chooseSeqIdLockedERKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEE",
@@ -718,6 +734,7 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15StepUnlockGuardD2Ev",
             "__ZN30qvac_lib_inference_addon_llama8batching7RequestC2EjO11PrefillPlanji",
             "__ZN12CacheManager17isFileInitializedERKNSt3__14__fs10filesystem4pathE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeD1Ev",
             "__ZN10LlamaModelC2EONSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_ONS0_13unordered_mapIS6_S6_NS0_4hashIS6_EENS0_8equal_toIS6_EENS4_INS0_4pairIKS6_S6_EEEEEE",
             "__ZN14TextLlmContext21rollbackFailedRequestEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler12submitLockedEONS1_13QueuedRequestE",
@@ -738,6 +755,7 @@
             "__ZN30qvac_lib_inference_addon_llama28applyGenerationParamHandlersEP8js_env_sRN28qvac_lib_inference_addon_cpp2js6ObjectER16GenerationParams",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15hasValidDriverFEv",
             "__ZN10LlamaModel19flushResidentCachesEv",
+            "__ZN14TextLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEENS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
             "__ZN18AsyncWeightsLoader30extractIndividualStreamedFilesEv",
             "__ZN14MtmdLlmContextC2ERK13common_paramsRK15LlmModelContextP12mtmd_contextiiy",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler5clearEv",
@@ -751,6 +769,7 @@
             "__ZN12CacheManager24isParentDirectoryMissingERKNSt3__14__fs10filesystem4pathE",
             "__ZN14MtmdLlmContext16sampleFromLogitsEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatch",
             "__ZN14MtmdLlmContext24rebuildSamplerFromLedgerERKN30qvac_lib_inference_addon_llama5cache6LedgerE",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime13resetSequenceEi",
             "__ZNK13ModelMetaData9tryGetU32EPKc",
             "__ZN12CacheManager10setRamTierENSt3__110shared_ptrIN30qvac_lib_inference_addon_llama8batching14SlotStateCacheEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher23planChunksForActiveSeqsERK10LlamaBatch",
@@ -760,11 +779,14 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler14enqueueSaveJobERKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEEb",
             "__ZNK12CacheManager24wasCacheUsedInLastPromptEv",
             "__ZN14TextLlmContextC2ERK13common_paramsRK15LlmModelContextii",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence13draftPreparedERKNS0_18SpeculativeRuntimeE",
             "__ZNK14TextLlmContext28appendCheckpointsToCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN12CacheManagerC1EP10LlmContextNSt3__18functionIFvbEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative22applySpeculativeConfigERKNS0_17SpeculativeConfigER13common_params",
             "__ZN30qvac_lib_inference_addon_llama5utils20getModelArchitectureEPK11llama_model",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17hasPendingCancelsEv",
             "__ZNK13ModelMetaData21hasOneBitQuantizationEv",
+            "__ZN14MtmdLlmContext25evalAudioChunkSpeculativeEPK16mtmd_input_chunkiiPi",
             "__ZNK10LlamaModel6cancelEv",
             "__ZN30qvac_lib_inference_addon_llama5utils33sequenceStateSnapshotFilesWrittenEv",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher22nextAwaitingCheckpointEv",
@@ -774,6 +796,7 @@
             "__ZN14MtmdLlmContext20evalMessageWithToolsERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEEbb",
             "__ZNK12CacheManager14hasActiveCacheEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler6cancelEjy",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence13setBatchStartEi",
             "__ZN14TextLlmContext10resetStateEb",
             "__ZN17backend_selection17tryMainGpuFromMapERNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEE",
             "__ZN14LlamaFinetuner30setPausedCheckpointStateSharedENSt3__110shared_ptrIN24llama_finetuning_helpers23TrainingCheckpointStateEEE",
@@ -804,18 +827,22 @@
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler13hasWorkLockedEv",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher6cancelEjRKNSt3__18functionIFvjEEE",
             "__ZThn48_NK14MtmdLlmContext28appendCheckpointsToCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
+            "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler24teardownPendingForLockedEj",
             "__ZN10LlamaModel22processConcurrentBatchERKNSt3__16vectorINS_6PromptENS0_9allocatorIS2_EEEEy",
+            "__ZN14TextLlmContext19generateSpeculativeERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEERj",
             "__Z33qvacLibInferenceAddonLlamaExportsP8js_env_sP10js_value_s",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler25ensureWorkerStartedLockedEv",
             "__ZN14MtmdLlmContext13onSequenceEndERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN24llama_finetuning_helpers23optEpochCallbackWrapperEbP16ggml_opt_contextP16ggml_opt_datasetP15ggml_opt_resultxxx",
             "__ZN14MtmdLlmContext8setNPastEi",
             "__ZThn24_N10LlamaModelD0Ev",
+            "__ZN14TextLlmContext23processSpeculativeBatchERK11llama_batch",
             "__ZN30qvac_lib_inference_addon_llama8batching7RequestC1EjO11PrefillPlanji",
             "__ZN30qvac_lib_inference_addon_llama5utils25partialSnapshotFromStreamERNSt3__113basic_istreamIcNS1_11char_traitsIcEEEEyiNS0_15SnapshotStorageERKNS1_12basic_stringIcS4_NS1_9allocatorIcEEEERKNS1_8functionIFvPKhmEEERNS0_21SequenceStateSnapshotE",
             "__ZN10LlamaModelD0Ev",
             "__ZN24llama_finetuning_helpers18schedulerLrForStepERKNS_20LoraLrSchedulerStateEx",
             "__ZN12CacheManager28persistedBackingStoreMissingERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
+            "__ZThn48_N14MtmdLlmContext23prepareSpeculativeDraftEiij",
             "__ZN14MtmdLlmContext23restoreCacheStateTokensERKNSt3__16vectorIiNS0_9allocatorIiEEEE",
             "__ZN14MtmdLlmContext9loadCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEb",
             "__ZN14TextLlmContext19acceptRestoredStateERKNSt3__16vectorIiNS0_9allocatorIiEEEERKNS0_12basic_stringIcNS0_11char_traitsIcEENS2_IcEEEE",
@@ -846,9 +873,10 @@
             "__ZN17backend_selection23shouldWarnAboutJitCacheEv",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request11isOptActiveERKNSt3__18optionalIS1_EE",
             "__ZNK30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot16avgConcurrentSeqEv",
-            "__ZThn48_N10LlamaModel11jobStartingEy",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence10afterDraftEv",
             "__ZN17backend_selection12parseMainGpuERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request15remainingToFeedEv",
+            "__ZThn48_N10LlamaModel11jobStartingEy",
             "__ZThn8_N10LlamaModel17setWeightsForFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEONS0_10unique_ptrINS0_15basic_streambufIcS3_EENS0_14default_deleteISB_EEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils16isGemma4BasenameENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
             "__ZN10LlamaModel4initEb",
@@ -856,6 +884,7 @@
             "__ZN10LlamaModel17processPromptImplERKNS_6PromptE",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher5clearERKNSt3__18functionIFvjEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request13exceededLimitEv",
+            "__ZN10LlamaModel13createContextEONSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEER13common_paramsNS0_10unique_ptrI18common_init_resultNS0_14default_deleteISB_EEEENSA_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENSC_ISH_EEEE",
             "__ZN14MtmdLlmContext10resetStateEb",
             "__ZN14TextLlmContext27confirmPendingResidentTokenEi",
             "__ZN24llama_finetuning_helpers25findLatestPauseCheckpointERKNSt3__14__fs10filesystem4pathE",
@@ -865,6 +894,8 @@
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher20completeMediaBarrierEjiRKNSt3__18functionIFvjimEEE",
             "__ZN14TextLlmContext17removeLastNTokensEi",
             "__ZNK10LlamaModel12runtimeStatsEv",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime9nDraftMaxEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence4bindEPKNS0_18SpeculativeRuntimeEi",
             "__ZN12CacheManager18restoreFromRamTierENSt3__18optionalIN30qvac_lib_inference_addon_llama8batching19SlotStateCacheEntryEEE",
             "__ZN10LlamaModel22processPromptBatchImplERKNSt3__16vectorINS_6PromptENS0_9allocatorIS2_EEEERKNS0_8functionIFbmjyEEERKNS8_IFvmjEEEy",
             "__ZN14TextLlmContext26flushPendingUtf8ToCallbackERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
@@ -880,6 +911,7 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchSchedulerC1E15LlmModelContextjjmiRK13common_paramsNSt3__18functionIFNS6_10unique_ptrI14SequenceDriverNS6_14default_deleteIS9_EEEES5_jiEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request21chunkConsumesAllUnfedEj",
             "__ZN10LlamaModel7processERKNSt3__13anyE",
+            "__ZThn48_N14MtmdLlmContext18onDraftLogitsReadyEiijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN26LlamaLazyInitializeBackend16initializeLockedERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_",
             "__ZN10LlamaModel32setFinetuneCancelSavesCheckpointEbRKNSt3__16vectorIyNS0_9allocatorIyEEEE",
             "__ZN13ModelMetaData5parseERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEERK10GGUFShardsbPKc",
@@ -894,6 +926,7 @@
             "__ZN10LlamaModel17processConcurrentERKNS_6PromptEy",
             "__ZN24llama_finetuning_helpers12readTextFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler26completeGroupRequestLockedERKNSt3__110shared_ptrINS1_10BatchGroupEEE",
+            "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher8setDraftEjNSt3__16vectorIiNS2_9allocatorIiEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler24recordPendingGroupCancelEy",
             "__ZNK14TextLlmContext10ctxCeilingEv",
             "__ZN14MtmdLlmContext17beginCacheRequestEv",
@@ -906,6 +939,7 @@
             "__ZN12CacheManager17acceptLoadedStateERNSt3__16vectorIiNS0_9allocatorIiEEEERKNS0_12basic_stringIcNS0_11char_traitsIcEENS2_IcEEEE",
             "__ZN14TextLlmContext32snapshotPreRequestRollbackAnchorEv",
             "__ZN17backend_selection30getUnknownFinetuneArchitectureEPK13ModelMetaData",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence5beginERKNSt3__16vectorIiNS2_9allocatorIiEEEE",
             "__ZN14MtmdLlmContext29clearCacheReconciliationStateEv",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request18isOptAwaitingMediaERKNSt3__18optionalIS1_EE",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request13isOptFinishedERKNSt3__18optionalIS1_EE",
@@ -914,18 +948,23 @@
             "__ZThn48_N14MtmdLlmContext9loadCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEb",
             "__ZThn40_NK10LlamaModel15consumeJobStatsEy",
             "__ZThn16_N10LlamaModelD1Ev",
+            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler30prepareSpeculativeDraftsLockedEPNSt3__111unique_lockINS2_5mutexEEE",
             "__ZN10LlamaModelC1EONSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_ONS0_13unordered_mapIS6_S6_NS0_4hashIS6_EENS0_8equal_toIS6_EENS4_INS0_4pairIKS6_S6_EEEEEE",
+            "__ZN14MtmdLlmContext23prepareSpeculativeDraftEiij",
             "__ZN14LlamaFinetuner15saveLoraAdapterEP18llama_adapter_loraRKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsE",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler7hasWorkEv",
             "__ZN14TextLlmContext22rollbackCurrentRequestERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN14TextLlmContext27discardPendingResidentTokenEv",
             "__ZThn48_N14TextLlmContext9loadCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEb",
-            "__ZN10LlamaModel13createContextEONSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEER13common_paramsNS0_10unique_ptrI18common_init_resultNS0_14default_deleteISB_EEEE",
+            "__ZN14MtmdLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEyNS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeFitModel6createERK13common_params",
             "__ZN19LlamaBackendsHandleC1EOS_",
             "__ZN12CacheManager11handleCacheER19ParsedPromptPayloadRKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEENS2_8functionIFS0_SA_EEESA_b",
             "__ZN17image_pixel_limit13takeMaxPixelsERNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEE",
+            "__ZN14MtmdLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEyNS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot21decodeTokensPerSecondEv",
             "__ZN14MtmdLlmContext17onPrefillCompleteEim",
+            "__ZN14MtmdLlmContext23processSpeculativeBatchERK11llama_batch",
             "__ZN17backend_selection25tryBackendOverrideFromMapERNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request22isOptGenerationPendingERKNSt3__18optionalIS1_EE",
             "__ZN14TextLlmContext29adoptCheckpointsFromCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
@@ -950,6 +989,7 @@
             "__ZNK14MtmdLlmContext28appendCheckpointsToCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN10LlamaModel19resolveChatAndToolsERKNS_6PromptE",
             "__ZThn48_NK14MtmdLlmContext8getNPastEv",
+            "__ZN14TextLlmContext18onDraftLogitsReadyEiijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZNK14MtmdLlmContext14getCacheTokensEv",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot5clearEv",
             "__ZN14TextLlmContext15reconcilePromptERKNSt3__16vectorIiNS0_9allocatorIiEEEEb",
@@ -974,6 +1014,7 @@
             "__ZNK10LlamaModel10cancelByIdEy",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17evictParkedLockedEj",
             "__ZN30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot5resetEv",
+            "__ZN14MtmdLlmContext19generateSpeculativeERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEERi",
             "__ZNK14MtmdLlmContext9saveCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN14MtmdLlmContext26flushPendingUtf8ToCallbackERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZNK10LlamaModel16jobTerminalStatsERKN30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshotERKNS1_20ObservedRequestStatsE",
@@ -983,8 +1024,10 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler11evictParkedEj",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher22completeCheckpointStopEj",
             "__ZNK14MtmdLlmContext17getVisionEncodeMsEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence5resetEv",
             "__ZN14MtmdLlmContext9onFailureERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler31admitPendingIntoFreeSlotsLockedEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime9forTargetEP13llama_context",
             "__ZN14TextLlmContext11evalMessageERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEEbb",
             "__ZN14TextLlmContext8setNPastEi",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot9adoptFileENSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEEiy",
@@ -993,14 +1036,20 @@
             "__ZThn48_N14MtmdLlmContext24snapshotPreRequestCursorEv",
             "__ZN10LlamaModel18initBatchSchedulerERNS_15ReloadableStateE",
             "__ZN17backend_selection20splitModeDeviceNamesERKNS_16BackendInterfaceERKNSt3__112basic_stringIcNS3_11char_traitsIcEENS3_9allocatorIcEEEE",
+            "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12draftDroppedEj",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime6createER13common_paramsP11llama_modelP13llama_context",
             "__ZThn48_N14TextLlmContext8onCancelERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN14MtmdLlmContext24snapshotPreRequestCursorEv",
             "__ZThn48_N14MtmdLlmContext9onFailureERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
+            "__ZN14MtmdLlmContext18onDraftLogitsReadyEiijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN24llama_finetuning_helpers16optEpochCallbackEbP16ggml_opt_contextP16ggml_opt_datasetP15ggml_opt_resultxxxPNS_23TrainingCheckpointStateE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative21requireDraftModelFileERKNS0_17SpeculativeConfigE",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12addRequestAtEjO11PrefillPlaniib",
-            "__ZNK14MtmdLlmContext16ledgerFromChunksERKN4mtmd12input_chunksE",
             "__ZN30qvac_lib_inference_addon_llama26applyFinetuneParamHandlersEP8js_env_sRN28qvac_lib_inference_addon_cpp2js6ObjectERNS_21LlamaFinetuningParamsE",
+            "__ZNK14MtmdLlmContext16ledgerFromChunksERKN4mtmd12input_chunksE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request15hasTokensToFeedEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative25captureDraftSequenceStateEP13llama_contextij",
+            "__ZN30qvac_lib_inference_addon_llama11speculative22parseSpeculativeConfigERNSt3__113unordered_mapINS1_12basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEES8_NS1_4hashIS8_EENS1_8equal_toIS8_EENS6_INS1_4pairIKS8_S8_EEEEEE",
             "__ZN17image_pixel_limit11checkBufferEPKhmy",
             "__ZN10LlamaModel6reloadENSt3__18optionalI23FinetuneConfigOverridesEE",
             "__ZN12CacheManagerC2EP10LlmContextNSt3__18functionIFvbEEE",
@@ -1009,6 +1058,7 @@
             "__ZN14MtmdLlmContext4stopEv",
             "__ZN14TextLlmContext24captureHistoryCheckpointEi",
             "__ZN22load_fit_normalization22productionDependenciesEPFv14ggml_log_levelPKcPvE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence12discardDraftEv",
             "__ZNK18AsyncWeightsLoader20shouldLendFirstShardERKNSt3__14__fs10filesystem4pathE",
             "__ZN14TextLlmContext24rebuildSamplerFromLedgerERKN30qvac_lib_inference_addon_llama5cache6LedgerE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler28serviceCheckpointStopsLockedEv",
@@ -1039,6 +1089,7 @@
             "__ZN14LlamaFinetuner32clearPausedCheckpointStateSharedEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17getOutputCallbackERNS1_9SlotStateEj",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler6submitEONS0_13SubmitRequestE",
+            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler29verifySpeculativeDraftsLockedEv",
             "__ZNK18AsyncWeightsLoader11isLastShardERKNSt3__14__fs10filesystem4pathE",
             "__ZN14TextLlmContext19appendResidentTokenEi",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request23isOptAwaitingCheckpointERKNSt3__18optionalIS1_EE",
@@ -1049,6 +1100,7 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler10notifyDoneEj",
             "__ZN24llama_finetuning_helpers23parseCheckpointMetadataERKNSt3__14__fs10filesystem4pathERNS_18CheckpointMetadataE",
             "__ZN14MtmdLlmContext32snapshotPreRequestRollbackAnchorEv",
+            "__ZN14MtmdLlmContext15checkAntipromptEmi",
             "__ZN14TextLlmContext15emitOutputPieceERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEES9_",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotC2EOS1_",
             "__ZN13ModelMetaData28FirstFileFromGgufStreamState7provideER13BorrowablePtrINSt3__115basic_streambufIcNS2_11char_traitsIcEEEEE",
@@ -1058,7 +1110,6 @@
             "__ZThn48_NK14MtmdLlmContext19residentStateTokensEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17resetRuntimeStatsEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler16saveConversationERKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEE",
-            "__ZN14MtmdLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEy",
             "__ZN24llama_finetuning_helpers15tokenizeDatasetEP13llama_contextRKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEE",
             "__ZThn48_N14MtmdLlmContext14preparePrefillERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERKNS1_INS1_IhNS3_IhEEEENS3_ISE_EEEERKNS1_I12PlannedMediaNS3_ISJ_EEEEbb",
             "__ZThn48_NK14TextLlmContext25getToolDefinitionsDroppedEv",
@@ -1073,6 +1124,7 @@
             "__ZN30qvac_lib_inference_addon_llama5utils30createPrivateSnapshotDirectoryERKNSt3__14__fs10filesystem4pathE",
             "__ZN14MtmdLlmContext27confirmPendingResidentTokenEi",
             "__ZN14TextLlmContext22configureReasoningTagsERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_RKNS0_8optionalIN30qvac_lib_inference_addon_llama5utils13ReasoningTagsEEE",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence19useTargetCheckpointEm",
             "__ZN14MtmdLlmContext27resetToolDefinitionsDroppedEv",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12chunkSizeForEj",
             "__ZN14LlamaFinetuner22prepareTrainingDatasetERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsE",
@@ -1082,6 +1134,7 @@
             "__ZN30qvac_lib_inference_addon_llama7parsers26parseReasoningBudgetConfigENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
             "__ZNK13ModelMetaData16checkInitializedEv",
             "__ZN10LlamaModel31closeFinetuneCancellationWindowEv",
+            "__ZN14MtmdLlmContext16emitSampledTokenEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatchm",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler18notifyDoneNoexceptEj",
             "__ZN12CacheManager24moveActiveCacheToRamTierEv",
             "__ZN14TextLlmContext29clearCacheReconciliationStateEv",