npm - @qvac/llm-llamacpp - Versions diffs - 0.61.2 → 0.63.0 - Mend

@qvac/llm-llamacpp 0.61.2 → 0.63.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (18) hide show

package/prebuilds/ios-arm64/qvac__llm-llamacpp.bare.exports CHANGED Viewed

@@ -7,7 +7,7 @@
     ],
     "current_versions": [
       {
-        "version": "0.61.2"
+        "version": "0.63"
       }
     ],
     "exported_symbols": [
@@ -61,10 +61,10 @@
             "__ZNSt3__19__any_imp17__unique_typeinfoIN30qvac_lib_inference_addon_llama16BatchTokenOutputEE4__idE",
             "__ZGVN28qvac_lib_inference_addon_cpp6logger8JsLogger12admin_mutex_E",
             "__ZN28qvac_lib_inference_addon_cpp6logger8JsLogger16async_initiated_E",
-            "__ZNSt3__19__any_imp17__unique_typeinfoIN28qvac_lib_inference_addon_cpp6Output6LogMsgEE4__idE",
             "__ZN30qvac_lib_inference_addon_llama5cache24CHECKPOINT_SECTION_MAGICE",
             "__ZNSt3__19__any_imp17__unique_typeinfoI22FinetuneTerminalResultE4__idE",
             "__ZN28qvac_lib_inference_addon_cpp11JsInterface13instancesMtx_E",
+            "__ZNSt3__19__any_imp17__unique_typeinfoIN28qvac_lib_inference_addon_cpp6Output6LogMsgEE4__idE",
             "__ZN30qvac_lib_inference_addon_llama22BATCH_OUTPUT_TYPE_NAMEE",
             "__ZN28qvac_lib_inference_addon_cpp6logger8JsLogger10log_queue_E",
             "__ZN30qvac_lib_inference_addon_llama5cache26CHECKPOINT_SECTION_VERSIONE",
@@ -84,9 +84,11 @@
             "__ZN24llama_finetuning_helpers23optEpochCallbackWrapperEbP16ggml_opt_contextP16ggml_opt_datasetP15ggml_opt_resultxxx",
             "__ZThn24_N10LlamaModelD0Ev",
             "__ZN14MtmdLlmContext8setNPastEi",
+            "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher13verifyDraftedERKNSt3__18functionIFNS1_12DraftOutcomeEjiiEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher10addRequestEONSt3__16vectorIiNS2_9allocatorIiEEEERj",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher19sampleAndAppendIdleERKNSt3__18functionIFijiEEE",
             "__ZN10LlamaModel30discardFinetuneCancelSaveModesERKNSt3__16vectorIyNS0_9allocatorIyEEEE",
+            "__ZN14TextLlmContext23processSpeculativeBatchERK11llama_batch",
             "__ZN30qvac_lib_inference_addon_llama8batching7RequestC1EjO11PrefillPlanji",
             "__ZN30qvac_lib_inference_addon_llama5utils25partialSnapshotFromStreamERNSt3__113basic_istreamIcNS1_11char_traitsIcEEEEyiNS0_15SnapshotStorageERKNS1_12basic_stringIcS4_NS1_9allocatorIcEEEERKNS1_8functionIFvPKhmEEERNS0_21SequenceStateSnapshotE",
             "__ZN10LlamaModelD0Ev",
@@ -110,6 +112,7 @@
             "__ZN30qvac_lib_inference_addon_llama5utils21streamPartialSnapshotERKNS0_21SequenceStateSnapshotERKNSt3__18functionIFvPKhmEEE",
             "__ZThn16_NK10LlamaModel6cancelEv",
             "__ZN17backend_selection30preferredBackendTypeFromStringERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence12prepareDraftEiiiRKNSt3__16vectorIiNS2_9allocatorIiEEEE",
             "__ZThn48_N14TextLlmContext24snapshotPreRequestCursorEv",
             "__ZN10LlamaModel9saveCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZNK10LlamaModel15consumeJobStatsEy",
@@ -122,10 +125,12 @@
             "__ZN30qvac_lib_inference_addon_llama5utils25selectReasoningBudgetTagsERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEES9_RKNS1_6vectorIS7_NS5_IS7_EEEERKNS1_8optionalINS0_13ReasoningTagsEEE",
             "__ZThn32_N10LlamaModelD0Ev",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request20isOptHasTokensToFeedERKNSt3__18optionalIS1_EE",
+            "__ZThn48_N14MtmdLlmContext23prepareSpeculativeDraftEiij",
             "__ZThn48_N14TextLlmContext20onGenerationFinishedERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE20GenerationStopReason",
             "__ZN12CacheManager14writeCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZThn48_N14MtmdLlmContext13onLogitsReadyEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatch",
             "__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",
             "__ZThn16_N10LlamaModelD0Ev",
             "__ZN12CacheManager14flushForUnloadEv",
@@ -148,15 +153,17 @@
             "__ZThn8_N10LlamaModel17setWeightsForFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEONS0_10unique_ptrINS0_15basic_streambufIcS3_EENS0_14default_deleteISB_EEEE",
             "__ZN14MtmdLlmContext24holdPendingResidentTokenEii",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler16discardKeyLockedERKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEE",
-            "__ZThn48_N10LlamaModel11jobStartingEy",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence10afterDraftEv",
             "__ZN12CacheManager7discardERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN17backend_selection12parseMainGpuERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
+            "__ZThn48_N10LlamaModel11jobStartingEy",
             "__ZN30qvac_lib_inference_addon_llama5utils24initializeReasoningStateEP13llama_contextRNS0_14ReasoningStateENS0_13ReasoningTagsE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15StepUnlockGuardC2ERS1_PNSt3__111unique_lockINS4_5mutexEEE",
             "__ZThn8_N10LlamaModelD1Ev",
+            "__ZN30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot21recordSpeculativeStepExNSt3__16chrono8durationIxNS2_5ratioILl1ELl1000000000EEEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler9numActiveEv",
-            "__ZNK30qvac_lib_inference_addon_llama8batching7Request15remainingToFeedEv",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12markFinishedEjNS0_10StopReasonE",
+            "__ZNK30qvac_lib_inference_addon_llama8batching7Request15remainingToFeedEv",
             "__ZThn48_N14TextLlmContext13onSequenceEndERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN10LlamaModel8getModelEv",
             "__ZN30qvac_lib_inference_addon_llama5utils16isGemma4BasenameENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
@@ -171,12 +178,14 @@
             "__Z30applyGenerationParamsToContextR13common_paramsRNSt3__110unique_ptrI14common_sampler20CommonSamplerDeleterEEP11llama_modelRK16GenerationParams",
             "__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",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler12runtimeStatsEv",
             "__ZN14MtmdLlmContext10resetStateEb",
             "__ZN14TextLlmContext27confirmPendingResidentTokenEi",
             "__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",
             "__ZN19LlamaBackendsHandleD2Ev",
             "__ZN24llama_finetuning_helpers25findLatestPauseCheckpointERKNSt3__14__fs10filesystem4pathE",
             "__ZN14MtmdLlmContext35refreshCurrentCacheTokensFromMemoryEv",
@@ -186,6 +195,7 @@
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher20completeMediaBarrierEjiRKNSt3__18functionIFvjimEEE",
             "__ZN14TextLlmContext17removeLastNTokensEi",
             "__ZNK10LlamaModel12runtimeStatsEv",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime9nDraftMaxEv",
             "__ZN14MtmdLlmContext10resetMediaEv",
             "__ZN30qvac_lib_inference_addon_llama8batching22finalizeTerminalDriverER14SequenceDriverNS0_10StopReasonEbRKNSt3__18functionIFvRKNS4_12basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot10adoptEmptyEi",
@@ -193,19 +203,20 @@
             "__ZN24llama_finetuning_helpers16parseLoraModulesERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZNK14TextLlmContext16cacheStateTokensEv",
             "__ZNK14MtmdLlmContext16cacheStateTokensEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence4bindEPKNS0_18SpeculativeRuntimeEi",
             "_bare_register_module_v0",
             "__ZN12CacheManager18restoreFromRamTierENSt3__18optionalIN30qvac_lib_inference_addon_llama8batching19SlotStateCacheEntryEEE",
             "__ZThn48_N14TextLlmContext24captureHistoryCheckpointEi",
             "__ZN10LlamaModel22processPromptBatchImplERKNSt3__16vectorINS_6PromptENS0_9allocatorIS2_EEEERKNS0_8functionIFbmjyEEERKNS8_IFvmjEEEy",
             "__ZN14TextLlmContext26flushPendingUtf8ToCallbackERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotaSEOS1_",
-            "__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",
             "__ZN19LlamaBackendsHandleaSEOS_",
             "__ZN14LlamaFinetuner22prepareDatasetFromPathERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsERKNSt3__112basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEEPKcSE_",
             "__ZN14LlamaFinetuner26validateModelForFinetuningEv",
             "__ZN10LlamaModel13processPromptERKNS_6PromptE",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime7processERK11llama_batch",
             "__ZN14LlamaFinetuner17createLrSchedulerERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsEx",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler16requestCancelAllEv",
             "__ZN30qvac_lib_inference_addon_llama5utils25requireToolChoiceHonouredE23common_chat_tool_choicebbPKc",
@@ -220,18 +231,22 @@
             "__ZN14MtmdLlmContext22configureReasoningTagsERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_RKNS0_8optionalIN30qvac_lib_inference_addon_llama5utils13ReasoningTagsEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchSchedulerC1E15LlmModelContextjjmiRK13common_paramsNSt3__18functionIFNS6_10unique_ptrI14SequenceDriverNS6_14default_deleteIS9_EEEES5_jiEEE",
             "__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",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request21chunkConsumesAllUnfedEj",
             "__ZN10LlamaModel7processERKNSt3__13anyE",
             "__ZN14MtmdLlmContextC1ERK13common_paramsRK15LlmModelContextP12mtmd_contextiiy",
             "__ZThn48_N14MtmdLlmContext8onCancelERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence8maxDraftEiii",
             "__ZN14MtmdLlmContext29adoptCheckpointsFromCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
             "__ZN26LlamaLazyInitializeBackend16initializeLockedERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_",
+            "__ZThn48_N14MtmdLlmContext18onDraftLogitsReadyEiijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN10LlamaModel32setFinetuneCancelSavesCheckpointEbRKNSt3__16vectorIyNS0_9allocatorIyEEEE",
             "__ZN13ModelMetaData5parseERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEERK10GGUFShardsbPKc",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18resetDraftSequenceEP13llama_contexti",
             "__ZNK10LlamaModel15cancelInferenceEv",
             "__ZN14LlamaFinetuner17clearPauseRequestEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler29writeStateToFileOrThrowLockedEjRKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEERKNS2_6vectorIiNS6_IiEEEERKNS2_5dequeINS_5cache10CheckpointENS6_ISI_EEEE",
@@ -240,37 +255,38 @@
             "__ZN12CacheManager13saveForCallerERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN14TextLlmContext18commitCacheRequestEv",
             "__ZN14TextLlmContext9onFailureERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
-            "__ZN17backend_selection26getEffectiveGpuDeviceCountERKNS_16BackendInterfaceE",
+            "__ZN14MtmdLlmContext23advanceDecodedTextTokenEi",
             "__ZN24llama_finetuning_helpers19savePauseCheckpointEP16ggml_opt_contextRNS_23TrainingCheckpointStateEbx",
             "__ZN14MtmdLlmContext14setCacheTokensEi",
-            "__ZN14TextLlmContext15checkAntipromptEv",
             "__ZN30qvac_lib_inference_addon_llama5utils25getThinkingForcedOpenTextERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEES9_",
+            "__ZN17backend_selection26getEffectiveGpuDeviceCountERKNS_16BackendInterfaceE",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotC1EOS1_",
             "__ZN24llama_finetuning_helpers25setCurrentCheckpointStateEPNS_23TrainingCheckpointStateE",
+            "__ZThn48_N14TextLlmContext23prepareSpeculativeDraftEiij",
             "__ZN10LlamaModel17processConcurrentERKNS_6PromptEy",
             "__ZN14MtmdLlmContext19appendResidentTokenEi",
             "__ZN24llama_finetuning_helpers12readTextFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler26completeGroupRequestLockedERKNSt3__110shared_ptrINS1_10BatchGroupEEE",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher9requestAtEj",
             "__ZN26LlamaLazyInitializeBackend17incrementRefCountEv",
-            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler24recordPendingGroupCancelEy",
+            "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher8setDraftEjNSt3__16vectorIiNS2_9allocatorIiEEEE",
             "__ZN14MtmdLlmContext27restorePreRequestCacheStateEv",
-            "__ZN14MtmdLlmContext15checkAntipromptEv",
+            "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher15markAllFinishedENS0_10StopReasonE",
             "__ZN14MtmdLlmContext17beginCacheRequestEv",
             "__ZN14MtmdLlmContext8onCancelERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
+            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler24recordPendingGroupCancelEy",
             "__ZN14MtmdLlmContext16evalMediaSegmentEmi",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher14firstFreeSeqIdEv",
             "__ZN30qvac_lib_inference_addon_llama5cache21readCheckpointSectionERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEEyRKNS0_6LedgerERKNS0_16CheckpointPolicyE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17cancelGroupQueuedEy",
-            "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher15markAllFinishedENS0_10StopReasonE",
+            "__ZThn48_N14TextLlmContext32snapshotPreRequestRollbackAnchorEv",
             "__ZN30qvac_lib_inference_addon_llama5utils26generationPromptTailLengthEP13llama_contextRKNSt3__112basic_stringIcNS3_11char_traitsIcEENS3_9allocatorIcEEEERKNS3_6vectorIiNS7_IiEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils34isQwen3ReasoningFamilyArchitectureENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
             "__ZNK14TextLlmContext10ctxCeilingEv",
             "__ZN10LlamaModel12formatPromptERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN14TextLlmContext27resetToolDefinitionsDroppedEv",
             "__ZN14LlamaFinetuner33clearCurrentCheckpointStateSharedEv",
-            "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher17nextAwaitingMediaEv",
-            "__ZThn48_N14TextLlmContext32snapshotPreRequestRollbackAnchorEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative16SpeculativeStats3addERKS1_",
             "__ZN30qvac_lib_inference_addon_llama5utils16isMedPsyBasenameENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
             "__ZN12CacheManager17acceptLoadedStateERNSt3__16vectorIiNS0_9allocatorIiEEEERKNS0_12basic_stringIcNS0_11char_traitsIcEENS2_IcEEEE",
             "__ZN14TextLlmContext8onCancelERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
@@ -280,6 +296,10 @@
             "__ZN17backend_selection30getUnknownFinetuneArchitectureEPK13ModelMetaData",
             "__ZThn48_N14MtmdLlmContext32snapshotPreRequestRollbackAnchorEv",
             "__ZN14TextLlmContextC1ERK13common_paramsRK15LlmModelContextii",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence5beginERKNSt3__16vectorIiNS2_9allocatorIiEEEE",
+            "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher17nextAwaitingMediaEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative25restoreDraftSequenceStateEP13llama_contextiRKNS0_18DraftSequenceStateEji",
+            "__ZN14TextLlmContext16emitSampledTokenEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatchm",
             "__ZN14MtmdLlmContext29clearCacheReconciliationStateEv",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request18isOptAwaitingMediaERKNSt3__18optionalIS1_EE",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request13isOptFinishedERKNSt3__18optionalIS1_EE",
@@ -291,46 +311,51 @@
             "__ZNK14LlamaFinetuner31getCurrentCheckpointStateSharedEv",
             "__ZThn16_N10LlamaModelD1Ev",
             "__ZThn48_N14MtmdLlmContext9loadCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEb",
+            "__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",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler25finalizeFinishedSequencesEv",
             "__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",
             "__ZN14TextLlmContext20onGenerationFinishedERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE20GenerationStopReason",
             "__Z23validateByteBufferCountRKNSt3__16vectorI12PlannedMediaNS_9allocatorIS1_EEEEm",
+            "__ZN14MtmdLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEyNS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeFitModel6createERK13common_params",
             "__ZN17backend_selection23getSplitDeviceSelectionEv",
             "__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",
-            "__ZN30qvac_lib_inference_addon_llama5utils21snapshotSequenceStateEP13llama_contextiiRNS0_21SequenceStateSnapshotENS0_15SnapshotStorageENS0_13SnapshotScopeERKNSt3__112basic_stringIcNS7_11char_traitsIcEENS7_9allocatorIcEEEE",
+            "__ZN14MtmdLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEyNS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
             "__ZNK10LlamaModel16supportsBatchingEv",
-            "__ZNK30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot21decodeTokensPerSecondEv",
+            "__ZN30qvac_lib_inference_addon_llama5utils21snapshotSequenceStateEP13llama_contextiiRNS0_21SequenceStateSnapshotENS0_15SnapshotStorageENS0_13SnapshotScopeERKNSt3__112basic_stringIcNS7_11char_traitsIcEENS7_9allocatorIcEEEE",
             "__ZN10LlamaModelD2Ev",
+            "__ZNK30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot21decodeTokensPerSecondEv",
             "__ZNK14TextLlmContext25getToolDefinitionsDroppedEv",
             "__ZN14MtmdLlmContext17onPrefillCompleteEim",
+            "__ZN14MtmdLlmContext23processSpeculativeBatchERK11llama_batch",
             "__ZN17backend_selection25tryBackendOverrideFromMapERNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEE",
-            "__ZN18AsyncWeightsLoader31lendFirstShardAndWaitForReleaseEONSt3__110unique_ptrINS0_15basic_streambufIcNS0_11char_traitsIcEEEENS0_14default_deleteIS5_EEEE",
+            "__ZN14TextLlmContext23prepareSpeculativeDraftEiij",
             "__ZN14TextLlmContext29adoptCheckpointsFromCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
             "__ZN22load_fit_normalization17tuneLoadConfigMapERNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEERK13ModelMetaDataRKNS0_8optionalIiEERK23FinetuneConfigOverridesbbbbb",
-            "__ZN30qvac_lib_inference_addon_llama8batching7Request22isOptGenerationPendingERKNSt3__18optionalIS1_EE",
+            "__ZN18AsyncWeightsLoader31lendFirstShardAndWaitForReleaseEONSt3__110unique_ptrINS0_15basic_streambufIcNS0_11char_traitsIcEEEENS0_14default_deleteIS5_EEEE",
             "__ZNK14MtmdLlmContext10ctxCeilingEv",
             "__ZThn48_N14TextLlmContext18adoptResidentStateERKNSt3__16vectorIiNS0_9allocatorIiEEEE",
-            "__ZThn48_N14MtmdLlmContext13onSequenceEndERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
+            "__ZN30qvac_lib_inference_addon_llama8batching7Request22isOptGenerationPendingERKNSt3__18optionalIS1_EE",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request16isPrefillPendingEv",
             "__ZThn32_NK10LlamaModel10cancelByIdEy",
             "__ZN30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot14accumulateSlotExxRKNS0_7RequestE",
             "__ZNK14MtmdLlmContext25getToolDefinitionsDroppedEv",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher13releaseBudgetEj",
             "__ZN12CacheManager10invalidateEv",
+            "__ZThn48_N14MtmdLlmContext13onSequenceEndERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN24llama_finetuning_helpers19loadPauseCheckpointERKNSt3__14__fs10filesystem4pathEP18llama_adapter_loraP11llama_modelP13llama_contextPP16ggml_opt_contextRNS_18CheckpointMetadataE",
             "__ZN10LlamaModel16beginFinetuneJobEy",
-            "__ZN14TextLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEE",
+            "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot11adoptBufferENSt3__16vectorIhNS2_9allocatorIhEEEEi",
             "__ZN14TextLlmContext21applyGenerationParamsERK16GenerationParams",
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher7isValidEj",
-            "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot11adoptBufferENSt3__16vectorIhNS2_9allocatorIhEEEEi",
             "__ZN22load_fit_normalization19normalizeLoadForFitERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEENS0_13unordered_mapIS6_S6_NS0_4hashIS6_EENS0_8equal_toIS6_EENS4_INS0_4pairIS7_S6_EEEEEERK13ModelMetaDataRK23FinetuneConfigOverridesRKNS_25NormalizationDependenciesE",
             "__ZN14TextLlmContext30capturePreRequestCacheSnapshotEv",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher15extractFinishedEv",
@@ -349,9 +374,11 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24generationBudgetExceededEjjij",
             "__ZNK14MtmdLlmContext28appendCheckpointsToCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZThn48_N14TextLlmContext29adoptCheckpointsFromCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
+            "__ZN14TextLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEENS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
             "__ZN14TextLlmContext17beginCacheRequestEv",
             "__ZN10LlamaModel19resolveChatAndToolsERKNS_6PromptE",
             "__ZThn48_NK14MtmdLlmContext8getNPastEv",
+            "__ZN14TextLlmContext18onDraftLogitsReadyEiijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZNK12CacheManager15isCacheDisabledEv",
             "__ZNK14MtmdLlmContext14getCacheTokensEv",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot5clearEv",
@@ -394,6 +421,7 @@
             "__ZN14MtmdLlmContext12tokenizeChatERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERN4mtmd12input_chunksEb",
             "__ZN10LlamaModel26validateBitnetQuantizationEv",
             "__ZN14MtmdLlmContext21rollbackFailedRequestEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence10addToBatchER11llama_batchii",
             "__ZThn8_N10LlamaModelD0Ev",
             "__ZN14MtmdLlmContext22commitCancelledRequestERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN12CacheManager9saveCacheEv",
@@ -404,10 +432,12 @@
             "__ZN14LlamaFinetuner32waitUntilFinetuningPauseCompleteEv",
             "__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",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher7advanceERKNSt3__18functionIFvjimEEE",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence12setPromptEndEi",
             "__ZN10LlamaModel9loadMediaERKNSt3__16vectorIhNS0_9allocatorIhEEEE",
             "__ZN24llama_finetuning_helpers21buildNextTokenDatasetERKNSt3__16vectorIiNS0_9allocatorIiEEEExx",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler22residentHitsForTestingEv",
@@ -418,6 +448,7 @@
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher22completeCheckpointStopEj",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler11evictParkedEj",
             "__ZNK14MtmdLlmContext17getVisionEncodeMsEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence5resetEv",
             "__ZThn40_N10LlamaModelD1Ev",
             "__ZN14MtmdLlmContext17initVisionContextEv",
             "__ZN14MtmdLlmContext9onFailureERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
@@ -427,8 +458,9 @@
             "__ZN30qvac_lib_inference_addon_llama5utils9getPromptEPK21common_chat_templatesR28common_chat_templates_inputs",
             "__ZThn48_NK14MtmdLlmContext14getKvCellsUsedEv",
             "__ZN14TextLlmContext16sampleFromLogitsEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatch",
-            "__ZNK30qvac_lib_inference_addon_llama8batching7Request10isFinishedEv",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime9forTargetEP13llama_context",
             "__ZN14TextLlmContext11evalMessageERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEEbb",
+            "__ZNK30qvac_lib_inference_addon_llama8batching7Request10isFinishedEv",
             "__ZN30qvac_lib_inference_addon_llama8batching20RuntimeStatsSnapshot16recordDecodeStepEyyyNSt3__16chrono8durationIxNS2_5ratioILl1ELl1000000000EEEEE",
             "__ZN14MtmdLlmContext20onGenerationFinishedERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE20GenerationStopReason",
             "__ZThn48_NK14TextLlmContext8getNPastEv",
@@ -449,24 +481,31 @@
             "__ZThn48_N14MtmdLlmContext24snapshotPreRequestCursorEv",
             "__ZN10LlamaModel18initBatchSchedulerERNS_15ReloadableStateE",
             "__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",
             "__ZN17backend_selection20splitModeDeviceNamesERKNS_16BackendInterfaceERKNSt3__112basic_stringIcNS3_11char_traitsIcEENS3_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler12submitLockedEONS1_13QueuedRequestE",
+            "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher12draftDroppedEj",
             "__ZN14MtmdLlmContext13onLogitsReadyEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatch",
+            "__ZN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime6createER13common_paramsP11llama_modelP13llama_context",
             "__ZNK18AsyncWeightsLoader25hasIndividualStreamedFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__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",
             "__ZNK14MtmdLlmContext14getKvCellsUsedEv",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request15hasTokensToFeedEv",
             "__ZN14TextLlmContext9loadCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEb",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request19isOptPrefillPendingERKNSt3__18optionalIS1_EE",
             "__ZN14MtmdLlmContext21applyGenerationParamsERK16GenerationParams",
+            "__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",
             "__ZN14LlamaFinetuner18prepareEvalDatasetERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsE",
@@ -484,10 +523,12 @@
             "__ZN14TextLlmContext24captureHistoryCheckpointEi",
             "__ZN22load_fit_normalization22productionDependenciesEPFv14ggml_log_levelPKcPvE",
             "__ZN30qvac_lib_inference_addon_llama28applyGenerationParamHandlersEP8js_env_sRN28qvac_lib_inference_addon_cpp2js6ObjectER16GenerationParams",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence12discardDraftEv",
             "__ZNK18AsyncWeightsLoader20shouldLendFirstShardERKNSt3__14__fs10filesystem4pathE",
             "__ZN10LlamaModel19flushResidentCachesEv",
             "__ZN14TextLlmContext24rebuildSamplerFromLedgerERKN30qvac_lib_inference_addon_llama5cache6LedgerE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler28serviceCheckpointStopsLockedEv",
+            "__ZN14TextLlmContextC1ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEENS3_IN30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntimeENS5_ISA_EEEE",
             "__ZN14MtmdLlmContext19acceptRestoredStateERKNSt3__16vectorIiNS0_9allocatorIiEEEERKNS0_12basic_stringIcNS0_11char_traitsIcEENS2_IcEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils20historyReasoningTagsEPK21common_chat_templatesPK11llama_modelb",
             "__ZN24llama_finetuning_helpers20clearPauseCheckpointERKNSt3__14__fs10filesystem4pathE",
@@ -498,24 +539,25 @@
             "__ZN14MtmdLlmContextC2ERK13common_paramsRK15LlmModelContextP12mtmd_contextiiy",
             "__ZN30qvac_lib_inference_addon_llama5utils19getHarmonyCallTokenEP13llama_context",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotD2Ev",
-            "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15hasValidDriverFEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler5clearEv",
+            "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15hasValidDriverFEv",
+            "__ZThn48_NK14MtmdLlmContext9saveCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZThn24_N10LlamaModel7processERKNSt3__13anyEy",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler29serviceNextMediaSegmentLockedEPNSt3__111unique_lockINS2_5mutexEEE",
-            "__ZThn48_NK14MtmdLlmContext9saveCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama5utils13isGemma4ModelEPK11llama_model",
+            "__ZN14TextLlmContext16generateResponseERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN10LlamaModel24validateCheckpointBudgetERNS_15ReloadableStateE",
             "__ZN22load_fit_normalization31validateMobileMultiDeviceConfigERKNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEE16llama_split_mode",
-            "__ZN24llama_finetuning_helpers24schedulerOptimizerParamsEPv",
-            "__ZN14TextLlmContext16generateResponseERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE",
             "__ZN17backend_selection23shouldWarnAboutJitCacheERKNS_11JitCacheEnvE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler19cancelPendingLockedEv",
+            "__ZN24llama_finetuning_helpers24schedulerOptimizerParamsEPv",
             "__ZN24llama_finetuning_helpers21tryHandlePauseRequestEP16ggml_opt_contextPNS_23TrainingCheckpointStateEbxx",
             "__ZN10LlamaModel21isMultiBatchActivatedERNS_15ReloadableStateE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchSchedulerC2E15LlmModelContextjjmiRK13common_paramsNSt3__18functionIFNS6_10unique_ptrI14SequenceDriverNS6_14default_deleteIS9_EEEES5_jiEEE",
             "__ZN12CacheManager24isParentDirectoryMissingERKNSt3__14__fs10filesystem4pathE",
             "__ZN14MtmdLlmContext16sampleFromLogitsEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatch",
             "__ZN14MtmdLlmContext24rebuildSamplerFromLedgerERKN30qvac_lib_inference_addon_llama5cache6LedgerE",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative18SpeculativeRuntime13resetSequenceEi",
             "__ZN14MtmdLlmContext15reconcilePromptE11PrefillPlanRKN30qvac_lib_inference_addon_llama5cache6LedgerEb",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15StepUnlockGuardD1Ev",
             "__ZN26LlamaLazyInitializeBackend10initializeERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_",
@@ -537,15 +579,19 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17getOutputCallbackERNS1_9SlotStateEj",
             "__ZNK12CacheManager24wasCacheUsedInLastPromptEv",
             "__ZN14TextLlmContextC2ERK13common_paramsRK15LlmModelContextii",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence13draftPreparedERKNS0_18SpeculativeRuntimeE",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler6submitEONS0_13SubmitRequestE",
+            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler29verifySpeculativeDraftsLockedEv",
             "__ZNK14TextLlmContext28appendCheckpointsToCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN12CacheManagerC1EP10LlmContextNSt3__18functionIFvbEEE",
             "__ZNK18AsyncWeightsLoader11isLastShardERKNSt3__14__fs10filesystem4pathE",
+            "__ZN30qvac_lib_inference_addon_llama11speculative22applySpeculativeConfigERKNS0_17SpeculativeConfigER13common_params",
             "__ZN14TextLlmContext19appendResidentTokenEi",
             "__ZN30qvac_lib_inference_addon_llama8batching7Request23isOptAwaitingCheckpointERKNSt3__18optionalIS1_EE",
             "__ZN30qvac_lib_inference_addon_llama5utils20getModelArchitectureEPK11llama_model",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17hasPendingCancelsEv",
             "__ZNK13ModelMetaData21hasOneBitQuantizationEv",
+            "__ZN14MtmdLlmContext25evalAudioChunkSpeculativeEPK16mtmd_input_chunkiiPi",
             "__ZThn48_N14MtmdLlmContext12syncPositionEi",
             "__ZNK30qvac_lib_inference_addon_llama8batching7Request17isPrefillCompleteEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler15StepUnlockGuardC1ERS1_PNSt3__111unique_lockINS4_5mutexEEE",
@@ -557,6 +603,7 @@
             "__ZNK30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher22nextAwaitingCheckpointEv",
             "__ZN14MtmdLlmContext32snapshotPreRequestRollbackAnchorEv",
             "__ZN14LlamaFinetuner24validateFinetuningParamsERKN30qvac_lib_inference_addon_llama21LlamaFinetuningParamsE",
+            "__ZN14MtmdLlmContext15checkAntipromptEmi",
             "__ZN14TextLlmContext15emitOutputPieceERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEES9_",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshotC2EOS1_",
             "__ZN13ModelMetaData28FirstFileFromGgufStreamState7provideER13BorrowablePtrINSt3__115basic_streambufIcNS2_11char_traitsIcEEEEE",
@@ -571,13 +618,13 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler6cancelEjy",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17resetRuntimeStatsEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler16saveConversationERKNSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEE",
-            "__ZN14MtmdLlmContextC2ER13common_paramsNSt3__110unique_ptrI18common_init_resultNS2_14default_deleteIS4_EEEEy",
+            "__ZN30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence13setBatchStartEi",
             "__ZN14TextLlmContext10resetStateEb",
-            "__ZThn48_N14MtmdLlmContext14preparePrefillERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERKNS1_INS1_IhNS3_IhEEEENS3_ISE_EEEERKNS1_I12PlannedMediaNS3_ISJ_EEEEbb",
             "__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",
             "__ZN17backend_selection17tryMainGpuFromMapERNSt3__113unordered_mapINS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES7_NS0_4hashIS7_EENS0_8equal_toIS7_EENS5_INS0_4pairIKS7_S7_EEEEEE",
             "__ZN14LlamaFinetuner30setPausedCheckpointStateSharedENSt3__110shared_ptrIN24llama_finetuning_helpers23TrainingCheckpointStateEEE",
-            "__ZThn48_NK14TextLlmContext25getToolDefinitionsDroppedEv",
             "__ZN30qvac_lib_inference_addon_llama5utils21SequenceStateSnapshot14seedForTestingENSt3__112basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEEi",
             "__ZN19LlamaBackendsHandleC1ERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_",
             "__ZN10LlamaModel7processERKNSt3__13anyEy",
@@ -592,6 +639,7 @@
             "__ZN14TextLlmContext22configureReasoningTagsERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEES8_RKNS0_8optionalIN30qvac_lib_inference_addon_llama5utils13ReasoningTagsEEE",
             "__ZN12CacheManager15hasTurnsToFlushEv",
             "__ZThn48_N14MtmdLlmContext18adoptResidentStateERKNSt3__16vectorIiNS0_9allocatorIiEEEE",
+            "__ZNK30qvac_lib_inference_addon_llama11speculative19SpeculativeSequence19useTargetCheckpointEm",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler10workerLoopEv",
             "__ZN17image_pixel_limit9checkFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEy",
             "__ZN14MtmdLlmContext27resetToolDefinitionsDroppedEv",
@@ -610,7 +658,7 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler21serviceSaveJobsLockedEv",
             "__ZN10LlamaModel31closeFinetuneCancellationWindowEv",
             "__ZN30qvac_lib_inference_addon_llama7logging17setVerbosityLevelERNSt3__113unordered_mapINS1_12basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEES8_NS1_4hashIS8_EENS1_8equal_toIS8_EENS6_INS1_4pairIKS8_S8_EEEEEE",
-            "__ZN24llama_finetuning_helpers27clearCurrentCheckpointStateEv",
+            "__ZN14MtmdLlmContext16emitSampledTokenEijRKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEEP10LlamaBatchm",
             "__ZN30qvac_lib_inference_addon_llama5utils20RequestRollbackState7restoreEP13llama_contexti",
             "__ZNK13ModelMetaData16checkInitializedEv",
             "__ZN12CacheManager24moveActiveCacheToRamTierEv",
@@ -618,7 +666,7 @@
             "__ZNK14TextLlmContext19residentStateTokensEv",
             "__ZN14TextLlmContext29clearCacheReconciliationStateEv",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler18failSaveJobsLockedEv",
-            "__ZN30qvac_lib_inference_addon_llama7parsers26parseReasoningBudgetConfigENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
+            "__ZN24llama_finetuning_helpers27clearCurrentCheckpointStateEv",
             "__ZN14MtmdLlmContext18adoptResidentStateERKNSt3__16vectorIiNS0_9allocatorIiEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching7RequestC2EjONSt3__16vectorIiNS2_9allocatorIiEEEEji",
             "__ZNK10LlamaModel13isBitnetModelEv",
@@ -628,11 +676,12 @@
             "__ZNK14TextLlmContext9saveCacheERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZThn48_NK14MtmdLlmContext28appendCheckpointsToCacheFileERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching19MultiRequestBatcher6cancelEjRKNSt3__18functionIFvjEEE",
-            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler18notifyDoneNoexceptEj",
+            "__ZN30qvac_lib_inference_addon_llama7parsers26parseReasoningBudgetConfigENSt3__117basic_string_viewIcNS1_11char_traitsIcEEEE",
             "__ZN30qvac_lib_inference_addon_llama8batching22aggregateObservedStatsERKNSt3__16vectorINS0_20ObservedRequestStatsENS1_9allocatorIS3_EEEE",
             "__ZN14MtmdLlmContext11evalMessageERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEEbb",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler23cancelKeyDeferredLockedEv",
             "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler13hasWorkLockedEv",
+            "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler18notifyDoneNoexceptEj",
             "__ZN30qvac_lib_inference_addon_llama5utils14isHarmonyModelEPK11llama_model",
             "__ZN10LlamaModel10resetStateEb",
             "__ZN10LlamaModel22processConcurrentBatchERKNSt3__16vectorINS_6PromptENS0_9allocatorIS2_EEEEy",
@@ -640,6 +689,8 @@
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler17prefillCompleteFnEv",
             "__ZN14MtmdLlmContext9loadMediaERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEE",
             "__ZN14MtmdLlmContext14preparePrefillERKNSt3__16vectorI15common_chat_msgNS0_9allocatorIS2_EEEERKNS1_I16common_chat_toolNS3_IS8_EEEERKNS1_INS1_IhNS3_IhEEEENS3_ISE_EEEERKNS1_I12PlannedMediaNS3_ISJ_EEEEbb",
+            "__ZNK30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler24teardownPendingForLockedEj",
+            "__ZN14TextLlmContext19generateSpeculativeERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEERj",
             "__Z33qvacLibInferenceAddonLlamaExportsP8js_env_sP10js_value_s",
             "__ZN30qvac_lib_inference_addon_llama8batching24ContinuousBatchScheduler25ensureWorkerStartedLockedEv",
             "__ZN14MtmdLlmContext13onSequenceEndERKNSt3__18functionIFvRKNS0_12basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEEEEE"
@@ -651,8 +702,8 @@
             "__ZN30qvac_lib_inference_addon_llama6cancelEP8js_env_sP18js_callback_info_s",
             "__ZN30qvac_lib_inference_addon_llama15fitResultObjectEP8js_env_sRKNS_9FitResultE",
             "__ZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask13onEnvTeardownINSt3__19monostateEEEvP22js_deferred_teardown_sPv",
-            "__ZN30qvac_lib_inference_addon_llama11activeSlotsEP8js_env_sP18js_callback_info_s",
             "__ZN28qvac_lib_inference_addon_cpp2js11JsAsyncTask13onCloseHandleIN30qvac_lib_inference_addon_llama9FitResultEEEvP11uv_handle_s",
+            "__ZN30qvac_lib_inference_addon_llama11activeSlotsEP8js_env_sP18js_callback_info_s",
             "__ZN28qvac_lib_inference_addon_cpp6logger8JsLogger13asyncCallbackEP10uv_async_s",
             "__ZN28qvac_lib_inference_addon_cpp11JsInterface6cancelEP8js_env_sP18js_callback_info_s",
             "__ZN28qvac_lib_inference_addon_cpp11JsInterface10activeJobsEP8js_env_sP18js_callback_info_s",

package/prebuilds/ios-arm64-simulator/qvac__llm-llamacpp.bare CHANGED Viewed

Binary file