LlamaCppModelParams constructor

const LlamaCppModelParams({
  1. int contextSize = 4096,
  2. ComputeDevice device = ComputeDevice.auto,
  3. int gpuLayers = maxGpuLayers,
  4. GpuBackend preferredBackend = GpuBackend.auto,
  5. ModelSplitMode splitMode = ModelSplitMode.layer,
  6. int mainGpu = 0,
  7. List<LoraAdapterConfig> loras = const [],
  8. String? chatTemplate,
  9. int numberOfThreads = 0,
  10. int numberOfThreadsBatch = 0,
  11. int batchSize = 0,
  12. int microBatchSize = 0,
  13. int maxParallelSequences = 1,
  14. int speculativeRollbackTokenMax = 0,
  15. bool useMmap = true,
  16. bool useMlock = false,
  17. bool loadMtp = false,
  18. FlashAttention flashAttention = FlashAttention.auto,
  19. KvCacheType cacheTypeK = KvCacheType.f16,
  20. KvCacheType cacheTypeV = KvCacheType.f16,
  21. bool? kvUnified,
  22. double? ropeFrequencyBase,
  23. double? ropeFrequencyScale,
  24. bool? preferMemory64,
  25. int? modelBytesHint,
})

Creates llama.cpp load parameters with ModelParams' defaults.

Implementation

const LlamaCppModelParams({
  super.contextSize,
  super.device,
  super.gpuLayers,
  super.preferredBackend,
  super.splitMode,
  super.mainGpu,
  super.loras,
  super.chatTemplate,
  super.numberOfThreads,
  super.numberOfThreadsBatch,
  super.batchSize,
  super.microBatchSize,
  super.maxParallelSequences,
  super.speculativeRollbackTokenMax,
  super.useMmap,
  super.useMlock,
  super.loadMtp,
  super.flashAttention,
  super.cacheTypeK,
  super.cacheTypeV,
  super.kvUnified,
  super.ropeFrequencyBase,
  super.ropeFrequencyScale,
  super.preferMemory64,
  super.modelBytesHint,
});