Class NativeAPI.EngineSettings
public static class NativeAPI.EngineSettings
- Inheritance
-
objectNativeAPI.EngineSettings
Methods
litert_lm_engine_settings_create(string, string, string?, string?)
Creates LiteRT LM Engine Settings. The caller is responsible for destroying the settings using litert_lm_engine_settings_delete(nint).
public static extern nint litert_lm_engine_settings_create(string modelPath, string backendStr, string? visionBackendStr, string? audioBackendStr)
Parameters
modelPathstringThe path to the model file.
backendStrstringThe backend to use (e.g., "cpu", "gpu").
visionBackendStrstringThe vision backend to use, or null if not set.
audioBackendStrstringThe audio backend to use, or null if not set.
Returns
- nint
A pointer to the created settings, or nint.Zero on failure.
litert_lm_engine_settings_create_from_raw_file_descriptor(int, string, string?, string?)
Creates LiteRT LM Engine Settings from a raw file descriptor. The engine takes ownership of the file descriptor and will close it when done. The caller is responsible for destroying the settings using litert_lm_engine_settings_delete(nint).
public static extern nint litert_lm_engine_settings_create_from_raw_file_descriptor(int fd, string backendStr, string? visionBackendStr, string? audioBackendStr)
Parameters
fdintThe file descriptor of the model.
backendStrstringThe backend to use (e.g., "cpu", "gpu").
visionBackendStrstringThe vision backend to use, or null if not set.
audioBackendStrstringThe audio backend to use, or null if not set.
Returns
- nint
A pointer to the created settings, or nint.Zero on failure.
litert_lm_engine_settings_delete(nint)
Destroys LiteRT LM Engine Settings.
public static extern void litert_lm_engine_settings_delete(nint settings)
Parameters
settingsnintThe settings to destroy.
litert_lm_engine_settings_enable_benchmark(nint)
Enables benchmarking for the engine.
public static extern void litert_lm_engine_settings_enable_benchmark(nint settings)
Parameters
settingsnintThe engine settings.
litert_lm_engine_settings_set_activation_data_type(nint, ActivationDataType)
Sets the activation data type.
public static extern void litert_lm_engine_settings_set_activation_data_type(nint settings, ActivationDataType activationDataType)
Parameters
settingsnintThe engine settings.
activationDataTypeActivationDataTypeThe activation data type.
litert_lm_engine_settings_set_audio_lora_rank(nint, int)
Sets the Audio LoRA rank for the engine.
public static extern void litert_lm_engine_settings_set_audio_lora_rank(nint settings, int loraRank)
Parameters
settingsnintThe engine settings.
loraRankintThe Audio LoRA rank.
litert_lm_engine_settings_set_audio_num_threads(nint, int)
Sets the number of threads for the audio CPU backend.
public static extern void litert_lm_engine_settings_set_audio_num_threads(nint settings, int numThreads)
Parameters
settingsnintThe engine settings.
numThreadsintThe number of threads.
litert_lm_engine_settings_set_cache_dir(nint, string)
Sets the cache directory for the engine.
public static extern void litert_lm_engine_settings_set_cache_dir(nint settings, string cacheDir)
Parameters
settingsnintThe engine settings.
cacheDirstringThe cache directory.
litert_lm_engine_settings_set_enable_speculative_decoding(nint, bool)
Sets whether to enable speculative decoding.
public static extern void litert_lm_engine_settings_set_enable_speculative_decoding(nint settings, bool enableSpeculativeDecoding)
Parameters
settingsnintThe engine settings.
enableSpeculativeDecodingboolWhether to enable speculative decoding.
litert_lm_engine_settings_set_litert_dispatch_lib_dir(nint, string)
Sets the LiteRT dispatch library directory for NPU backend.
public static extern void litert_lm_engine_settings_set_litert_dispatch_lib_dir(nint settings, string libDir)
Parameters
settingsnintThe engine settings.
libDirstringThe dispatch library directory.
litert_lm_engine_settings_set_lora_rank(nint, int)
Sets the LoRA rank for the engine.
public static extern void litert_lm_engine_settings_set_lora_rank(nint settings, int loraRank)
Parameters
settingsnintThe engine settings.
loraRankintThe LoRA rank.
litert_lm_engine_settings_set_max_num_images(nint, int)
Sets the maximum number of images for the engine. This is only used for the legacy implementation of the engine.
public static extern void litert_lm_engine_settings_set_max_num_images(nint settings, int maxNumImages)
Parameters
settingsnintThe engine settings.
maxNumImagesintThe maximum number of images.
litert_lm_engine_settings_set_max_num_tokens(nint, int)
Sets the maximum number of tokens for the engine.
public static extern void litert_lm_engine_settings_set_max_num_tokens(nint settings, int maxNumTokens)
Parameters
settingsnintThe engine settings.
maxNumTokensintThe maximum number of tokens.
litert_lm_engine_settings_set_num_decode_tokens(nint, int)
Sets the number of decode tokens for benchmarking.
public static extern void litert_lm_engine_settings_set_num_decode_tokens(nint settings, int numDecodeTokens)
Parameters
settingsnintThe engine settings.
numDecodeTokensintThe number of decode tokens.
litert_lm_engine_settings_set_num_prefill_tokens(nint, int)
Sets the number of prefill tokens for benchmarking.
public static extern void litert_lm_engine_settings_set_num_prefill_tokens(nint settings, int numPrefillTokens)
Parameters
settingsnintThe engine settings.
numPrefillTokensintThe number of prefill tokens.
litert_lm_engine_settings_set_num_threads(nint, int)
Sets the number of threads for the CPU backend.
public static extern void litert_lm_engine_settings_set_num_threads(nint settings, int numThreads)
Parameters
settingsnintThe engine settings.
numThreadsintThe number of threads.
litert_lm_engine_settings_set_parallel_file_section_loading(nint, bool)
Sets whether the engine should load different sections of the litertlm file in parallel. Defaults to true.
public static extern void litert_lm_engine_settings_set_parallel_file_section_loading(nint settings, bool parallelFileSectionLoading)
Parameters
settingsnintThe engine settings.
parallelFileSectionLoadingboolWhether to load in parallel.
litert_lm_engine_settings_set_prefill_chunk_size(nint, int)
Sets the prefill chunk size for the engine. Only applicable for CPU backend with dynamic models.
public static extern void litert_lm_engine_settings_set_prefill_chunk_size(nint settings, int prefillChunkSize)
Parameters
settingsnintThe engine settings.
prefillChunkSizeintThe prefill chunk size.
litert_lm_engine_settings_set_supported_audio_lora_ranks(nint, int[], nuint)
Sets the supported Audio LoRA ranks for the engine.
public static extern int litert_lm_engine_settings_set_supported_audio_lora_ranks(nint settings, int[] loraRanks, nuint numRanks)
Parameters
settingsnintThe engine settings.
loraRanksint[]An array of supported Audio LoRA ranks.
numRanksnuintThe number of ranks in the array.
Returns
- int
0 on success, non-zero on failure.
litert_lm_engine_settings_set_supported_lora_ranks(nint, int[], nuint)
Sets the supported LoRA ranks for the engine.
public static extern int litert_lm_engine_settings_set_supported_lora_ranks(nint settings, int[] loraRanks, nuint numRanks)
Parameters
settingsnintThe engine settings.
loraRanksint[]An array of supported LoRA ranks.
numRanksnuintThe number of ranks in the array.
Returns
- int
0 on success, non-zero on failure.