/aosp_15_r20/external/executorch/examples/models/llava/runner/ |
H A D | llava_runner.cpp | 16 #include <executorch/extension/llm/tokenizer/bpe_tokenizer.h> 23 namespace llm = ::executorch::extension::llm; 43 stats_.model_load_start_ms = llm::time_in_ms(); in load() 46 tokenizer_ = std::make_unique<llm::BPETokenizer>(); in load() 55 text_prefiller_ = std::make_unique<llm::TextPrefiller>( in load() 65 text_token_generator_ = std::make_unique<llm::TextTokenGenerator>( in load() 73 stats_.model_load_end_ms = llm::time_in_ms(); in load() 78 std::vector<llm::Image>& images, in prefill_images() 103 std::function<void(const ::executorch::extension::llm::Stats&)> in generate_from_pos() 113 stats_.first_token_ms = llm::time_in_ms(); in generate_from_pos() [all …]
|
H A D | llava_runner.h | 9 // A simple multimodal LLM runner that includes preprocessing and post 20 #include <executorch/extension/llm/runner/multimodal_runner.h> 25 : public ::executorch::extension::llm::MultimodalRunner { 35 std::vector<::executorch::extension::llm::Image> images, 39 std::function<void(const ::executorch::extension::llm::Stats&)> 46 * @param start_pos The starting position in KV cache of the input in the LLM. 51 std::vector<::executorch::extension::llm::Image>& images, 57 * @param start_pos The starting position in KV cache of the input in the LLM. 74 * @param start_pos The starting position in KV cache of the input in the LLM. 85 std::function<void(const ::executorch::extension::llm::Stats&)>
|
H A D | CMakeLists.txt | 31 "${EXECUTORCH_ROOT}/extension/llm/sampler/sampler.cpp" 32 "${EXECUTORCH_ROOT}/extension/llm/tokenizer/bpe_tokenizer.cpp" 35 # extension llm runner lib 37 ${CMAKE_CURRENT_SOURCE_DIR}/../../../../extension/llm/runner 38 ${CMAKE_CURRENT_BINARY_DIR}/../../../../extension/llm/runner
|
/aosp_15_r20/external/executorch/examples/models/llama/runner/ |
H A D | runner.cpp | 16 #include <executorch/extension/llm/runner/util.h> 19 #include <executorch/extension/llm/tokenizer/bpe_tokenizer.h> 27 namespace llm = ::executorch::extension::llm; 84 tokenizer_ = std::make_unique<llm::BPETokenizer>(); in load() 123 text_decoder_runner_ = std::make_unique<llm::TextDecoderRunner>( in load() 128 text_prefiller_ = std::make_unique<llm::TextPrefiller>( in load() 133 text_token_generator_ = std::make_unique<llm::TextTokenGenerator>( in load() 155 std::function<void(const llm::Stats&)> stats_callback, in generate() 162 stats_.model_load_start_ms = llm::time_in_ms(); in generate() 164 stats_.model_load_end_ms = llm::time_in_ms(); in generate() [all …]
|
H A D | runner.h | 20 #include <executorch/extension/llm/runner/irunner.h> 21 #include <executorch/extension/llm/runner/stats.h> 22 #include <executorch/extension/llm/runner/text_decoder_runner.h> 23 #include <executorch/extension/llm/runner/text_prefiller.h> 24 #include <executorch/extension/llm/runner/text_token_generator.h> 25 #include <executorch/extension/llm/tokenizer/tokenizer.h> 30 class ET_EXPERIMENTAL Runner : public executorch::extension::llm::IRunner { 43 std::function<void(const ::executorch::extension::llm::Stats&)> 59 std::unique_ptr<::executorch::extension::llm::Tokenizer> tokenizer_; 61 std::unique_ptr<::executorch::extension::llm::TextDecoderRunner> [all …]
|
H A D | targets.bzl | 8 …orch/configurations:optimized_native_cpu_ops_oss", "//executorch/extension/llm/custom_ops:custom_o… 10 …return ["//executorch/configurations:optimized_native_cpu_ops", "//executorch/extension/llm/custom… 42 "//executorch/extension/llm/runner:irunner", 43 "//executorch/extension/llm/runner:stats", 44 "//executorch/extension/llm/runner:text_decoder_runner" + aten_suffix, 45 "//executorch/extension/llm/runner:text_prefiller" + aten_suffix, 46 "//executorch/extension/llm/runner:text_token_generator" + aten_suffix, 54 "//executorch/extension/llm/tokenizer:bpe_tokenizer",
|
/aosp_15_r20/external/executorch/extension/llm/runner/ |
H A D | multimodal_runner.h | 9 // A simple multimodal LLM runner that includes preprocessing and post 22 #include <executorch/extension/llm/runner/image.h> 23 #include <executorch/extension/llm/runner/image_prefiller.h> 24 #include <executorch/extension/llm/runner/stats.h> 25 #include <executorch/extension/llm/runner/text_decoder_runner.h> 26 #include <executorch/extension/llm/runner/text_prefiller.h> 27 #include <executorch/extension/llm/runner/text_token_generator.h> 28 #include <executorch/extension/llm/sampler/sampler.h> 29 #include <executorch/extension/llm/tokenizer/tokenizer.h> 34 namespace llm { [all …]
|
H A D | text_prefiller.h | 10 // LLM. 14 #include <executorch/extension/llm/runner/text_decoder_runner.h> 15 #include <executorch/extension/llm/tokenizer/tokenizer.h> 20 namespace llm { 29 * Prefill an LLM Module with the given text input. 30 * @param prompt_tokens The text prompt tokens to the LLM Module. Encoded by 32 * @param start_pos The starting position in KV cache of the input in the LLM 34 * @return The next token of the LLM Module after prefill. 46 } // namespace llm 54 using ::executorch::extension::llm::TextPrefiller;
|
H A D | image_prefiller.h | 9 // Given a image tensor, prefill the KV cache of a multimodal LLM. 13 #include <executorch/extension/llm/runner/image.h> 19 namespace llm { 28 * Prefill an LLM Module with the given image input. 29 * @param image The image input to the multimodal LLM. 30 * @param start_pos The starting position in KV cache of the input in the LLM. 32 * @return The next token of the LLM Module after prefill. 47 } // namespace llm 55 using ::executorch::extension::llm::ImagePrefiller;
|
H A D | text_decoder_runner.h | 9 // Given inputs, run a text decoder in LLM and return the output. 13 #include <executorch/extension/llm/sampler/sampler.h> 21 namespace llm { 34 * Run LLM text decoder with inputs to generate next token. 35 * @param input The input to the LLM Module. 36 * @param start_pos The starting position in KV cache of the input in the LLM 38 * @return The output of the LLM Module. This will be a tensor of logits. 106 } // namespace llm 114 using ::executorch::extension::llm::TextDecoderRunner;
|
H A D | stats.h | 9 // Runner stats for LLM 11 #include <executorch/extension/llm/runner/util.h> 19 namespace llm { 163 } // namespace llm 168 namespace llm { 171 using ::executorch::extension::llm::kTopp; 172 using ::executorch::extension::llm::print_report; 173 using ::executorch::extension::llm::Stats; 174 } // namespace llm
|
H A D | text_decoder_runner.cpp | 11 #include <executorch/extension/llm/runner/text_decoder_runner.h> 15 #include <executorch/extension/llm/runner/stats.h> 19 namespace llm { namespace 49 "More then one output returned from executing LLM."); in step() 52 "Non Tensor Output returned from executing LLM"); in step() 63 "More then one output returned from executing LLM."); in step() 66 "Non Tensor Output returned from executing LLM"); in step() 73 } // namespace llm
|
H A D | text_token_generator.h | 12 #include <executorch/extension/llm/runner/stats.h> 13 #include <executorch/extension/llm/runner/text_decoder_runner.h> 14 #include <executorch/extension/llm/tokenizer/tokenizer.h> 19 namespace llm { 145 } // namespace llm 153 using ::executorch::extension::llm::TextTokenGenerator;
|
H A D | irunner.h | 9 // An interface for LLM runners. Developers can create their own runner that 17 #include <executorch/extension/llm/runner/stats.h> 22 namespace llm { 39 std::function<void(const ::executorch::extension::llm::Stats&)> 48 } // namespace llm
|
H A D | util.h | 20 namespace llm { 72 } // namespace llm 81 using ::executorch::extension::llm::get_rss_bytes; 82 using ::executorch::extension::llm::safe_printf; 83 using ::executorch::extension::llm::time_in_ms;
|
/aosp_15_r20/external/executorch/extension/android/jni/ |
H A D | jni_layer_llama.cpp | 18 #include <executorch/extension/llm/runner/image.h> 19 #include <executorch/extension/llm/runner/irunner.h> 36 namespace llm = ::executorch::extension::llm; 101 void onStats(const llm::Stats& result) const { in onStats() 119 std::unique_ptr<llm::IRunner> runner_; 120 std::unique_ptr<llm::MultimodalRunner> multi_modal_runner_; 173 // Interpret the model type as LLM in ExecuTorchLlamaJni() 190 std::vector<llm::Image> images; in generate() 198 llm::Image image_runner{image_data, width, height, channels}; in generate() 206 [callback](const llm::Stats& result) { callback->onStats(result); }, in generate() [all …]
|
/aosp_15_r20/external/googleapis/google/cloud/dialogflow/cx/v3beta1/ |
H A D | generative_settings.proto | 45 // Prompt text that is sent to a LLM on no-match default, placeholders are 63 // Settings for knowledge connector. These parameters are used for LLM prompt 69 // represents. Used for knowledge connector LLM prompt and for knowledge 73 // Name of the virtual agent. Used for LLM prompt. Can be left empty. 79 // Company description, used for LLM prompt, e.g. "a family company selling 87 // Whether to disable fallback to Data Store search results (in case the LLM 108 // LLM model settings. 112 // Settings for LLM models. 114 // The selected LLM model.
|
/aosp_15_r20/external/executorch/examples/models/llama/ |
H A D | TARGETS | 59 "//executorch/extension/llm/custom_ops:model_sharding_py", 60 "//executorch/extension/llm/custom_ops:custom_ops_aot_lib", 111 "//executorch/extension/llm/custom_ops:custom_ops_aot_py", 112 "//executorch/extension/llm/export:export_lib", 156 "//executorch/extension/llm/export:export_lib", 157 "//executorch/extension/llm/tokenizer:tokenizer_py_lib", 192 "//executorch/extension/llm/custom_ops:custom_ops_aot_lib", 207 "//executorch/extension/llm/custom_ops:custom_ops_aot_lib",
|
/aosp_15_r20/external/executorch/examples/mediatek/executor_runner/ |
H A D | mtk_llama_runner.h | 15 #include <executorch/extension/llm/runner/irunner.h> 16 #include <executorch/extension/llm/runner/stats.h> 17 #include <executorch/extension/llm/tokenizer/bpe_tokenizer.h> 18 #include <executorch/extension/llm/tokenizer/tiktoken.h> 26 using Stats = ::executorch::llm::Stats; 31 using executorch::extension::llm::Tokenizer; 35 class MTKLlamaRunner : public executorch::extension::llm::IRunner {
|
/aosp_15_r20/external/pytorch/.github/workflows/ |
H A D | llm_td_retrieval.yml | 11 llm-retrieval: 35 repository: osalpekar/llm-target-determinator 37 path: llm-target-determinator 55 cd "${GITHUB_WORKSPACE}/llm-target-determinator" 79 cd "${GITHUB_WORKSPACE}"/llm-target-determinator/assets 92 cd "${GITHUB_WORKSPACE}"/llm-target-determinator 110 path: llm-target-determinator/assets/mappings.zip
|
/aosp_15_r20/external/executorch/extension/llm/tokenizer/ |
H A D | tiktoken.h | 11 #include <executorch/extension/llm/tokenizer/tokenizer.h> 19 namespace llm { 80 } // namespace llm 88 using ::executorch::extension::llm::Decoder; 89 using ::executorch::extension::llm::Encoder; 90 using ::executorch::extension::llm::Re2UPtr; 91 using ::executorch::extension::llm::Tiktoken;
|
H A D | bpe_tokenizer.h | 11 #include <executorch/extension/llm/tokenizer/tokenizer.h> 16 namespace llm { 47 } // namespace llm 55 using ::executorch::extension::llm::BPETokenizer; 56 using ::executorch::extension::llm::TokenIndex;
|
/aosp_15_r20/external/googleapis/google/cloud/dialogflow/cx/v3/ |
H A D | generative_settings.proto | 45 // Prompt text that is sent to a LLM on no-match default, placeholders are 63 // Settings for knowledge connector. These parameters are used for LLM prompt 69 // represents. Used for knowledge connector LLM prompt and for knowledge 73 // Name of the virtual agent. Used for LLM prompt. Can be left empty. 79 // Company description, used for LLM prompt, e.g. "a family company selling 87 // Whether to disable fallback to Data Store search results (in case the LLM
|
/aosp_15_r20/external/executorch/extension/android/src/main/java/org/pytorch/executorch/ |
H A D | LlamaModule.java | 49 /** Constructs a LLM Module for a model with given path, tokenizer, and temperature. */ 132 * @param startPos The starting position in KV cache of the input in the LLM. 133 * @return The updated starting position in KV cache of the input in the LLM. 152 * @param startPos The starting position in KV cache of the input in the LLM. It's passed as 156 * @return The updated starting position in KV cache of the input in the LLM. 175 * @param startPos The starting position in KV cache of the input in the LLM.
|
/aosp_15_r20/external/executorch/.github/workflows/ |
H A D | upload-android-test-specs.yml | 7 - extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml 13 - extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml 35 path: extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml 48 …zonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/android-llm-device-farm-test-sp… 82 SPEC_FILE: android-llm-device-farm-test-spec.yml
|