remove group_beam_searcher.hpp; copy fast_tokenizer

Wovchena · May 22, 2024 · 75b7c37 · 75b7c37
1 parent 82a9449
commit 75b7c37
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 18 deletions.
diff --git a/src/cpp/CMakeLists.txt b/src/cpp/CMakeLists.txt
@@ -49,8 +49,6 @@ add_library(${TARGET_NAME} SHARED ${SOURCE_FILES})
 add_library(openvino::${TARGET_NAME} ALIAS ${TARGET_NAME})
 
 target_include_directories(${TARGET_NAME}
-    # TODO: remove it, because beam_search algo should not be exposed to end users
-    PRIVATE "$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/../../text_generation/causal_lm/cpp/>"
     PUBLIC "$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>" "$<INSTALL_INTERFACE:runtime/include>")
 
 target_link_libraries(${TARGET_NAME} PUBLIC openvino::runtime PRIVATE nlohmann_json::nlohmann_json jinja2cpp)
@@ -66,6 +64,13 @@ add_custom_command(TARGET ${TARGET_NAME} POST_BUILD
         "${CMAKE_CURRENT_SOURCE_DIR}/../python/openvino_genai/$<TARGET_FILE_NAME:${TARGET_NAME}>"
     COMMENT "Copy ${TARGET_NAME} to src/python/openvino_genai")
 
+# Copy libcore_tokenizers.so to build_dir/openvino_tokenizers/src/
+add_custom_command(TARGET ${TARGET_NAME} POST_BUILD
+    COMMAND "${CMAKE_COMMAND}" -E copy
+        "${CMAKE_BINARY_DIR}/_deps/fast_tokenizer-src/lib/libcore_tokenizers.so"
+        "${CMAKE_BINARY_DIR}/openvino_tokenizers/src/"
+    COMMENT "Copy libcore_tokenizers.so to build_dir/openvino_tokenizers/src/")
+
 install(TARGETS ${TARGET_NAME} LIBRARY DESTINATION . COMPONENT core_genai RUNTIME DESTINATION . COMPONENT core_genai)
 
 # - Windows: `<openvino_dir>\runtime\bin\intel64\Release\`

diff --git a/src/cpp/src/group_beam_searcher.cpp b/src/cpp/src/group_beam_searcher.cpp
@@ -2,8 +2,8 @@
 // SPDX-License-Identifier: Apache-2.0
 
 #include <openvino/runtime/tensor.hpp>
-#include "group_beam_searcher.hpp"
 #include "generation_config_helper.hpp"
+#include "openvino/genai/llm_pipeline.hpp"
 #include "utils.hpp"
 
 namespace {

diff --git a/src/cpp/src/group_beam_searcher.hpp b/src/cpp/src/group_beam_searcher.hpp
diff --git a/src/cpp/src/llm_pipeline.cpp b/src/cpp/src/llm_pipeline.cpp
@@ -14,7 +14,6 @@
 #include "openvino/genai/llm_pipeline.hpp"
 #include "utils.hpp"
 #include "generation_config_helper.hpp"
-#include "group_beam_searcher.hpp"
 #include "text_callback_streamer.hpp"
 
 
@@ -29,6 +28,8 @@ ov::EncodedResults greedy_decoding(
     bool is_chat_conversation = false
 );
 
+EncodedResults beam_search(ov::InferRequest& lm, ov::Tensor prompts, ov::Tensor attentin_mask, GenerationConfig config);
+
 
 class LLMPipeline::LLMPipelineImpl {
 public:

diff --git a/tests/python_tests/test_generate_api.py b/tests/python_tests/test_generate_api.py
@@ -28,10 +28,10 @@ def run_hf_ov_genai_comparison(model_fixture, generation_config, prompt):
     hf_output = tokenizer.decode(hf_encoded_output[0, encoded_prompt.shape[1]:])
 
     device = 'CPU'
-    ov_tokenziers_path = '../../build/openvino_tokenizers/src/'
+    ov_tokenizers_path = '../../build/openvino_tokenizers/src/'
     import openvino_genai as ov_genai
 
-    pipe = ov_genai.LLMPipeline(path, device, {}, ov_tokenziers_path)
+    pipe = ov_genai.LLMPipeline(path, device, {}, ov_tokenizers_path)
     ov_output = pipe.generate(prompt, **generation_config)
 
     if hf_output != ov_output: