Merge branch 'master' into ak/wwb_no_model_comparison

openvinotoolkit · Nov 22, 2024 · dbc7cbd · dbc7cbd
2 parents 0225d10 + ff8846a
commit dbc7cbd
Show file tree

Hide file tree

Showing 5 changed files with 15 additions and 9 deletions.
diff --git a/samples/CMakeLists.txt b/samples/CMakeLists.txt
@@ -14,8 +14,12 @@ add_subdirectory(cpp/text2image)
 add_subdirectory(cpp/visual_language_chat)
 add_subdirectory(cpp/whisper_speech_recognition)
 
-install(FILES requirements.txt DESTINATION samples
-        COMPONENT cpp_samples_genai)
+install(FILES
+        deployment-requirements.txt
+        export-requirements.txt
+        requirements.txt
+    DESTINATION samples
+    COMPONENT cpp_samples_genai)
 
 install(DIRECTORY
             cpp/beam_search_causal_lm

diff --git a/src/python/CMakeLists.txt b/src/python/CMakeLists.txt
@@ -182,12 +182,14 @@ if(pybind11_stubgen_AVAILABLE)
         VERBATIM)
 
     add_custom_target(${TARGET_NAME}_stub ALL DEPENDS ${output_file})
-else()
+elseif(OpenVINODeveloperPackage_FOUND)
     # Produce warning message at build time as well
     add_custom_command(OUTPUT pybind11_stub_gen_not_found.txt
         COMMAND ${CMAKE_COMMAND}
             -E cmake_echo_color --red "Warning: Please, install ${pybind11_stubgen_dep}")
     add_custom_target(${TARGET_NAME}_stub ALL DEPENDS pybind11_stub_gen_not_found.txt)
+else()
+    add_custom_target(${TARGET_NAME}_stub ALL)
 endif()
 
 add_dependencies(${TARGET_NAME}_stub ${TARGET_NAME})
diff --git a/tools/llm_bench/llm_bench_utils/metrics_print.py b/tools/llm_bench/llm_bench_utils/metrics_print.py
@@ -149,15 +149,15 @@ def output_avg_statis_tokens(prompt_dict, prompt_idx_list, iter_data_list, batch
             avg_input_size = int(avg_input_size / index_num)
             if avg_2nd_tokens_latency > 0:
                 avg_2nd_token_tput = (1 / avg_2nd_tokens_latency) * batch_size * 1000
-            latency_unit = 'token' if is_text_gen is True else 'step'
+            tput_unit = latency_unit = 'token' if is_text_gen is True else 'step'
             if batch_size > 1:
                 if is_text_gen is True:
                     latency_unit = '{}tokens'.format(batch_size)
                 else:
                     latency_unit = '{}steps'.format(batch_size)
             avg_1st_token_latency = 'NA' if avg_1st_token_latency < 0 else f'{avg_1st_token_latency:.2f} ms/{latency_unit}'
             avg_2nd_tokens_latency = 'NA' if avg_2nd_tokens_latency < 0 else f'{avg_2nd_tokens_latency:.2f} ms/{latency_unit}'
-            avg_2nd_token_tput = 'NA' if avg_2nd_tokens_latency == 'NA' else f'{avg_2nd_token_tput:.2f} {latency_unit}s/s'
+            avg_2nd_token_tput = 'NA' if avg_2nd_tokens_latency == 'NA' else f'{avg_2nd_token_tput:.2f} {tput_unit}s/s'
             prefix = f'[ INFO ] [Average] P[{p_idx}]L[{loop_idx}]' if loop_idx != -1 else f'[ INFO ] [Average] P[{p_idx}]'
             if is_text_gen is True:
                 output_info = ''

diff --git a/tools/llm_bench/task/speech_to_text_generation.py b/tools/llm_bench/task/speech_to_text_generation.py
@@ -51,10 +51,10 @@ def run_speech_2_txt_generation(input_param, args, md5_list, iter_data_list):
         )
         end = time.perf_counter()
         perf_metrics = result_text.perf_metrics
-        first_token_time = perf_metrics.get_ttft().mean / args["batch_size"]
+        first_token_time = perf_metrics.get_ttft().mean
         second_tokens_durations = (
             np.array(perf_metrics.raw_metrics.m_new_token_times[1:])
-            - np.array(perf_metrics.raw_metrics.m_new_token_times[:-1]) / args["batch_size"]
+            - np.array(perf_metrics.raw_metrics.m_new_token_times[:-1])
         ).tolist()
         tm_list = (np.array([first_token_time] + second_tokens_durations) / 1000).tolist()
         tm_infer_list = []

diff --git a/tools/llm_bench/task/text_generation.py b/tools/llm_bench/task/text_generation.py
@@ -240,10 +240,10 @@ def run_text_generation_genai(input_text, num, model, tokenizer, args, iter_data
         per_token_time = generation_time * 1000 / (num_tokens / args['batch_size'])
     else:
         log.warning("No generated tokens")
-    first_token_time = (perf_metrics.get_ttft().mean - perf_metrics.raw_metrics.tokenization_durations[-1] / 1000) / args["batch_size"]
+    first_token_time = (perf_metrics.get_ttft().mean - perf_metrics.raw_metrics.tokenization_durations[-1] / 1000) * args["batch_size"]
     second_tokens_durations = (
         np.array(perf_metrics.raw_metrics.m_new_token_times[1:])
-        - np.array(perf_metrics.raw_metrics.m_new_token_times[:-1]) / args["batch_size"]
+        - np.array(perf_metrics.raw_metrics.m_new_token_times[:-1])
     ).tolist()
 
     tm_list = np.array([first_token_time] + second_tokens_durations) / 1000