fix python

samples/python/chat_sample/chat_sample.py

@@ -6,9 +6,8 @@
 import openvino_genai
 
 
-def streamer(subword) -> openvino_genai.StreamerRunningStatus:
+def streamer(subword):
     print(subword, end='', flush=True)
-
     return openvino_genai.StreamerRunningStatus.RUNNING
 
 def main():

samples/python/prompt_lookup_decoding_lm/prompt_lookup_decoding_lm.py

@@ -5,11 +5,10 @@
 import argparse
 import openvino_genai
 
-def streamer(subword): 
-        print(subword, end='', flush=True) 
-        # Return flag corresponds whether generation should be stopped. 
-        # False means continue generation. 
-        return False
+def streamer(subword):
+    print(subword, end='', flush=True) 
+    # Return flag corresponds whether generation should be stopped. 
+    return openvino_genai.StreamerRunningStatus.RUNNING
 
 def main():
     parser = argparse.ArgumentParser()

samples/python/speculative_decoding_lm/speculative_decoding_lm.py

@@ -8,10 +8,9 @@
 import threading
 
 def streamer(subword): 
-        print(subword, end='', flush=True) 
-        # Return flag corresponds whether generation should be stopped. 
-        # False means continue generation. 
-        return False
+    print(subword, end='', flush=True) 
+    # Return flag corresponds whether generation should be stopped. 
+    return openvino_genai.StreamerRunningStatus.RUNNING
 
 def main():
     parser = argparse.ArgumentParser()

samples/python/visual_language_chat/visual_language_chat.py

@@ -11,7 +11,7 @@
 from pathlib import Path
 
 
-def streamer(subword: str) -> bool:
+def streamer(subword: str) -> openvino_genai.StreamerRunningStatus:
     '''
 
     Args:
@@ -25,6 +25,8 @@ def streamer(subword: str) -> bool:
     # No value is returned as in this example we don't want to stop the generation in this method.
     # "return None" will be treated the same as "return False".
 
+    return openvino_genai.StreamerRunningStatus.RUNNING
+
 
 def read_image(path: str) -> Tensor:
     '''
@@ -66,7 +68,7 @@ def main():
     config = openvino_genai.GenerationConfig()
     config.max_new_tokens = 100
 
-    pipe.start_chat()
+    # pipe.start_chat()
     prompt = input('question:\n')
     pipe.generate(prompt, images=rgbs, generation_config=config, streamer=streamer)
 
@@ -77,7 +79,7 @@ def main():
         except EOFError:
             break
         pipe.generate(prompt, generation_config=config, streamer=streamer)
-    pipe.finish_chat()
+    # pipe.finish_chat()
 
 
 if '__main__' == __name__:

src/python/py_utils.cpp

@@ -336,15 +336,6 @@ ov::genai::StreamerVariant pystreamer_to_streamer(const PyBindStreamerVariant& p
     ov::genai::StreamerVariant streamer = std::monostate();
 
     std::visit(overloaded {
-    [&streamer](const std::function<bool(py::str)>& py_callback){
-        // Wrap python streamer with manual utf-8 decoding. Do not rely
-        // on pybind automatic decoding since it raises exceptions on incomplete strings.
-        auto callback_wrapped = [py_callback](std::string subword) -> bool {
-            auto py_str = PyUnicode_DecodeUTF8(subword.data(), subword.length(), "replace");
-            return py_callback(py::reinterpret_borrow<py::str>(py_str));
-        };
-        streamer = callback_wrapped;
-    },
     [&streamer](const std::function<ov::genai::StreamerRunningStatus(py::str)>& py_callback){
         // Wrap python streamer with manual utf-8 decoding. Do not rely
         // on pybind automatic decoding since it raises exceptions on incomplete strings.

src/python/py_utils.hpp

@@ -18,7 +18,7 @@ namespace ov::genai::pybind::utils {
 
 // When StreamerVariant is used utf-8 decoding is done by pybind and can lead to exception on incomplete texts.
 // Therefore strings decoding should be handled with PyUnicode_DecodeUTF8(..., "replace") to not throw errors.
-using PyBindStreamerVariant = std::variant<std::function<StreamerRunningStatus(py::str)>, std::function<bool(py::str)>, std::shared_ptr<StreamerBase>, std::monostate>;
+using PyBindStreamerVariant = std::variant<std::function<StreamerRunningStatus(py::str)>, std::shared_ptr<StreamerBase>, std::monostate>;
 
 template <class... Ts>
 struct overloaded : Ts... {