feat: chain of thought

main.py

@@ -1,3 +1,4 @@
+import importlib
 import os
 import re
 
@@ -8,7 +9,7 @@ from src.prompt_system import system_base_definition
 from src.prompt_tasks import general_guidelines, executor_file_task, requirements_file_task, \
     test_executor_file_task, docker_file_task, client_file_task, streamlit_file_task
 from src.utils.io import recreate_folder
-from src.utils.string import find_differences
+from src.utils.string_tools import find_differences
 
 
 def extract_content_from_result(plain_text, file_name):
@@ -17,7 +18,7 @@ def extract_content_from_result(plain_text, file_name):
     if match:
         return match.group(1).strip()
     else:
-        return None
+        raise ValueError(f'Could not find {file_name} in result')
 
 
 def extract_and_write(plain_text, dest_folder):
@@ -28,7 +29,7 @@ def extract_and_write(plain_text, dest_folder):
             f.write(clean)
 
 
-def write_config_yml(executor_name):
+def write_config_yml(executor_name, dest_folder):
     config_content = f'''
 jtype: {executor_name}
 py_modules:
@@ -36,7 +37,7 @@ py_modules:
 metas:
   name: {executor_name}
     '''
-    with open('executor/config.yml', 'w') as f:
+    with open(os.path.join(dest_folder, 'config.yml'), 'w') as f:
         f.write(config_content)
 
 
@@ -69,7 +70,9 @@ def build_prototype_implementation(executor_description, executor_name, input_do
             + docker_file_task()
             + client_file_task()
             + streamlit_file_task()
-            + "First, write down some non-obvious thoughts about the challenges of the task and how you handle them. "
+            + "First, write down some non-obvious thoughts about the challenges of the task and give multiple approaches on how you handle them. "
+              "For example, there are different libraries you could use. "
+              "Discuss the pros and cons for all of these approaches and then decide for one of the approaches. "
               "Then write as I told you. "
     )
     plain_text = gpt.get_response(system_definition, user_query)
@@ -81,14 +84,18 @@ def build_production_ready_implementation(all_executor_files_string):
             system_base_definition
             + f"The user gives you the code of the executor and all other files needed ({', '.join([e[0] for e in FILE_AND_TAG_PAIRS])}) "
               f"The files may contain bugs. Fix all of them. "
-              f"Some files might have only prototype implementations and are not production ready. Add all the missing code. "
-              f"Some imports might be missing. Make sure to add them. Output all the files in the same format like given to you. "
+
     )
     user_query = (
-        "Fix all files, add all missing code and imports. Make it production ready. "
+        'Make it production ready. '
+        "Fix all files and add all missing code. "
         "Keep the same format as given to you. "
-        "First write down some non-obvious thoughts about what parts could need an adjustment and why. "
-        "Then write as I told you. "
+        f"Some files might have only prototype implementations and are not production ready. Add all the missing code. "
+        f"Some imports might be missing. Make sure to add them. "
+        f"Some libraries might be missing. Make sure to install them in the requirements.txt and Dockerfile. "
+        "First write down an extensive list of obvious and non-obvious thoughts about what parts could need an adjustment and why. "
+        "Think about if all the changes are required and finally decide for the changes you want to make. "
+        f"Output all the files even the ones that did not change. "
         "Here are the files: \n\n"
         + all_executor_files_string
     )
@@ -119,60 +126,61 @@ def main(
 ):
     recreate_folder(EXECUTOR_FOLDER_v1)
     recreate_folder(EXECUTOR_FOLDER_v2)
+    recreate_folder('flow')
 
     all_executor_files_string = build_prototype_implementation(executor_description, executor_name, input_doc_field, input_modality,
                                                 output_doc_field, output_modality, test_in, test_out)
     extract_and_write(all_executor_files_string, EXECUTOR_FOLDER_v1)
-
+    write_config_yml(executor_name, EXECUTOR_FOLDER_v1)
     file_name_to_content_v1 = get_all_executor_files_with_content(EXECUTOR_FOLDER_v1)
-
     all_executor_files_string_no_instructions = files_to_string(file_name_to_content_v1)
 
     all_executor_files_string_improved = build_production_ready_implementation(all_executor_files_string_no_instructions)
-
     extract_and_write(all_executor_files_string_improved, EXECUTOR_FOLDER_v2)
+    write_config_yml(executor_name, EXECUTOR_FOLDER_v2)
 
-    write_config_yml(executor_name)
+    jina_cloud.push_executor(EXECUTOR_FOLDER_v2)
 
-    jina_cloud.push_executor()
+    host = jina_cloud.deploy_flow(executor_name, do_validation, 'flow')
 
-    host = jina_cloud.deploy_flow(executor_name, do_validation)
+    update_client_line_in_file(os.path.join(EXECUTOR_FOLDER_v1, CLIENT_FILE_NAME), host)
+    update_client_line_in_file(os.path.join(EXECUTOR_FOLDER_v1, STREAMLIT_FILE_NAME), host)
+    update_client_line_in_file(os.path.join(EXECUTOR_FOLDER_v2, CLIENT_FILE_NAME), host)
+    update_client_line_in_file(os.path.join(EXECUTOR_FOLDER_v2, STREAMLIT_FILE_NAME), host)
 
-    update_client_line_in_file(f'executor/{CLIENT_FILE_NAME}', host)
-    update_client_line_in_file(f'executor/{STREAMLIT_FILE_NAME}', host)
     if do_validation:
-        pass
+        importlib.import_module("executor_v1.client")
 
     return get_all_executor_files_with_content(EXECUTOR_FOLDER_v2)
 
 
 if __name__ == '__main__':
-    ######### Level 2 task #########
-    main(
-        executor_name='My3DTo2DExecutor',
-        executor_description="The executor takes 3D objects in obj format as input and outputs a 2D image projection of that object",
-        input_modality='3d',
-        input_doc_field='blob',
-        output_modality='image',
-        output_doc_field='blob',
-        test_in='https://raw.githubusercontent.com/makehumancommunity/communityassets-wip/master/clothes/leotard_fs/leotard_fs.obj',
-        test_out='the output should be exactly one image in png format',
-        do_validation=False
-    )
-
-    ######### Level 1 task #########
+    # ######### Level 2 task #########
     # main(
-    #     executor_name='MyCoolOcrExecutor',
-    #     executor_description="OCR detector",
-    #     input_modality='image',
-    #     input_doc_field='uri',
-    #     output_modality='text',
-    #     output_doc_field='text',
-    #     test_in='https://miro.medium.com/v2/resize:fit:1024/0*4ty0Adbdg4dsVBo3.png',
-    #     test_out='> Hello, world!_',
+    #     executor_name='My3DTo2DExecutor',
+    #     executor_description="The executor takes 3D objects in obj format as input and outputs a 2D image projection of that object",
+    #     input_modality='3d',
+    #     input_doc_field='blob',
+    #     output_modality='image',
+    #     output_doc_field='blob',
+    #     test_in='https://raw.githubusercontent.com/makehumancommunity/communityassets-wip/master/clothes/leotard_fs/leotard_fs.obj',
+    #     test_out='the output should be exactly one image in png format',
     #     do_validation=False
     # )
 
+    ######## Level 1 task #########
+    main(
+        executor_name='MyCoolOcrExecutor',
+        executor_description="OCR detector",
+        input_modality='image',
+        input_doc_field='uri',
+        output_modality='text',
+        output_doc_field='text',
+        test_in='https://miro.medium.com/v2/resize:fit:1024/0*4ty0Adbdg4dsVBo3.png',
+        test_out='> Hello, world!_',
+        do_validation=False
+    )
+
     # main(
     #     executor_name='MySentimentAnalyzer',
     #     executor_description="Sentiment analysis executor",

server.py

@@ -29,9 +29,9 @@ class CreateResponse(BaseModel):
     message: Optional[str]
 
 @app.post("/create", response_model=CreateResponse)
-async def create_endpoint(request: CreateRequest):
+def create_endpoint(request: CreateRequest):
 
-    result = await main(
+    result = main(
         executor_name=request.executor_name,
         executor_description=request.executor_description,
         input_modality=request.input_modality,
@@ -55,7 +55,7 @@ app.add_middleware(
 
 # Add a custom exception handler for RequestValidationError
 @app.exception_handler(RequestValidationError)
-async def validation_exception_handler(request: Request, exc: RequestValidationError):
+def validation_exception_handler(request: Request, exc: RequestValidationError):
     return JSONResponse(
         status_code=422,
         content={"detail": exc.errors()},

src/gpt.py

@@ -4,7 +4,8 @@ from time import sleep
 import openai
 from openai.error import RateLimitError, Timeout
 
-from src.utils.string import print_colored
+from src.utils.io import timeout_generator_wrapper
+from src.utils.string_tools import print_colored
 
 openai.api_key = os.environ['OPENAI_API_KEY']
 
@@ -13,7 +14,7 @@ def get_response(system_definition, user_query):
     print_colored('user_query', user_query, 'blue')
     for i in range(10):
         try:
-            response = openai.ChatCompletion.create(
+            response_generator = openai.ChatCompletion.create(
                 temperature=0,
                 max_tokens=5_000,
                 model="gpt-4",
@@ -32,15 +33,17 @@ def get_response(system_definition, user_query):
 
                 ]
             )
+            response_generator_with_timeout = timeout_generator_wrapper(response_generator, 5)
+
             complete_string = ''
-            for chunk in response:
+            for chunk in response_generator_with_timeout:
                 delta = chunk['choices'][0]['delta']
                 if 'content' in delta:
                     content = delta['content']
                     print_colored('' if complete_string else 'Agent response:', content, 'green', end='')
                     complete_string += content
             return complete_string
-        except (RateLimitError, Timeout) as e:
+        except (RateLimitError, Timeout, ConnectionError) as e:
             print(e)
             print('retrying')
             sleep(3)

src/jina_cloud.py

@@ -1,4 +1,3 @@
-import asyncio
 import os
 from multiprocessing.connection import Client
 
@@ -9,8 +8,8 @@ from jina import Flow
 from src.constants import FLOW_URL_PLACEHOLDER
 
 
-def push_executor():
-    cmd = 'jina hub push executor/. --verbose'
+def push_executor(dir_path):
+    cmd = f'jina hub push {dir_path}/. --verbose'
     os.system(cmd)
 
 def get_user_name():
@@ -25,7 +24,7 @@ def deploy_on_jcloud(flow_yaml):
 
 
 
-def deploy_flow(executor_name, do_validation):
+def deploy_flow(executor_name, do_validation, dest_folder):
     flow = f'''
 jtype: Flow
 with:
@@ -47,7 +46,8 @@ executors:
         instance: C4
         capacity: spot
 '''
-    full_flow_path = os.path.join('executor', 'flow.yml')
+    full_flow_path = os.path.join(dest_folder,
+                     'flow.yml')
     with open(full_flow_path, 'w') as f:
         f.write(flow)
 

src/prompt_system.py

@@ -6,14 +6,16 @@ executor_example = "Here is an example of how an executor can be defined. It alw
 # this executor takes ... as input and returns ... as output
 # it processes each document in the following way: ...
 from jina import Executor, requests, DocumentArray, Document
-class MyExecutor(Executor):
+class MyInfoExecutor(Executor):
     def __init__(self, **kwargs):
         super().__init__()
 
     @requests
     def foo(self, docs: DocumentArray, **kwargs) => DocumentArray:
         for d in docs:
-            d.text = 'hello world'"
+            d.load_uri_to_blob()
+            d.tags['my_info'] = {'byte_length': len(d.blob)}
+            d.blob = None
         return docs
 '''
 "An executor gets a DocumentArray as input and returns a DocumentArray as output. "

src/prompt_tasks.py

@@ -73,7 +73,7 @@ def docker_file_task():
         "It is important to make sure that all libs are installed that are required by the python packages. "
         "Usually libraries are installed with apt-get. "
         "Add the config.yml file to the Dockerfile. "
-        "The base image of the Dockerfile is FROM jinaai/jina:3.14.2-dev18-py310-standard. "
+        "The base image of the Dockerfile is FROM jinaai/jina:3.14.1-py39-standard. "
         'The entrypoint is ENTRYPOINT ["jina", "executor", "--uses", "config.yml"] '
         "The Dockerfile runs the test during the build process. ",
         DOCKER_FILE_TAG,

src/utils/io.py

@@ -1,8 +1,43 @@
 import os
 import shutil
-
+import concurrent.futures
+import concurrent.futures
+from typing import Generator
 
 def recreate_folder(folder_path):
     if os.path.exists(folder_path) and os.path.isdir(folder_path):
         shutil.rmtree(folder_path)
     os.makedirs(folder_path)
+
+
+class GenerationTimeoutError(Exception):
+    pass
+
+def timeout_generator_wrapper(generator, timeout):
+    def generator_func():
+        for item in generator:
+            yield item
+
+    def wrapper() -> Generator:
+        gen = generator_func()
+        while True:
+            try:
+                with concurrent.futures.ThreadPoolExecutor() as executor:
+                    future = executor.submit(next, gen)
+                    yield future.result(timeout=timeout)
+            except StopIteration:
+                break
+            except concurrent.futures.TimeoutError:
+                raise GenerationTimeoutError(f"Generation took longer than {timeout} seconds")
+
+    return wrapper()
+
+# def my_generator():
+#     for i in range(10):
+#         sleep(3)
+#         yield 1
+#
+#
+# my_generator_with_timeout = timeout_generator_wrapper(my_generator, 2.9)
+# for item in my_generator_with_timeout():
+#     print(item)