feat: playground

micro_chain.py

@@ -6,7 +6,7 @@ from src import gpt, jina_cloud
 from src.constants import FILE_AND_TAG_PAIRS
 from src.jina_cloud import build_docker
 from src.prompt_tasks import general_guidelines, executor_file_task, chain_of_thought_creation, test_executor_file_task, \
-    chain_of_thought_optimization, requirements_file_task, docker_file_task
+    chain_of_thought_optimization, requirements_file_task, docker_file_task, not_allowed
 from src.utils.io import recreate_folder, persist_file
 from src.utils.string_tools import print_colored
 
@@ -15,17 +15,19 @@ def wrap_content_in_code_block(executor_content, file_name, tag):
     return f'**{file_name}**\n```{tag}\n{executor_content}\n```\n\n'
 
 
-def main(
+
+
+def create_executor(
         executor_description,
         input_modality,
         output_modality,
         test_scenario,
-        do_validation=True
+        executor_name
 ):
     input_doc_field = 'text' if input_modality == 'text' else 'blob'
     output_doc_field = 'text' if output_modality == 'text' else 'blob'
     # random integer at the end of the executor name to avoid name clashes
-    executor_name = f'MicroChainExecutor{random.randint(0, 1000_000)}'
+
     recreate_folder('executor')
     EXECUTOR_FOLDER_v1 = 'executor/v1'
     recreate_folder(EXECUTOR_FOLDER_v1)
@@ -40,9 +42,9 @@ def main(
     )
     conversation = gpt.Conversation()
     conversation.query(user_query)
-    executor_content_raw = conversation.query(chain_of_thought_optimization('python', 'executor.py'))
+    executor_content_raw = conversation.query(f"General rules: " + not_allowed() + chain_of_thought_optimization('python', 'executor.py'))
     executor_content = extract_content_from_result(executor_content_raw, 'executor.py')
-    persist_file(executor_content, 'executor.py')
+    persist_file(executor_content, EXECUTOR_FOLDER_v1 + '/executor.py')
 
     print_colored('', '############# Test Executor #############', 'red')
     user_query = (
@@ -53,11 +55,12 @@ def main(
     conversation = gpt.Conversation()
     conversation.query(user_query)
     test_executor_content_raw = conversation.query(
+        f"General rules: " + not_allowed() +
         chain_of_thought_optimization('python', 'test_executor.py')
         + "Don't add any additional tests. "
     )
     test_executor_content = extract_content_from_result(test_executor_content_raw, 'test_executor.py')
-    persist_file(test_executor_content, 'test_executor.py')
+    persist_file(test_executor_content, EXECUTOR_FOLDER_v1 + '/test_executor.py')
 
     print_colored('', '############# Requirements #############', 'red')
     user_query = (
@@ -71,7 +74,7 @@ def main(
     requirements_content_raw = conversation.query(chain_of_thought_optimization('', 'requirements.txt') + "Keep the same version of jina ")
 
     requirements_content = extract_content_from_result(requirements_content_raw, 'requirements.txt')
-    persist_file(requirements_content, 'requirements.txt')
+    persist_file(requirements_content, EXECUTOR_FOLDER_v1 + '/requirements.txt')
 
     print_colored('', '############# Dockerfile #############', 'red')
     user_query = (
@@ -83,24 +86,50 @@ def main(
     )
     conversation = gpt.Conversation()
     conversation.query(user_query)
-    dockerfile_content_raw = conversation.query(chain_of_thought_optimization('dockerfile', 'Dockerfile'))
+    dockerfile_content_raw = conversation.query(f"General rules: " + not_allowed() + chain_of_thought_optimization('dockerfile', 'Dockerfile'))
     dockerfile_content = extract_content_from_result(dockerfile_content_raw, 'Dockerfile')
-    persist_file(dockerfile_content, 'Dockerfile')
+    persist_file(dockerfile_content, EXECUTOR_FOLDER_v1 + '/Dockerfile')
 
     write_config_yml(executor_name, EXECUTOR_FOLDER_v1)
 
-    for i in range(1, 20):
+def create_playground(executor_name, executor_path, host):
+    print_colored('', '############# Playground #############', 'red')
+
+    file_name_to_content = get_all_executor_files_with_content(executor_path)
+    user_query = (
+            general_guidelines()
+            + wrap_content_in_code_block(file_name_to_content['executor.py'], 'executor.py', 'python')
+            + wrap_content_in_code_block(file_name_to_content['test_executor.py'], 'test_executor.py', 'python')
+            + f'''
+Create a playground for the executor {executor_name} using streamlit. 
+The executor is hosted on {host}. 
+This is an example how you can connect to the executor assuming the document (d) is already defined:
+from jina import Client, Document, DocumentArray
+client = Client(host='{host}')
+response = client.post('/process', inputs=DocumentArray([d]))
+print(response[0].text) # can also be blob in case of image/audio..., this should be visualized in the streamlit app
+'''
+    )
     conversation = gpt.Conversation()
+    conversation.query(user_query)
+    playground_content_raw = conversation.query(f"General rules: " + not_allowed() + chain_of_thought_optimization('python', 'playground.py'))
+    playground_content = extract_content_from_result(playground_content_raw, 'playground.py')
+    persist_file(playground_content, f'{executor_path}/playground.py')
+
+def debug_executor():
+    for i in range(1, 20):
+
         error = build_docker(f'executor/v{i}')
         if error:
             recreate_folder(f'executor/v{i + 1}')
             file_name_to_content = get_all_executor_files_with_content(f'executor/v{i}')
             all_files_string = files_to_string(file_name_to_content)
             user_query = (
-                'Here are all the files I use:\n'
+                f"General rules: " + not_allowed()
+                + 'Here are all the files I use:\n'
                 + all_files_string
                 + 'I got the following error:\n'
-                + error
+                + error + '\n'
                 + 'Think quickly about possible reasons. '
                   'Then output the files that need change. '
                   "Don't output files that don't need change. "
@@ -111,6 +140,7 @@ def main(
                    f"...code...\n"
                    f"```\n\n"
             )
+            conversation = gpt.Conversation()
             returned_files_raw = conversation.query(user_query)
             for file_name, tag in FILE_AND_TAG_PAIRS:
                 updated_file = extract_content_from_result(returned_files_raw, file_name)
@@ -121,33 +151,64 @@ def main(
                 persist_file(content, f'executor/v{i + 1}/{file_name}')
         else:
             break
+    return f'executor/v{i}'
 
 
-    jina_cloud.push_executor('executor')
+def main(
-    host = jina_cloud.deploy_flow(executor_name, do_validation, 'flow')
+        executor_description,
-
+        input_modality,
-    # create playgorund and client.py
+        output_modality,
-
+        test_scenario,
+):
+    executor_name = f'MicroChainExecutor{random.randint(0, 1000_000)}'
+    create_executor(executor_description, input_modality, output_modality, test_scenario, executor_name)
+    executor_path = debug_executor()
+    print('Executor can be built locally, now we will push it to the cloud.')
+    jina_cloud.push_executor(executor_path)
+    print('Deploy a jina flow')
+    host = jina_cloud.deploy_flow(executor_name, 'flow')
+    print(f'Flow is deployed create the playground for {host}')
+    executor_name = 'MicroChainExecutor48442'
+    executor_path = 'executor/v2'
+    host = 'grpcs://mybelovedocrflow-24a412bc63.wolf.jina.ai'
+    create_playground(executor_name, executor_path, host)
 
 if __name__ == '__main__':
     # ######## Level 1 task #########
+    main(
+        executor_description="The executor takes a pdf file as input, parses it and returns the text.",
+        input_modality='pdf',
+        output_modality='text',
+        test_scenario='Takes https://www2.deloitte.com/content/dam/Deloitte/de/Documents/about-deloitte/Deloitte-Unternehmensgeschichte.pdf and returns a string that is at least 100 characters long',
+    )
+    # money prompt: $0.56
+    # money generation: $0.22
+    # total money: $0.78
+
+
+    # ######## Level 2 task #########
     # main(
     #     executor_description="OCR detector",
     #     input_modality='image',
-    #     # input_doc_field='blob',
     #     output_modality='text',
-    #     # output_doc_field='text',
     #     test_scenario='Takes https://miro.medium.com/v2/resize:fit:1024/0*4ty0Adbdg4dsVBo3.png as input and returns a string that contains "Hello, world"',
-    #     do_validation=False
     # )
 
-    ######### Level 2 task #########
+
-    main(
+    # ######## Level 3 task #########
-        executor_description="The executor takes 3D objects in obj format as input "
+    # main(
-                             "and outputs a 2D image projection of that object where the full object is shown. ",
+    #     executor_description="The executor takes an mp3 file as input and returns bpm and pitch in the tags.",
-        input_modality='3d',
+    #     input_modality='audio',
-        output_modality='image',
+    #     output_modality='tags',
-        test_scenario='Test that 3d object from https://raw.githubusercontent.com/polygonjs/polygonjs-assets/master/models/wolf.obj '
+    #     test_scenario='Takes https://miro.medium.com/v2/resize:fit:1024/0*4ty0Adbdg4dsVBo3.png as input and returns a string that contains "Hello, world"',
-                      'is put in and out comes a 2d rendering of it',
+    # )
-        do_validation=False
+
-    )
+    ######### Level 4 task #########
+    # main(
+    #     executor_description="The executor takes 3D objects in obj format as input "
+    #                          "and outputs a 2D image projection of that object where the full object is shown. ",
+    #     input_modality='3d',
+    #     output_modality='image',
+    #     test_scenario='Test that 3d object from https://raw.githubusercontent.com/polygonjs/polygonjs-assets/master/models/wolf.obj '
+    #                   'is put in and out comes a 2d rendering of it',
+    # )

requirements.txt

@@ -1,5 +1,6 @@
-jina[perf]==3.14.2.dev18
+jina==3.14.1
-openai
+pyrender~=0.1.45
-ptest
+trimesh~=3.10.0
-jcloud
+numpy~=1.22.3
-uvicorn
+Pillow~=9.0.1
+requests~=2.27.1

src/gpt.py

@@ -11,6 +11,8 @@ from src.utils.string_tools import print_colored
 
 openai.api_key = os.environ['OPENAI_API_KEY']
 
+total_chars_prompt = 0
+total_chars_generation = 0
 
 class Conversation:
     def __init__(self):
@@ -26,6 +28,7 @@ class Conversation:
 
 
 def get_response(prompt_list: List[Tuple[str, str]]):
+    global total_chars_prompt, total_chars_generation
     for i in range(10):
         try:
             response_generator = openai.ChatCompletion.create(
@@ -42,7 +45,7 @@ def get_response(prompt_list: List[Tuple[str, str]]):
                 ]
             )
             response_generator_with_timeout = timeout_generator_wrapper(response_generator, 5)
-
+            total_chars_prompt += sum(len(prompt[1]) for prompt in prompt_list)
             complete_string = ''
             for chunk in response_generator_with_timeout:
                 delta = chunk['choices'][0]['delta']
@@ -50,6 +53,13 @@ def get_response(prompt_list: List[Tuple[str, str]]):
                     content = delta['content']
                     print_colored('' if complete_string else 'assistent', content, 'green', end='')
                     complete_string += content
+                    total_chars_generation += len(content)
+            print('\n')
+            money_prompt = round(total_chars_prompt / 3.4 * 0.03 / 1000, 2)
+            money_generation = round(total_chars_generation / 3.4 * 0.06 / 1000, 2)
+            print('money prompt:', f'${money_prompt}')
+            print('money generation:', f'${money_generation}')
+            print('total money:', f'${money_prompt + money_generation}')
             print('\n')
             return complete_string
         except (RateLimitError, Timeout, ConnectionError, GenerationTimeoutError) as e:

src/jina_cloud.py

@@ -1,5 +1,4 @@
 import os
-from multiprocessing.connection import Client
 import subprocess
 import re
 
@@ -7,7 +6,6 @@ import hubble
 from jcloud.flow import CloudFlow
 from jina import Flow
 
-from src.constants import FLOW_URL_PLACEHOLDER
 
 
 def push_executor(dir_path):
@@ -26,7 +24,7 @@ def deploy_on_jcloud(flow_yaml):
 
 
 
-def deploy_flow(executor_name, do_validation, dest_folder):
+def deploy_flow(executor_name, dest_folder):
     flow = f'''
 jtype: Flow
 with:
@@ -53,7 +51,6 @@ executors:
     with open(full_flow_path, 'w') as f:
         f.write(flow)
 
-    if do_validation:
     print('try local execution')
     flow = Flow.load_config(full_flow_path)
     with flow:
@@ -86,7 +83,7 @@ def build_docker(path):
         lines = error_message.split('\n')
         relevant_lines = []
 
-        pattern = re.compile(r"^#\d+ \[\d+/\d+\]")  # Pattern to match lines like "#11 [7/8]"
+        pattern = re.compile(r"^#\d+ \[[ \d]+/[ \d]+\]")  # Pattern to match lines like "#11 [7/8]"
         last_matching_line_index = None
 
         for index, line in enumerate(lines):

src/prompt_system.py

@@ -66,7 +66,13 @@ Here is an example of how a DocumentArray can be defined:
 from jina import DocumentArray, Document
 
 d1 = Document(text='hello')
-d2 = Document(blob=b'\\x89PNG\\r\\n\\x1a\\n\\x00\\x00\\x00\\rIHDR\\x00\\x00\\x03L\\x00\\x00\\x01\\x18\\x08\\x06\\x00\\x00\\x00o...')
+
+# you can load binary data into a document
+url = 'https://...'
+response = requests.get(url)
+obj_data = response.content
+d2 = Document(blob=obj_data) # blob is bytes like b'\\x89PNG\\r\\n\\x1a\\n\
+
 d3 = Document(tensor=numpy.array([1, 2, 3]), chunks=[Document(uri=/local/path/to/file)]
 d4 = Document(
    uri='https://docs.docarray.org/img/logo.png',
@@ -76,13 +82,13 @@ d5 = Document()
 d5.tensor = np.ones((2,4))
 d5.uri = 'https://audio.com/audio.mp3'
 d6 = Document()
-d6.blob = b'RIFF\\x00\\x00\\x00\\x00WAVEfmt \\x10\\x00...'
+d6.blob # like b'RIFF\\x00\\x00\\x00\\x00WAVEfmt \\x10\\x00...'
 docs = DocumentArray([
    d1, d2, d3, d4
 ])
 # the document has a helper function load_uri_to_blob:
 # For instance, d4.load_uri_to_blob() downloads the file from d4.uri and stores it in d4.blob. 
-# If d4.uri was something like 'https://website.web/img.jpg', then d4.blob would be something like  b'\xff\xd8\xff\xe0\x00\x10JFIF\x00\x01\x01... 
+# If d4.uri was something like 'https://website.web/img.jpg', then d4.blob would be something like  b'\\xff\\xd8\\xff\\xe0\\x00\\x10JFIF\\x00\\x01\\x01... 
 '''
 
 
@@ -94,10 +100,10 @@ Here is an example of a client file:
 ```python
 from jina import Client, Document, DocumentArray
 client = Client(host='{FLOW_URL_PLACEHOLDER}')
-d = Document(uri='data/img.png')
+d = Document(uri='...')
 d.load_uri_to_blob()
 response = client.post('/process', inputs=DocumentArray([d]))
-response[0].summary()
+print(response[0].text)
 ```
 '''
 

src/prompt_tasks.py

@@ -35,27 +35,12 @@ It matches the following description: '{executor_description}'.
 It gets a DocumentArray as input where each document has the input modality '{input_modality}' and can be accessed via document.{input_doc_field}.
 It returns a DocumentArray as output where each document has the output modality '{output_modality}' that is stored in document.{output_doc_field}.
 Have in mind that d.uri is never a path to a local file. It is always a url.
-The executor is not allowed to use the GPU.
+''' + not_allowed(),
-The executor is not allowed to access a database.
-The executor is not allowed to access a display.
-The executor is not allowed to access external apis. 
-''',
                  EXECUTOR_FILE_TAG,
                  EXECUTOR_FILE_NAME
                  )
 
 
-def requirements_file_task():
-    return _task(
-        "Write the content of the requirements.txt file. "
-        "Make sure to include pytest. "
-        "Make sure that jina==3.14.1. "
-        "All versions are fixed using ~=, ==, <, >, <=, >=. The package versions should not have conflicts. ",
-        REQUIREMENTS_FILE_TAG,
-        REQUIREMENTS_FILE_NAME
-    )
-
-
 def test_executor_file_task(executor_name, test_scenario):
     return _task(
         "Write a small unit test for the executor. "
@@ -65,11 +50,23 @@ def test_executor_file_task(executor_name, test_scenario):
             if test_scenario else ""
         )
         + "Use the following import to import the executor: "
-          f"from executor import {executor_name} ",
+          f"from executor import {executor_name} "
+        + not_allowed()
+        + "The test is not allowed to open local files. ",
         TEST_EXECUTOR_FILE_TAG,
         TEST_EXECUTOR_FILE_NAME
     )
 
+def requirements_file_task():
+    return _task(
+        "Write the content of the requirements.txt file. "
+        "Make sure to include pytest. "
+        "Make sure that jina==3.14.1. "
+        "All versions are fixed using ~=, ==, <, >, <=, >=. The package versions should not have conflicts. ",
+        REQUIREMENTS_FILE_TAG,
+        REQUIREMENTS_FILE_NAME
+    )
+
 
 def docker_file_task():
     return _task(
@@ -80,8 +77,9 @@ def docker_file_task():
         "Be aware that the machine the docker container is running on does not have a GPU - only CPU. "
         "Add the config.yml file to the Dockerfile. "
         "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 entrypoint is ENTRYPOINT ["jina", "executor", "--uses", "config.yml"]. '
-        "The Dockerfile runs the test during the build process. ",
+        'Make sure the all files are in the /workdir. '
+        "The Dockerfile runs the test during the build process. " + not_allowed(),
         DOCKER_FILE_TAG,
         DOCKER_FILE_NAME
     )
@@ -106,8 +104,9 @@ def streamlit_file_task():
 def chain_of_thought_creation():
     return (
         "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. "
+        "For example, there are different libraries you could use and not all of them obay the rules: "
-        "Discuss the pros and cons for all of these approaches and then decide for one of the approaches. "
+        + not_allowed()
+        + "Discuss the pros and cons for all of these approaches and then decide for one of the approaches. "
         "Then write as I told you. "
     )
 
@@ -124,3 +123,11 @@ def chain_of_thought_optimization(tag_name, file_name):
         tag_name,
         file_name
     )
+
+def not_allowed():
+    return '''
+The executor is not allowed to use the GPU.
+The executor is not allowed to access a database.
+The executor is not allowed to access a display.
+The executor is not allowed to access external apis. 
+'''