Merge pull request #4 from jina-ai/fix_json_only

fix: json only

README.md

@@ -104,9 +104,6 @@ graph TB
 
 ```
 
-
-
-
 1. GPT Deploy identifies several strategies to implement your task.
 2. It tests each strategy until it finds one that works.
 3. For each strategy, it creates the following files:
@@ -123,6 +120,14 @@ graph TB
 
 ## Examples
 
+### Animal Detector
+```bash
+
+gptdeploy create --description "Given an image, return the image with bounding boxes of all animals (https://pjreddie.com/media/files/yolov3.weights, https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3.cfg)" --test "https://images.unsplash.com/photo-1444212477490-ca407925329e contains animals"
+```
+
+<img src="res/animal_detector_example.png" alt="Animal Detector" width="600" />
+
 ### Meme Generator
 ```bash
 gptdeploy create --description "Generate a meme from an image and a caption" --test "Surprised Pikachu: https://media.wired.com/photos/5f87340d114b38fa1f8339f9/master/w_1600%2Cc_limit/Ideas_Surprised_Pikachu_HD.jpg, TOP:When you discovered GPTDeploy"
@@ -193,19 +198,12 @@ gptdeploy create --description "Visualize the Mandelbrot set with custom paramet
 
 [//]: # (## TO BE TESTED)
 
-[//]: # (### ASCII Art Generator)
-
-[//]: # (```bash)
-
-[//]: # (gptdeploy create --description "Convert image to ASCII art" --test "https://images.unsplash.com/photo-1602738328654-51ab2ae6c4ff")
-
-[//]: # (```)
 
 [//]: # (### Password Strength Checker)
 
 [//]: # (```bash)
 
-[//]: # (gptdeploy create --description "Given a password, return a score from 1 to 10 indicating the strength of the password" --test "Pa$$w0rd")
+[//]: # (gptdeploy create --description "Given a password, return a score from 1 to 10 indicating the strength of the password" --test "Pa$$w0rd => 1/5, !Akfdh%.ytRadf => 5/5")
 
 [//]: # (```)
 
@@ -389,11 +387,14 @@ gptdeploy create --description "Visualize the Mandelbrot set with custom paramet
 [//]: # (```)
 
 [//]: # ()
-[//]: # (### Bounding Box)
+
+
+
+[//]: # (### ASCII Art Generator)
 
 [//]: # (```bash)
 
-[//]: # (gptdeploy create --description "Given an image, return the bounding boxes of all animals" --test "...")
+[//]: # (gptdeploy create --description "Convert image to ASCII art" --test "https://images.unsplash.com/photo-1602738328654-51ab2ae6c4ff")
 
 [//]: # (```)
 
@@ -404,7 +405,7 @@ Use natural language interface to create, deploy and update your microservice in
 If you want to contribute to this project, feel free to open a PR or an issue.
 In the following, you can find a list of things that need to be done.
 
-Critical:
+next steps:
 - [ ] check if windows and linux support works
 - [ ] support gpt3.5-turbo
 - [ ] add video to README.md

src/executor_factory.py

@@ -196,7 +196,8 @@ print(response[0].text) # can also be blob in case of image/audio..., this shoul
                             + error_before) if error_before else '')
                         + '\n\nNow, I get the following error:\n'
                         + error + '\n'
-                        + 'Think quickly about possible reasons. '
+                        + 'Think quickly about possible reasons the error might caused by. '
+                          'Decide which files need to be changed. '
                           'Then output the files that need change. '
                           "Don't output files that don't need change. "
                           "If you output a file, then write the complete file. "

src/gpt.py

@@ -58,7 +58,7 @@ class GPTSession:
         print('Estimated costs on openai.com:')
         # print('money prompt:', f'${money_prompt}')
         # print('money generation:', f'${money_generation}')
-        print('total money so far:', f'${money_prompt + money_generation}')
+        print('total money spent so far:', f'${money_prompt + money_generation}')
         print('\n')
 
     def get_conversation(self):

src/prompt_system.py

@@ -8,6 +8,7 @@ Here is an example of how an executor can be defined. It always starts with a co
 ```python
 # this executor binary files as input and returns the length of each binary file as output
 from jina import Executor, requests, DocumentArray, Document
+import json
 class MyInfoExecutor(Executor):
     def __init__(self, **kwargs):
         super().__init__()
@@ -15,8 +16,9 @@ class MyInfoExecutor(Executor):
     @requests() # each executor must have exactly this decorator without parameters
     def foo(self, docs: DocumentArray, **kwargs) => DocumentArray:
         for d in docs:
-            d.load_uri_to_blob()
-            d.blob = None
+            content = json.loads(d.text)
+            ...
+            d.text = json.dumps(modified_content)
         return docs
 ```
 
@@ -29,64 +31,29 @@ A Document is a python class that represents a single document.
 Here is the protobuf definition of a Document:
 
 message DocumentProto {{
-  // A hexdigest that represents a unique document ID
-  string id = 1;
-
-  oneof content {{
-    // the raw binary content of this document, which often represents the original document when comes into jina
-    bytes blob = 2;
-
-    // the ndarray of the image/audio/video document
-    NdArrayProto tensor = 3;
-
-    // a text document
-    string text = 4;
-  }}
-
-  // a uri of the document is a remote url starts with http or https or data URI scheme
-  string uri = 5;
-
-  // list of the sub-documents of this document (recursive structure)
-  repeated DocumentProto chunks = 6;
-
-  // the matched documents on the same level (recursive structure)
-  repeated DocumentProto matches = 7;
-
-  // the embedding of this document
-  NdArrayProto embedding = 8;
+  // used to store json data the executor gets and returns
+  string text = 1;
 }}
 
-Here is an example of how a DocumentArray can be defined:
+Here are examples of how a DocumentArray can be defined:
 
 from jina import DocumentArray, Document
+import json
 
-d1 = Document(text='hello')
+d1 = Document(text=json.dumps({{'he_says': 'hello'}}))
 
 # 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...'
+base64_data = base64.b64encode(png_data).decode('utf-8')
+d2 = Document(text=json.dumps({{'image': base64_data}}))
 
-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',
-)
-d5 = Document()
-d5.tensor = np.ones((2,4))
-d5.uri = 'https://audio.com/audio.mp3'
-d6 = Document()
-d6.blob # like b'RIFF\\x00\\x00\\x00\\x00WAVEfmt \\x10\\x00...'
-docs = DocumentArray([
-   d1, d2, d3, d4
-])
-d7 = Document()
-d7.text = 'test string'
-d8 = Document()
-d8.text = json.dumps([{{"id": "1", "text": ["hello", 'test']}}, {{"id": "2", "text": "world"}}])
-# 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... 
+array = numpy.array([1, 2, 3])
+array_list = array.tolist()
+d3 = Document(text=json.dumps(array_list))
+d4 = Document()
+d4.text = '{{"uri": "https://.../logo.png"}}'
 '''
 
 

src/prompt_tasks.py

@@ -134,6 +134,5 @@ The executor must not access external apis except unless it is explicitly mentio
 The executor must not load data from the local file system unless it was created by the executor itself.
 The executor must not use a pre-trained model unless it is explicitly mentioned in the description.
 The executor must not train a model.
-The executor must not use Document.tags.
-The executor must only use Document.uri, Document.blob and Document.text.
+The executor must not use any attribute of Document accept Document.text.
 '''