🛺 fix: auto refinement parsing

dev_gpt/options/generate/chains/auto_refine_description.py

@@ -1,7 +1,7 @@
 import json
 
 from dev_gpt.apis.gpt import ask_gpt
-from dev_gpt.options.generate.parser import identity_parser
+from dev_gpt.options.generate.parser import identity_parser, optional_tripple_back_tick_parser
 from dev_gpt.options.generate.prompt_factory import context_to_string
 from dev_gpt.options.generate.tools.tools import get_available_tools
 
@@ -14,12 +14,12 @@ def auto_refine_description(context):
     )
     context['request_schema'] = ask_gpt(
         generate_request_schema_prompt,
-        identity_parser,
+        optional_tripple_back_tick_parser,
         context_string=context_to_string(context)
     )
     context['response_schema'] = ask_gpt(
         generate_output_schema_prompt,
-        identity_parser,
+        optional_tripple_back_tick_parser,
         context_string=context_to_string(context)
     )
     context['microservice_description'] = ask_gpt(
@@ -36,12 +36,13 @@ def auto_refine_description(context):
 better_description_prompt = f'''{{context_string}}
 Update the description of the Microservice to make it more precise without adding or removing information.
 Note: the output must be a list of tasks the Microservice has to perform.
-Note: you can uses two tools if necessary:
+Note: you can uses the following tools if necessary:
 {get_available_tools()}
-Example for the description: "return an image representing the current weather for a given location."
+Example for the description: "return an image representing the current weather for a given location." \
+when the tools gpt-3.5-turbo and google-search are available:
 1. get the current weather information from the https://openweathermap.org/ API
-2. generate a Google search query to find the image matching the weather information and the location by using gpt-3.5-turbo
-3. find the image by using the Google search API
+2. generate a Google search query to find the image matching the weather information and the location by using gpt-3.5-turbo (a)
+3. find the image by using the Google search API (b)
 4. return the image as a base64 encoded string'''
 
 generate_request_schema_prompt = '''{context_string}

dev_gpt/options/generate/chains/question_answering.py

@@ -15,20 +15,18 @@ def is_question_false(question):
 
 def answer_yes_no_question(text, question):
     pros_and_cons = ask_gpt(
-        pros_and_cons_prompt.format(
-            question=question,
-            text=text,
-        ),
-        identity_parser,
+        pros_and_cons_prompt,
+        question=question,
+        text=text,
     )
 
     return ask_gpt(
-        question_prompt.format(
-            text=text,
-            question=question,
-            pros_and_cons=pros_and_cons,
-        ),
-        boolean_parser)
+        question_prompt,
+        boolean_parser,
+        text=text,
+        question=question,
+        pros_and_cons=pros_and_cons,
+    )
 
 pros_and_cons_prompt = '''\
 # Context
@@ -42,5 +40,7 @@ question_prompt = '''\
 {text}
 # Question
 {question}
+# Pros and Cons
+{pros_and_cons}
 Note: You must answer "yes" or "no".
 '''

dev_gpt/options/generate/parser.py

@@ -5,6 +5,12 @@ import re
 def identity_parser(x):
     return x
 
+def optional_tripple_back_tick_parser(x):
+    if '```' in x:
+        pattern = r'```(.+)```'
+        x = re.findall(pattern, x, re.DOTALL)[-1]
+    return x.strip()
+
 def boolean_parser(x):
     return 'yes' in x.lower()
 

dev_gpt/options/generate/prompt_factory.py

@@ -8,7 +8,6 @@ def context_to_string(context):
     for k, v in context.items():
         if isinstance(v, dict):
             v = json.dumps(v, indent=4)
-        v = make_prompt_friendly(v)
         context_strings.append(f'''\
 {k}:
 ```

test/integration/test_generator.py

@@ -174,11 +174,10 @@ def test_generation_level_5_company_logos(microservice_dir, mock_input_sequence)
     os.environ['VERBOSE'] = 'true'
     generator = Generator(
         f'''\
-Given a list of email addresses, get all company names from them.
+Given a list of email addresses, get all unique company names from them.
 For all companies, get the company logo.
 All logos need to be arranged on a square.
-The square is returned as png.
-''',
+The square is returned as png.''',
         str(microservice_dir),
         'gpt-3.5-turbo',
         # self_healing=False,