Support for list in list of custom objects

Author: alexeichhorn

tests/test_parser.py

@@ -288,12 +288,12 @@ def test_parse_subtype_output(self):
 STRING 1: s1
 STRING 2: s2
 STRING 3: s3
-ITEM SUBTITLE 1: subtitle one
-ITEM DESCRIPTION 1: description one
-ITEM ABSTRACT 1: Just an abstract...
-ITEM SUBTITLE 2: subtitle two
-ITEM DESCRIPTION 2: Description TWO
-ITEM ABSTRACT 2: More abstract...
+ITEM 1 SUBTITLE: subtitle one
+ITEM 1 DESCRIPTION: description one
+ITEM 1 ABSTRACT: Just an abstract...
+ITEM 2 SUBTITLE: subtitle two
+ITEM 2 DESCRIPTION: Description TWO
+ITEM 2 ABSTRACT: More abstract...
 
 SUBITEM TITLE: A subitem title (!!)
 
@@ -343,7 +343,8 @@ class InnerElement(BaseLLMArrayElement):
             description: str | None = LLMArrayElementOutput(lambda pos: f"Put the {pos.ordinal} item description here")
             abstract: str = LLMArrayElementOutput(lambda _: "...", multiline=True)
             inner_item: InnerItem
-            # inner_elements: list[InnerElement] = LLMArrayOutput(2, instruction=lambda _: "...")
+            inner_elements: list[InnerElement] = LLMArrayOutput(2, instruction=lambda _: "...")
+            multiline_text: list[str] = LLMArrayOutput((0, None), lambda _: "...", multiline=True)
 
         class DirectItem(BaseLLMResponse):
             class InnerDirectElement(BaseLLMArrayElement):
@@ -362,25 +363,38 @@ def test_parse_ultra_subtype_output(self):
 
 SUBITEM TITLE: subtitle
 disregarded!
-SUBITEM X SUBTITLE 1: sub1
+SUBITEM X 1 SUBTITLE: sub1
 ? also disregarded ?
-SUBITEM X SUBTITLE 2: sub2
-
-ITEM SUBTITLE 1: First item subtitle
-ITEM DESCRIPTION 1: first descr.
-ITEM ABSTRACT 1: Jsut some random abstract
-ITEM INNER ITEM TITLE 1: INNER TITLE 1
-ITEM INNER ITEM DESCRIPTION 1: DESCription 1
-ITEM INNER ELEMENT VALUE 1: 1.0
-ITEM INNER ELEMENT IS ACCURATE 1: yes
-ITEM INNER ELEMENT VALUE 2: 3.14
-ITEM INNER ELEMENT IS ACCURATE 2: False
-
-ITEM SUBTITLE 2: Second item subtitle
-ITEM ABSTRACT 2: Another abstract but this time
+SUBITEM X 2 SUBTITLE: sub2
+
+ITEM 1 SUBTITLE: First item subtitle
+ITEM 1 DESCRIPTION: first descr.
+ITEM 1 ABSTRACT: Jsut some random abstract
+ITEM 1 INNER ITEM TITLE: INNER TITLE 1
+ITEM 1 INNER ITEM DESCRIPTION: DESCription 1
+ITEM 1 INNER ELEMENT 1 VALUE: 1.0
+ITEM 1 INNER ELEMENT 1 IS ACCURATE: yes
+ITEM 1 INNER ELEMENT 2 VALUE: 3.14
+ITEM 1 INNER ELEMENT 2 IS ACCURATE: False
+ITEM 1 MULTILINE TEXT 1: line 1
+line 2
+line 3
+ITEM 1 MULTILINE TEXT 2: line 4
+line 5
+line 6
+ITEM 1 MULTILINE TEXT 3: line 7
+line 8
+line 9
+
+ITEM 2 SUBTITLE: Second item subtitle
+ITEM 2 ABSTRACT: Another abstract but this time
 with multiple lines
-ITEM INNER ITEM TITLE 2: tt2
-ITEM INNER ITEM DESCRIPTION 2: dd2
+ITEM 2 INNER ITEM TITLE: tt2
+ITEM 2 INNER ITEM DESCRIPTION: dd2
+ITEM 2 INNER ELEMENT 1 VALUE: 2.0
+ITEM 2 INNER ELEMENT 1 IS ACCURATE: no
+ITEM 2 INNER ELEMENT 2 VALUE: 8.958
+ITEM 2 INNER ELEMENT 2 IS ACCURATE: True
 """
         parsed_output = self.UltraSubtypeTestOutput.parse_response(completion_output)
         assert parsed_output.title == "Main head title"
@@ -395,11 +409,22 @@ def test_parse_ultra_subtype_output(self):
         assert parsed_output.items[0].abstract == "Jsut some random abstract"
         assert parsed_output.items[0].inner_item.title == "INNER TITLE 1"
         assert parsed_output.items[0].inner_item.description == "DESCription 1"
+        assert len(parsed_output.items[0].inner_elements) == 2
+        assert parsed_output.items[0].inner_elements[0].value == 1.0
+        assert parsed_output.items[0].inner_elements[0].is_accurate == True
+        assert parsed_output.items[0].inner_elements[1].value == 3.14
+        assert parsed_output.items[0].inner_elements[1].is_accurate == False
+        assert parsed_output.items[0].multiline_text == ["line 1\nline 2\nline 3", "line 4\nline 5\nline 6", "line 7\nline 8\nline 9"]
 
         assert parsed_output.items[1].subtitle == "Second item subtitle"
         assert parsed_output.items[1].description is None
         assert parsed_output.items[1].abstract == "Another abstract but this time\nwith multiple lines"
         assert parsed_output.items[1].inner_item.title == "tt2"
         assert parsed_output.items[1].inner_item.description == "dd2"
+        assert len(parsed_output.items[1].inner_elements) == 2
+        assert parsed_output.items[1].inner_elements[0].value == 2.0
+        assert parsed_output.items[1].inner_elements[0].is_accurate == False
+        assert parsed_output.items[1].inner_elements[1].value == 8.958
+        assert parsed_output.items[1].inner_elements[1].is_accurate == True
 
     # endregion

tests/test_prompt.py

@@ -129,12 +129,12 @@ def test_subtype_output_fields(self):
 STRING 2: <Put the second string here>
 ...
 
-ITEM SUBTITLE 1: <Put the first subtitle here>
-ITEM DESCRIPTION 1: <Put the first item description here>
-ITEM ABSTRACT 1: <Some instruction>
-ITEM SUBTITLE 2: <Put the second subtitle here>
-ITEM DESCRIPTION 2: <Put the second item description here>
-ITEM ABSTRACT 2: <Some instruction>
+ITEM 1 SUBTITLE: <Put the first subtitle here>
+ITEM 1 DESCRIPTION: <Put the first item description here>
+ITEM 1 ABSTRACT: <Some instruction>
+ITEM 2 SUBTITLE: <Put the second subtitle here>
+ITEM 2 DESCRIPTION: <Put the second item description here>
+ITEM 2 ABSTRACT: <Some instruction>
 ...
 
 SUBITEM TITLE: <Put the title here>
@@ -180,28 +180,28 @@ def test_ultra_subtype_output_fields(self):
 TITLE: <Put the title here>
 
 SUBITEM TITLE: <Put the title here>
-SUBITEM X SUBTITLE 1: <Put the first subtitle here>
-SUBITEM X SUBTITLE 2: <Put the second subtitle here>
+SUBITEM X 1 SUBTITLE: <Put the first subtitle here>
+SUBITEM X 2 SUBTITLE: <Put the second subtitle here>
 ...
 
-ITEM SUBTITLE 1: <Put the first subtitle here>
-ITEM DESCRIPTION 1: <Put the first description here>
-ITEM ABSTRACT 1: <Put the first abstract here>
+ITEM 1 SUBTITLE: <Put the first subtitle here>
+ITEM 1 DESCRIPTION: <Put the first description here>
+ITEM 1 ABSTRACT: <Put the first abstract here>
 
-ITEM INNER ITEM TITLE 1: <Put the title here>
-ITEM INNER ITEM DESCRIPTION 1: <Put the description here>
+ITEM 1 INNER ITEM TITLE: <Put the title here>
+ITEM 1 INNER ITEM DESCRIPTION: <Put the description here>
 
-ITEM INNER ELEMENT VALUE 1: <Put the first value here>
-ITEM INNER ELEMENT VALUE 2: <Put the second value here>
-ITEM SUBTITLE 2: <Put the second subtitle here>
-ITEM DESCRIPTION 2: <Put the second description here>
-ITEM ABSTRACT 2: <Put the second abstract here>
+ITEM 1 INNER ELEMENT 1 VALUE: <Put the first value here>
+ITEM 1 INNER ELEMENT 2 VALUE: <Put the second value here>
+ITEM 2 SUBTITLE: <Put the second subtitle here>
+ITEM 2 DESCRIPTION: <Put the second description here>
+ITEM 2 ABSTRACT: <Put the second abstract here>
 
-ITEM INNER ITEM TITLE 2: <Put the title here>
-ITEM INNER ITEM DESCRIPTION 2: <Put the description here>
+ITEM 2 INNER ITEM TITLE: <Put the title here>
+ITEM 2 INNER ITEM DESCRIPTION: <Put the description here>
 
-ITEM INNER ELEMENT VALUE 1: <Put the first value here>
-ITEM INNER ELEMENT VALUE 2: <Put the second value here>
+ITEM 2 INNER ELEMENT 1 VALUE: <Put the first value here>
+ITEM 2 INNER ELEMENT 2 VALUE: <Put the second value here>
 ...
 \"""
 """.strip()

typegpt/parser.py

@@ -48,7 +48,7 @@ def _regex_for_field(self, field: LLMFieldInfo) -> str:
                 count_regex = "\\d{1,3}"
 
             if field_type := if_array_element_list_type(field.type_):
-                return rf"(?:^|\n){field.name} (?P<subfield_name>((?!:|\n)[\S ])+) (?P<i>{count_regex}): ?(?P<content>({exclusion_cases_regex}[\s\S])+)"
+                return rf"(?:^|\n){field.name} (?P<i>{count_regex}) (?P<subfield_name>((?!:|\n)[\S ])+): ?(?P<content>({exclusion_cases_regex}[\s\S])+)"
             else:
                 return rf"(?:^|\n){field.name} {count_regex}: ?(?P<content>({exclusion_cases_regex}[\s\S])+)"
 

typegpt/prompt_builder.py

@@ -15,8 +15,12 @@ def _generate_array(self, field: LLMFieldInfo, info: LLMArrayOutputInfo) -> str:
 
         if element_type := if_array_element_list_type(field.type_):
             subfields = list(element_type.__fields__.values())
-            subprompt_factory = OutputPromptFactory(subfields, name_prefixes=self.name_prefixes + [field.name])
-            examples = [subprompt_factory._generate_schema(offset=i + 1) for i in range(max_count)]
+            # subprompt_factory = OutputPromptFactory(subfields, name_prefixes=self.name_prefixes + [field.name])
+            # examples = [subprompt_factory._generate_schema(offset=i + 1) for i in range(max_count)]
+            examples = [
+                OutputPromptFactory(subfields, name_prefixes=self.name_prefixes + [field.name + f" {i+1}"])._generate_schema(offset=i + 1)
+                for i in range(max_count)
+            ]
         else:
             field_name = " ".join(self.name_prefixes + [field.name])
             examples = [f"{field_name} {i+1}: <{info.instruction(ExamplePosition(i+1))}>" for i in range(max_count)]
@@ -31,8 +35,8 @@ def _generate_schema(self, offset: int = 0) -> str:
 
         for field in self.fields:
             field_name = " ".join(self.name_prefixes + [field.name])
-            if offset > 0:
-                field_name += f" {offset}"
+            # if offset > 0:
+            #     field_name += f" {offset}"
 
             if isinstance(field.info, LLMOutputInfo):
                 if field_type := if_response_type(field.type_):