janus-llm 3.4.2__tar.gz → 3.5.0__tar.gz

{janus_llm-3.4.2 → janus_llm-3.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: janus-llm
-Version: 3.4.2
+Version: 3.5.0
 Summary: A transcoding library using LLMs.
 Home-page: https://github.com/janus-llm/janus-llm
 License: Apache 2.0

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/__init__.py

@@ -5,7 +5,7 @@ from langchain_core._api.deprecation import LangChainDeprecationWarning
 from janus.converter.translate import Translator
 from janus.metrics import *  # noqa: F403
 
-__version__ = "3.4.2"
+__version__ = "3.5.0"
 
 # Ignoring a deprecation warning from langchain_core that I can't seem to hunt down
 warnings.filterwarnings("ignore", category=LangChainDeprecationWarning)

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/cli.py

@@ -2,7 +2,7 @@ import json
 import logging
 import os
 from pathlib import Path
-from typing import Optional
+from typing import List, Optional
 
 import click
 import typer
@@ -12,6 +12,8 @@ from rich.console import Console
 from rich.prompt import Confirm
 from typing_extensions import Annotated
 
+from janus.converter.aggregator import Aggregator
+from janus.converter.converter import Converter
 from janus.converter.diagram import DiagramGenerator
 from janus.converter.document import Documenter, MadLibsDocumenter, MultiDocumenter
 from janus.converter.requirements import RequirementsDocumenter
@@ -439,6 +441,126 @@ def document(
     documenter.translate(input_dir, output_dir, overwrite, collection)
 
 
+def get_subclasses(cls):
+    return set(cls.__subclasses__()).union(
+        set(s for c in cls.__subclasses__() for s in get_subclasses(c))
+    )
+
+
+@app.command()
+def aggregate(
+    input_dir: Annotated[
+        Path,
+        typer.Option(
+            "--input",
+            "-i",
+            help="The directory containing the source code to be translated. "
+            "The files should all be in one flat directory.",
+        ),
+    ],
+    language: Annotated[
+        str,
+        typer.Option(
+            "--language",
+            "-l",
+            help="The language of the source code.",
+            click_type=click.Choice(sorted(LANGUAGES)),
+        ),
+    ],
+    output_dir: Annotated[
+        Path,
+        typer.Option(
+            "--output-dir", "-o", help="The directory to store the translated code in."
+        ),
+    ],
+    llm_name: Annotated[
+        str,
+        typer.Option(
+            "--llm",
+            "-L",
+            help="The custom name of the model set with 'janus llm add'.",
+        ),
+    ] = "gpt-4o",
+    max_prompts: Annotated[
+        int,
+        typer.Option(
+            "--max-prompts",
+            "-m",
+            help="The maximum number of times to prompt a model on one functional block "
+            "before exiting the application. This is to prevent wasting too much money.",
+        ),
+    ] = 10,
+    overwrite: Annotated[
+        bool,
+        typer.Option(
+            "--overwrite/--preserve",
+            help="Whether to overwrite existing files in the output directory",
+        ),
+    ] = False,
+    temperature: Annotated[
+        float,
+        typer.Option("--temperature", "-t", help="Sampling temperature.", min=0, max=2),
+    ] = 0.7,
+    collection: Annotated[
+        str,
+        typer.Option(
+            "--collection",
+            "-c",
+            help="If set, will put the translated result into a Chroma DB "
+            "collection with the name provided.",
+        ),
+    ] = None,
+    splitter_type: Annotated[
+        str,
+        typer.Option(
+            "-S",
+            "--splitter",
+            help="Name of custom splitter to use",
+            click_type=click.Choice(list(CUSTOM_SPLITTERS.keys())),
+        ),
+    ] = "file",
+    intermediate_converters: Annotated[
+        List[str],
+        typer.Option(
+            "-C",
+            "--converter",
+            help="Name of an intermediate converter to use",
+            click_type=click.Choice([c.__name__ for c in get_subclasses(Converter)]),
+        ),
+    ] = ["Documenter"],
+):
+    converter_subclasses = get_subclasses(Converter)
+    converter_subclasses_map = {c.__name__: c for c in converter_subclasses}
+    model_arguments = dict(temperature=temperature)
+    collections_config = get_collections_config()
+    converters = []
+    for ic in intermediate_converters:
+        converters.append(
+            converter_subclasses_map[ic](
+                model=llm_name,
+                model_arguments=model_arguments,
+                source_language=language,
+                max_prompts=max_prompts,
+                db_path=db_loc,
+                db_config=collections_config,
+                splitter_type=splitter_type,
+            )
+        )
+
+    aggregator = Aggregator(
+        intermediate_converters=converters,
+        model=llm_name,
+        model_arguments=model_arguments,
+        source_language=language,
+        max_prompts=max_prompts,
+        db_path=db_loc,
+        db_config=collections_config,
+        splitter_type=splitter_type,
+        prompt_template="basic_aggregation",
+    )
+    aggregator.translate(input_dir, output_dir, overwrite, collection)
+
+
 @app.command(
     help="Diagram input code using an LLM.",
     no_args_is_help=True,

janus_llm-3.5.0/janus/converter/aggregator.py

@@ -0,0 +1,52 @@
+from copy import deepcopy
+from typing import List
+
+from janus.converter.converter import Converter
+from janus.language.block import CodeBlock, TranslatedCodeBlock
+
+
+class Aggregator(Converter):
+    def __init__(
+        self,
+        intermediate_converters: List[Converter],
+        separator: str = "\n==============\n",
+        **kwargs,
+    ):
+        """
+        Initialization Method
+        Arguments:
+            intermediate_converters - list of converters to use
+            separator: separator string to partition different outputs before combination
+        """
+        self._intermediate_converters = intermediate_converters
+        self._separator = separator
+        super().__init__(**kwargs)
+        self._load_parameters()
+
+    def _iterative_translate(self, root: CodeBlock) -> TranslatedCodeBlock:
+        res = TranslatedCodeBlock(root, language=self._target_language)
+        return self._recursive_translate(res)
+
+    def _recursive_translate(self, root: TranslatedCodeBlock) -> None:
+        """
+        Recursively translates code blocks from a bottom up approach
+        """
+        original_text = root.original.text
+        if len(root.children) > 0:
+            for c in root.children:
+                self._recursive_translate(c)
+            root.original.text = self._combine_blocks(root.children, self._separator)
+        else:
+            int_reps = [deepcopy(root) for ic in self._intermediate_converters]
+            for ic, r in zip(self._intermediate_converters, int_reps):
+                ic._add_translation(r)
+            root.original.text = self._combine_blocks(int_reps, self._separator)
+        self._add_translation(root)
+        root.original.text = original_text
+        return root
+
+    def _combine_blocks(self, blocks: List[TranslatedCodeBlock], separator: str) -> str:
+        """
+        Combines code blocks into a single piece of text
+        """
+        return separator.join([block.text for block in blocks])

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/converter/converter.py

@@ -7,7 +7,6 @@ from typing import Any, List, Optional, Tuple
 from langchain.output_parsers import RetryWithErrorOutputParser
 from langchain_core.exceptions import OutputParserException
 from langchain_core.language_models import BaseLanguageModel
-from langchain_core.output_parsers import BaseOutputParser
 from langchain_core.prompts import ChatPromptTemplate, SystemMessagePromptTemplate
 from langchain_core.runnables import RunnableLambda, RunnableParallel
 from openai import BadRequestError, RateLimitError
@@ -26,7 +25,7 @@ from janus.language.splitter import (
 from janus.llm import load_model
 from janus.llm.model_callbacks import get_model_callback
 from janus.llm.models_info import MODEL_PROMPT_ENGINES
-from janus.parsers.code_parser import GenericParser
+from janus.parsers.parser import GenericParser, JanusParser
 from janus.parsers.refiner_parser import RefinerParser
 from janus.refiners.refiner import BasicRefiner, Refiner
 from janus.utils.enums import LANGUAGES
@@ -129,7 +128,7 @@ class Converter:
         self._llm: BaseLanguageModel
         self._prompt: ChatPromptTemplate
 
-        self._parser: BaseOutputParser = GenericParser()
+        self._parser: JanusParser = GenericParser()
         self._combiner: Combiner = Combiner()
 
         self._refiner_type: str
@@ -328,6 +327,7 @@ class Converter:
         "_prompt_template_name",
         "_source_language",
         "_model_name",
+        "_parser",
     )
     def _load_prompt(self) -> None:
         """Load the prompt according to this instance's attributes.
@@ -340,6 +340,9 @@ class Converter:
             prompt_template=self._prompt_template_name,
         )
         self._prompt = prompt_engine.prompt
+        self._prompt = self._prompt.partial(
+            format_instructions=self._parser.get_format_instructions()
+        )
 
     @run_if_changed("_db_path", "_db_config")
     def _load_vectorizer(self) -> None:
@@ -445,6 +448,15 @@ class Converter:
             except FileSizeError:
                 log.warning("Current tile is too large for basic splitter, skipping")
                 continue
+            except ValueError as e:
+                if str(e).startswith(
+                    "Error raised by bedrock service"
+                ) and "maximum context length" in str(e):
+                    log.warning(
+                        "Input is too large for this model's context length, skipping"
+                    )
+                    continue
+                raise e
 
             # Don't attempt to write files for which translation failed
             if not out_block.translated:
@@ -599,7 +611,7 @@ class Converter:
         to the cube root of self.max_retries, so the total calls to the
         LLM will be roughly as expected (up to sqrt(self.max_retries) over)
         """
-        self._parser.set_reference(block.original)
+        input = self._parser.parse_input(block.original)
 
         # Retries with just the output and the error
         n1 = round(self.max_prompts ** (1 / 2))
@@ -607,15 +619,12 @@ class Converter:
         # Retries with the input, output, and error
         n2 = round(self.max_prompts // n1)
 
-        # Retries with just the input
         if not self.skip_context:
             self._make_prompt_additions(block)
         if not self.skip_refiner:  # Make replacements in the prompt
             refine_output = RefinerParser(
                 parser=self._parser,
-                initial_prompt=self._prompt.format(
-                    **{"SOURCE_CODE": block.original.text}
-                ),
+                initial_prompt=self._prompt.format(**{"SOURCE_CODE": input}),
                 refiner=self._refiner,
                 max_retries=n1,
                 llm=self._llm,
@@ -626,13 +635,14 @@ class Converter:
                 parser=self._parser,
                 max_retries=n1,
             )
+
         completion_chain = self._prompt | self._llm
         chain = RunnableParallel(
             completion=completion_chain, prompt_value=self._prompt
         ) | RunnableLambda(lambda x: refine_output.parse_with_prompt(**x))
         for _ in range(n2):
             try:
-                return chain.invoke({"SOURCE_CODE": block.original.text})
+                return chain.invoke({"SOURCE_CODE": input})
             except OutputParserException:
                 pass
 

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/converter/diagram.py

@@ -52,7 +52,7 @@ class DiagramGenerator(Documenter):
         self._load_diagram_prompt_engine()
 
     def _run_chain(self, block: TranslatedCodeBlock) -> str:
-        self._parser.set_reference(block.original)
+        input = self._parser.parse_input(block.original)
         n1 = round(self.max_prompts ** (1 / 3))
 
         # Retries with the input, output, and error
@@ -67,7 +67,7 @@ class DiagramGenerator(Documenter):
                 parser=self._diagram_parser,
                 initial_prompt=self._diagram_prompt.format(
                     **{
-                        "SOURCE_CODE": block.original.text,
+                        "SOURCE_CODE": input,
                         "DOCUMENTATION": documentation_text,
                         "DIAGRAM_TYPE": self._diagram_type,
                     }
@@ -81,7 +81,7 @@ class DiagramGenerator(Documenter):
                 parser=self._diagram_parser,
                 initial_prompt=self._diagram_prompt.format(
                     **{
-                        "SOURCE_CODE": block.original.text,
+                        "SOURCE_CODE": input,
                         "DIAGRAM_TYPE": self._diagram_type,
                     }
                 ),
@@ -103,7 +103,7 @@ class DiagramGenerator(Documenter):
                 if self._add_documentation:
                     return chain.invoke(
                         {
-                            "SOURCE_CODE": block.original.text,
+                            "SOURCE_CODE": input,
                             "DOCUMENTATION": documentation_text,
                             "DIAGRAM_TYPE": self._diagram_type,
                         }
@@ -111,7 +111,7 @@ class DiagramGenerator(Documenter):
                 else:
                     return chain.invoke(
                         {
-                            "SOURCE_CODE": block.original.text,
+                            "SOURCE_CODE": input,
                             "DIAGRAM_TYPE": self._diagram_type,
                         }
                     )

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/llm/models_info.py

@@ -210,7 +210,9 @@ def get_available_model_names() -> list[str]:
     return avaialable_models
 
 
-def load_model(user_model_name: str) -> tuple[BaseLanguageModel, int, dict[str, float]]:
+def load_model(
+    user_model_name: str,
+) -> tuple[BaseLanguageModel, str, int, dict[str, float]]:
     if not MODEL_CONFIG_DIR.exists():
         MODEL_CONFIG_DIR.mkdir(parents=True)
     model_config_file = MODEL_CONFIG_DIR / f"{user_model_name}.json"

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/parsers/_tests/test_code_parser.py

@@ -1,11 +1,12 @@
 import unittest
 
-from janus.parsers.code_parser import CodeParser, JanusParser
+from janus.parsers.code_parser import CodeParser
+from janus.parsers.parser import GenericParser
 
 
 class TestJanusParser(unittest.TestCase):
     def setUp(self):
-        self.parser = JanusParser()
+        self.parser = GenericParser()
 
     def test_parse_combined_output(self):
         text = "test text"

janus_llm-3.5.0/janus/parsers/code_parser.py

@@ -0,0 +1,27 @@
+import re
+
+from langchain_core.exceptions import OutputParserException
+from langchain_core.messages import BaseMessage
+
+from janus.parsers.parser import JanusParser
+from janus.utils.logger import create_logger
+
+log = create_logger(__name__)
+
+
+class CodeParser(JanusParser):
+    language: str
+
+    def parse(self, text: str | BaseMessage) -> str:
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
+        pattern = rf"```[^\S\r\n]*(?:{self.language}[^\S\r\n]*)?\n?(.*?)\n*```"
+        code = re.search(pattern, text, re.DOTALL)
+        if code is None:
+            raise OutputParserException(
+                "Code not find code between triple square brackets"
+            )
+        return str(code.group(1))
+
+    def get_format_instructions(self) -> str:
+        return "Output must contain text contained within triple square brackets (```)"

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/parsers/doc_parser.py

@@ -3,13 +3,12 @@ import re
 
 from langchain.output_parsers import PydanticOutputParser
 from langchain.output_parsers.json import parse_json_markdown
-from langchain.schema.output_parser import BaseOutputParser
 from langchain_core.exceptions import OutputParserException
-from langchain_core.messages import AIMessage
+from langchain_core.messages import BaseMessage
 from langchain_core.pydantic_v1 import BaseModel, Field
 
 from janus.language.block import CodeBlock
-from janus.parsers.code_parser import JanusParser
+from janus.parsers.parser import JanusParser
 from janus.utils.logger import create_logger
 
 log = create_logger(__name__)
@@ -32,18 +31,21 @@ class MultiDoc(BaseModel):
     )
 
 
-class MultiDocumentationParser(PydanticOutputParser, JanusParser):
+class MultiDocumentationParser(JanusParser, PydanticOutputParser):
     block_name: str = ""
 
     def __init__(self):
         PydanticOutputParser.__init__(self, pydantic_object=MultiDoc)
 
-    def set_reference(self, block: CodeBlock):
-        self.block_name = block.name
+    def parse_input(self, block: CodeBlock) -> str:
+        text = super().parse_input(block)
+        self.block_name = str(block.name)
+        return text
+
+    def parse(self, text: str | BaseMessage) -> str:
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
 
-    def parse(self, text: str) -> str:
-        if isinstance(text, AIMessage):
-            text = text.content
         try:
             docs = json.loads(super().parse(text).json())
         except (OutputParserException, json.JSONDecodeError):
@@ -81,22 +83,25 @@ class MultiDocumentationParser(PydanticOutputParser, JanusParser):
 
     @property
     def _type(self) -> str:
-        return self.__class__.name
+        return str(self.__class__.name)
 
 
-class MadlibsDocumentationParser(BaseOutputParser[str], JanusParser):
+class MadlibsDocumentationParser(JanusParser):
     expected_keys: set[str]
 
     def __init__(self):
         super().__init__(expected_keys=[])
 
-    def set_reference(self, block: CodeBlock):
-        comment_ids = re.findall(r"<(?:BLOCK|INLINE)_COMMENT (\w{8})>", block.text)
+    def parse_input(self, block: CodeBlock):
+        # TODO: Perform comment stripping/placeholding here rather than in script
+        text = super().parse_input(block)
+        comment_ids = re.findall(r"<(?:BLOCK|INLINE)_COMMENT (\w{8})>", text)
         self.expected_keys = set(comment_ids)
 
-    def parse(self, text: str) -> str:
-        if isinstance(text, AIMessage):
-            text = text.content
+    def parse(self, text: str | BaseMessage) -> str:
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
+
         try:
             obj = parse_json_markdown(text)
         except json.JSONDecodeError as e:
@@ -166,4 +171,4 @@ class MadlibsDocumentationParser(BaseOutputParser[str], JanusParser):
 
     @property
     def _type(self) -> str:
-        return self.__class__.name
+        return str(self.__class__.name)

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/parsers/eval_parser.py

@@ -1,9 +1,10 @@
 import json
 
 from langchain.output_parsers import PydanticOutputParser
+from langchain_core.messages import BaseMessage
 from langchain_core.pydantic_v1 import BaseModel, Field, validator
 
-from janus.parsers.code_parser import JanusParser
+from janus.parsers.parser import JanusParser
 from janus.utils.logger import create_logger
 
 log = create_logger(__name__)
@@ -28,7 +29,7 @@ class Eval(BaseModel):
 
         return v
 
-    def __add__(self, other):
+    def __add__(self, other: "Eval"):
         if isinstance(other, int) and other == 0:
             return self.copy()
         return Eval.construct(
@@ -38,10 +39,10 @@ class Eval(BaseModel):
             completeness=self.completeness + other.completeness,
         )
 
-    def __radd__(self, other):
+    def __radd__(self, other) -> "Eval":
         return self.__add__(other)
 
-    def __truediv__(self, other):
+    def __truediv__(self, other) -> "Eval":
         if isinstance(other, int):
             return Eval.construct(
                 syntax=self.syntax / other,
@@ -57,11 +58,13 @@ class Eval(BaseModel):
         )
 
 
-class EvaluationParser(PydanticOutputParser, JanusParser):
+class EvaluationParser(JanusParser, PydanticOutputParser):
     def __init__(self):
         PydanticOutputParser.__init__(self, pydantic_object=Eval)
 
-    def parse(self, text: str) -> str:
+    def parse(self, text: str | BaseMessage) -> str:
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
         eval = super().parse(text)
         return json.dumps(eval.json())
 
@@ -75,6 +78,6 @@ class EvaluationParser(PydanticOutputParser, JanusParser):
         Returns:
             A parsed version of the text.
         """
-        objs = [super().parse(line.strip()) for line in text.split("\n")]
-        avg_obj = sum(objs) / len(objs)
+        objs: list[Eval] = [super().parse(line.strip()) for line in text.split("\n")]
+        avg_obj: Eval = sum(objs) / len(objs)
         return json.dumps(avg_obj.json())

janus_llm-3.5.0/janus/parsers/parser.py

@@ -0,0 +1,51 @@
+from langchain.schema.output_parser import BaseOutputParser
+from langchain_core.messages import BaseMessage
+from langchain_core.output_parsers import StrOutputParser
+
+from janus.language.block import CodeBlock
+from janus.language.splitter import EmptyTreeError
+from janus.utils.logger import create_logger
+
+log = create_logger(__name__)
+
+
+class JanusParser(BaseOutputParser[str]):
+    def parse_input(self, block: CodeBlock) -> str:
+        """Parse the input block into raw string input ready to be passed to
+        an LLM. Also perform any processing or saving of metadata.
+
+        Arguments:
+            block: The CodeBlock to be processed
+
+        Returns:
+            A parsed version of the input text
+        """
+        if block.text is None:
+            raise EmptyTreeError("No text in input CodeBlock!")
+        return block.text
+
+    def parse_combined_output(self, text: str) -> str:
+        """Parse the output text from the LLM when multiple inputs are combined
+
+        Arguments:
+            text: The output text from the LLM
+
+        Returns:
+            A parsed version of the text
+        """
+        return text
+
+    def parse_into_block(self, text: str | BaseMessage, block: CodeBlock):
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
+        block.text = text
+
+
+class GenericParser(JanusParser, StrOutputParser):
+    def parse(self, text: str | BaseMessage) -> str:
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
+        return text
+
+    def get_format_instructions(self) -> str:
+        return "Output should be a string"

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/parsers/refiner_parser.py

@@ -1,11 +1,13 @@
 from langchain_core.exceptions import OutputParserException
 from langchain_core.language_models import BaseLanguageModel
+from langchain_core.messages import BaseMessage
 from langchain_core.output_parsers import BaseOutputParser
 
+from janus.parsers.parser import JanusParser
 from janus.refiners.refiner import Refiner
 
 
-class RefinerParser(BaseOutputParser):
+class RefinerParser(JanusParser):
     """Parser for performing refinement with a refiner
 
     Properties:
@@ -25,15 +27,7 @@ class RefinerParser(BaseOutputParser):
     refiner: Refiner
     max_retries: int
 
-    def parse(self, text: str) -> str:
-        """Parses the text using the refiner
-
-        Arguments:
-            text: text to parse
-
-        Returns:
-            Parsed text
-        """
+    def parse(self, text: str | BaseMessage) -> str:
         last_prompt = self.initial_prompt
         for _ in range(self.max_retries):
             try:
@@ -46,6 +40,7 @@ class RefinerParser(BaseOutputParser):
                 new_chain = new_prompt | self.llm
                 text = new_chain.invoke(prompt_arguments)
                 last_prompt = new_prompt.format(**prompt_arguments)
+
         raise OutputParserException(
             f"Error: unable to correct output after {self.max_retries} attempts"
         )

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/parsers/reqs_parser.py

@@ -2,29 +2,28 @@ import json
 import re
 
 from langchain.output_parsers.json import parse_json_markdown
-from langchain.schema.output_parser import BaseOutputParser
 from langchain_core.exceptions import OutputParserException
-from langchain_core.messages import AIMessage
+from langchain_core.messages import BaseMessage
 
-from janus.language.block import CodeBlock
-from janus.parsers.code_parser import JanusParser
+from janus.parsers.parser import JanusParser
 from janus.utils.logger import create_logger
 
 log = create_logger(__name__)
 
 
-class RequirementsParser(BaseOutputParser[str], JanusParser):
-    block_name: str = ""
+class RequirementsParser(JanusParser):
+    expected_keys: set[str]
 
     def __init__(self):
         super().__init__(expected_keys=[])
 
-    def set_reference(self, block: CodeBlock):
-        self.block_name = block.name
+    def parse(self, text: str | BaseMessage) -> str:
+        if isinstance(text, BaseMessage):
+            text = str(text.content)
 
-    def parse(self, text: str) -> str:
-        if isinstance(text, AIMessage):
-            text = text.content
+        # TODO: This is an incorrect implementation (lstrip and rstrip take character
+        #       lists and strip any instances of those characters, not the full str)
+        #       Should be replaced with a regex search, see CodeParser
         text = text.lstrip("```json")
         text = text.rstrip("```")
         try:
@@ -70,4 +69,4 @@ class RequirementsParser(BaseOutputParser[str], JanusParser):
 
     @property
     def _type(self) -> str:
-        return self.__class__.name
+        return str(self.__class__.name)

{janus_llm-3.4.2 → janus_llm-3.5.0}/janus/parsers/uml.py

@@ -4,6 +4,7 @@ from pathlib import Path
 from typing import List, Tuple
 
 from langchain_core.exceptions import OutputParserException
+from langchain_core.messages import BaseMessage
 
 from janus.parsers.code_parser import CodeParser
 from janus.utils.logger import create_logger
@@ -12,7 +13,7 @@ log = create_logger(__name__)
 
 
 class UMLSyntaxParser(CodeParser):
-    def _get_uml_output(self, file: str) -> Tuple[str, str]:
+    def _get_uml_output(self, file: Path) -> Tuple[str, str]:
         # NOTE: running subprocess with shell=False, added nosec to label that we know
         # risk exists
         try:
@@ -33,7 +34,7 @@ class UMLSyntaxParser(CodeParser):
     def _get_errs(self, s: str) -> List[str]:
         return [x.group() for x in re.finditer(r"Error (.*)\n", s)]
 
-    def parse(self, text: str) -> str:
+    def parse(self, text: str | BaseMessage) -> str:
         text = super().parse(text)
         janus_path = Path.home().expanduser() / Path(".janus")
         if not janus_path.exists():

{janus_llm-3.4.2 → janus_llm-3.5.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "janus-llm"
-version = "3.4.2"
+version = "3.5.0"
 description = "A transcoding library using LLMs."
 authors = ["Michael Doyle <mdoyle@mitre.org>", "Chris Glasz <cglasz@mitre.org>",
            "Chris Tohline <ctohline@mitre.org>", "William Macke <wmacke@mitre.org>",

janus_llm-3.4.2/janus/parsers/code_parser.py

@@ -1,59 +0,0 @@
-import re
-
-from langchain.schema.output_parser import BaseOutputParser
-from langchain_core.exceptions import OutputParserException
-from langchain_core.messages import BaseMessage
-from langchain_core.output_parsers import StrOutputParser
-
-from janus.language.block import CodeBlock
-from janus.utils.logger import create_logger
-
-log = create_logger(__name__)
-
-
-class JanusParser:
-    def parse_combined_output(self, text: str) -> str:
-        """Parse the output text from the LLM when multiple inputs are combined
-
-        Arguments:
-            text: The output text from the LLM
-
-        Returns:
-            A parsed version of the text
-        """
-        if isinstance(text, BaseMessage):
-            text = text.content
-        return text
-
-    def parse_into_block(self, text: str, block: CodeBlock):
-        if isinstance(text, BaseMessage):
-            text = text.content
-        block.text = text
-
-    def set_reference(self, block: CodeBlock):
-        pass
-
-
-class GenericParser(StrOutputParser, JanusParser):
-    def parse(self, text: str) -> str:
-        if isinstance(text, BaseMessage):
-            return text.content
-        return text
-
-
-class CodeParser(BaseOutputParser[str], JanusParser):
-    language: str
-
-    def parse(self, text: str) -> str:
-        if isinstance(text, BaseMessage):
-            text = text.content
-        pattern = rf"```[^\S\r\n]*(?:{self.language}[^\S\r\n]*)?\n?(.*?)\n*```"
-        code = re.search(pattern, text, re.DOTALL)
-        if code is None:
-            raise OutputParserException(
-                "Code not find code between triple square brackets"
-            )
-        return str(code.group(1))
-
-    def get_format_instructions(self) -> str:
-        return "Output must contain text contained within triple square brackets (```)"