fabricatio 0.2.10.dev0__cp312-cp312-manylinux_2_34_x86_64.whl → 0.2.11.dev0__cp312-cp312-manylinux_2_34_x86_64.whl

fabricatio/actions/article.py

@@ -78,7 +78,7 @@ class FixArticleEssence(Action):
         out = []
         count = 0
         for a in article_essence:
-            if key := (bib_mgr.get_cite_key(a.title) or bib_mgr.get_cite_key_fuzzy(a.title)):
+            if key := (bib_mgr.get_cite_key_by_title(a.title) or bib_mgr.get_cite_key_fuzzy(a.title)):
                 a.title = bib_mgr.get_title_by_key(key) or a.title
                 a.authors = bib_mgr.get_author_by_key(key) or a.authors
                 a.publication_year = bib_mgr.get_year_by_key(key) or a.publication_year
@@ -145,7 +145,8 @@ class GenerateInitialOutline(Action, Propose):
             f"{(article_proposal.as_prompt())}\n\nNote that you should use `{article_proposal.language}` to write the `ArticleOutline`\n"
             f"Design each chapter of a proper and academic and ready for release manner.\n"
             f"You Must make sure every chapter have sections, and every section have subsections.\n"
-            f"Make the chapter and sections and subsections bing divided into a specific enough article component.",
+            f"Make the chapter and sections and subsections bing divided into a specific enough article component.\n"
+            f"Every chapter must have sections, every section must have subsections.",
         )
 
         return ok(
@@ -226,3 +227,12 @@ class GenerateArticle(Action, Censor):
         )
 
         return article
+
+
+class LoadArticle(Action):
+    """Load the article from the outline and typst code."""
+
+    output_key: str = "loaded_article"
+
+    async def _execute(self, article_outline: ArticleOutline, typst_code: str, **cxt) -> Article:
+        return Article.from_mixed_source(article_outline, typst_code)

fabricatio/actions/article_rag.py

@@ -1,16 +1,114 @@
 """A module for writing articles using RAG (Retrieval-Augmented Generation) capabilities."""
 
 from asyncio import gather
-from typing import Optional
+from pathlib import Path
+from typing import List, Optional
 
+from fabricatio import BibManager
 from fabricatio.capabilities.censor import Censor
+from fabricatio.capabilities.propose import Propose
 from fabricatio.capabilities.rag import RAG
+from fabricatio.journal import logger
 from fabricatio.models.action import Action
-from fabricatio.models.extra.article_main import Article, ArticleSubsection
+from fabricatio.models.extra.aricle_rag import ArticleChunk, CitationManager
+from fabricatio.models.extra.article_essence import ArticleEssence
+from fabricatio.models.extra.article_main import Article, ArticleChapter, ArticleSection, ArticleSubsection
+from fabricatio.models.extra.article_outline import ArticleOutline
 from fabricatio.models.extra.rule import RuleSet
 from fabricatio.utils import ok
 
 
+class WriteArticleContentRAG(Action, RAG, Propose):
+    """Write an article based on the provided outline."""
+
+    ref_limit: int = 100
+    """The limit of references to be retrieved"""
+    extractor_model: str
+    """The model to use for extracting the content from the retrieved references."""
+    query_model: str
+    """The model to use for querying the database"""
+
+    async def _execute(
+        self,
+        article_outline: ArticleOutline,
+        writing_ruleset: RuleSet,
+        collection_name: str = "article_chunks",
+        **cxt,
+    ) -> Article:
+        article = Article.from_outline(article_outline).update_ref(article_outline)
+        await gather(
+            *[
+                self._inner(article, article_outline, chap, sec, subsec)
+                for chap, sec, subsec in article.iter_subsections()
+            ]
+        )
+        return article.convert_tex()
+
+    async def _inner(
+        self,
+        article: Article,
+        article_outline: ArticleOutline,
+        chap: ArticleChapter,
+        sec: ArticleSection,
+        subsec: ArticleSubsection,
+    ) -> ArticleSubsection:
+        ref_q = ok(
+            await self.arefined_query(
+                f"{article_outline.display()}\n\nAbove is my article outline, I m writing graduate thesis titled `{article.title}`. "
+                f"More specifically, i m witting the Chapter `{chap.title}` >> Section `{sec.title}` >> Subsection `{subsec.title}`.\n"
+                f"I need to search related references to build up the content of the subsec mentioned above, which is `{subsec.title}`.\n"
+                f"plus, you can search required formulas by using latex equation code.\n"
+                f"provide 10 queries as possible, to get best result!\n"
+                f"You should provide both English version and chinese version of the refined queries!\n",
+                model=self.query_model,
+            ),
+            "Failed to refine query.",
+        )
+        ret = await self.aretrieve(ref_q, ArticleChunk, final_limit=self.ref_limit, result_per_query=25)
+        ret.reverse()
+        cm = CitationManager().update_chunks(ret)
+
+        raw_paras = await self.aask(
+            f"{cm.as_prompt()}\nAbove is some related reference retrieved for you. When need to cite some of them ,you MUST follow the academic convention,"
+            f"{article.referenced.display()}\n\nAbove is my article outline, I m writing graduate thesis titled `{article.title}`. "
+            f"More specifically, i m witting the Chapter `{chap.title}` >> Section `{sec.title}` >> Subsection `{subsec.title}`.\n"
+            f"Please help me write the paragraphs of the subsec mentioned above, which is `{subsec.title}`\n"
+            f"You can output the written paragraphs directly, without explanation. you should use `{subsec.language}`, and maintain academic writing style."
+            f"In addition,you MUST follow the academic convention and use [[1]] to cite the first reference, and use [[9]] to cite the second reference, and so on.\n"
+            f"It 's greatly recommended to cite multiple references that stands for the same opinion at a single sentences, like [[1,5,9]] for 1th,5th and 9th references,[[1-9,16]] for 1th to 9th and 16th references.\n"
+            f"citation number is REQUIRED to cite any reference!\n"
+            f"for paragraphs that need write equation you should also no forget to doing so. wrapp inline equation using $ $, and wrapp block equation using $$ $$.\n"
+        )
+
+        raw_paras = (
+            raw_paras.replace(r" \( ", "$").replace(r" \) ", "$").replace("\\[\n", "$$\n").replace("\n\\]", "\n$$")
+        )
+
+        new_subsec = ok(
+            await self.propose(
+                ArticleSubsection,
+                f"{raw_paras}\nAbove is the subsection titled `{subsec.title}`.\n"
+                f"I need you to extract the content to update my subsection obj provided below.\n"
+                f"Everything is build upon the typst language, which is similar to latex, \n"
+                f"so reference annotation like `[[1]]` for 1th reference or `[[2,6]]` for 2th and 6th reference or "
+                f"`[[1,5,9]]` for 1th,5th and 9th references or "
+                f"`[[1-9,16]]` for 1th to 9th and 16th references\n"
+                f"Those reference mark shall not be omitted during the extraction\n"
+                f"Wrapp inline expression using $ $, and wrapp block equation using $$ $$\n\n\n"
+                f"{subsec.display()}",
+                model=self.extractor_model,
+            ),
+            "Failed to propose new subsection.",
+        )
+
+        for p in new_subsec.paragraphs:
+            p.content = cm.apply(p.content)
+
+        subsec.update_from(new_subsec)
+        logger.debug(f"{subsec.title}:rpl\n{subsec.display()}")
+        return subsec
+
+
 class TweakArticleRAG(Action, RAG, Censor):
     """Write an article based on the provided outline.
 
@@ -39,7 +137,7 @@ class TweakArticleRAG(Action, RAG, Censor):
         twk_rag_ruleset: Optional[RuleSet] = None,
         parallel: bool = False,
         **cxt,
-    ) -> Optional[Article]:
+    ) -> Article:
         """Write an article based on the provided outline.
 
         This method processes the article outline, either in parallel or sequentially, by enhancing each subsection
@@ -53,7 +151,7 @@ class TweakArticleRAG(Action, RAG, Censor):
             **cxt: Additional context parameters.
 
         Returns:
-            Optional[Article]: The processed article with enhanced subsections and applied censoring rules.
+            Article: The processed article with enhanced subsections and applied censoring rules.
         """
         self.view(collection_name)
 
@@ -86,20 +184,43 @@ class TweakArticleRAG(Action, RAG, Censor):
         """
         refind_q = ok(
             await self.arefined_query(
-                f"{article.referenced.as_prompt()}\n"
-                f"# Subsection requiring reference enhancement\n"
-                f"{subsec.display()}\n"
-                f"# Requirement\n"
-                f"Search related articles in the base to find reference candidates, "
-                f"provide queries in both `English` and `{subsec.language}` can get more accurate results.",
+                f"{article.referenced.as_prompt()}\n# Subsection requiring reference enhancement\n{subsec.display()}\n"
             )
         )
         await self.censor_obj_inplace(
             subsec,
             ruleset=ruleset,
-            reference=f"{await self.aretrieve_compact(refind_q, final_limit=self.ref_limit)}\n\n"
+            reference=f"{'\n\n'.join(d.display() for d in await self.aretrieve(refind_q, document_model=ArticleEssence, final_limit=self.ref_limit))}\n\n"
             f"You can use Reference above to rewrite the `{subsec.__class__.__name__}`.\n"
             f"You should Always use `{subsec.language}` as written language, "
             f"which is the original language of the `{subsec.title}`. "
             f"since rewrite a `{subsec.__class__.__name__}` in a different language is usually a bad choice",
         )
+
+
+class ChunkArticle(Action):
+    """Chunk an article into smaller chunks."""
+
+    output_key: str = "article_chunks"
+    """The key used to store the output of the action."""
+    max_chunk_size: Optional[int] = None
+    """The maximum size of each chunk."""
+    max_overlapping_rate: Optional[float] = None
+    """The maximum overlapping rate between chunks."""
+
+    async def _execute(
+        self,
+        article_path: str | Path,
+        bib_manager: BibManager,
+        max_chunk_size: Optional[int] = None,
+        max_overlapping_rate: Optional[float] = None,
+        **_,
+    ) -> List[ArticleChunk]:
+        return ArticleChunk.from_file(
+            article_path,
+            bib_manager,
+            max_chunk_size=ok(max_chunk_size or self.max_chunk_size, "No max_chunk_size provided!"),
+            max_overlapping_rate=ok(
+                max_overlapping_rate or self.max_overlapping_rate, "No max_overlapping_rate provided!"
+            ),
+        )

fabricatio/actions/fs.py

@@ -0,0 +1,25 @@
+"""A module for file system utilities."""
+
+from pathlib import Path
+from typing import Any, List, Mapping, Self
+
+from fabricatio.fs import safe_text_read
+from fabricatio.journal import logger
+from fabricatio.models.action import Action
+from fabricatio.models.generic import FromMapping
+
+
+class ReadText(Action, FromMapping):
+    """Read text from a file."""
+    output_key: str = "read_text"
+    read_path: str | Path
+    """Path to the file to read."""
+
+    async def _execute(self, *_: Any, **cxt) -> str:
+        logger.info(f"Read text from {Path(self.read_path).as_posix()} to {self.output_key}")
+        return safe_text_read(self.read_path)
+
+    @classmethod
+    def from_mapping(cls, mapping: Mapping[str, str | Path], **kwargs: Any) -> List[Self]:
+        """Create a list of ReadText actions from a mapping of output_key to read_path."""
+        return [cls(read_path=p, output_key=k, **kwargs) for k, p in mapping.items()]

fabricatio/actions/output.py

@@ -1,11 +1,11 @@
 """Dump the finalized output to a file."""
 
 from pathlib import Path
-from typing import Any, Iterable, List, Optional, Type
+from typing import Any, Iterable, List, Mapping, Optional, Type
 
 from fabricatio.journal import logger
 from fabricatio.models.action import Action
-from fabricatio.models.generic import FinalizedDumpAble, PersistentAble
+from fabricatio.models.generic import FinalizedDumpAble, FromMapping, PersistentAble
 from fabricatio.models.task import Task
 from fabricatio.utils import ok
 
@@ -115,7 +115,7 @@ class RetrieveFromPersistent[T: PersistentAble](Action):
         return self.retrieve_cls.from_persistent(self.load_path)
 
 
-class RetrieveFromLatest[T: PersistentAble](RetrieveFromPersistent[T]):
+class RetrieveFromLatest[T: PersistentAble](RetrieveFromPersistent[T], FromMapping):
     """Retrieve the object from the latest persistent file in the dir at `load_path`."""
 
     async def _execute(self, /, **_) -> Optional[T]:
@@ -130,6 +130,20 @@ class RetrieveFromLatest[T: PersistentAble](RetrieveFromPersistent[T]):
         logger.error(f"Path {self.load_path} is not a directory")
         return None
 
+    @classmethod
+    def from_mapping(
+        cls,
+        mapping: Mapping[str, str | Path],
+        *,
+        retrieve_cls: Type[T],
+        **kwargs,
+    ) -> List["RetrieveFromLatest[T]"]:
+        """Create a list of `RetrieveFromLatest` from the mapping."""
+        return [
+            cls(retrieve_cls=retrieve_cls, load_path=Path(p).as_posix(), output_key=o, **kwargs)
+            for o, p in mapping.items()
+        ]
+
 
 class GatherAsList(Action):
     """Gather the objects from the context as a list.

fabricatio/actions/rag.py

@@ -2,37 +2,57 @@
 
 from typing import List, Optional
 
-from questionary import text
-
 from fabricatio.capabilities.rag import RAG
+from fabricatio.config import configs
 from fabricatio.journal import logger
 from fabricatio.models.action import Action
-from fabricatio.models.generic import Vectorizable
+from fabricatio.models.extra.rag import MilvusClassicModel, MilvusDataBase
 from fabricatio.models.task import Task
+from fabricatio.utils import ok
 
 
 class InjectToDB(Action, RAG):
     """Inject data into the database."""
 
     output_key: str = "collection_name"
+    collection_name: str = "my_collection"
+    """The name of the collection to inject data into."""
 
-    async def _execute[T: Vectorizable](
-        self, to_inject: Optional[T] | List[Optional[T]], collection_name: str = "my_collection",override_inject:bool=False, **_
+    async def _execute[T: MilvusDataBase](
+            self, to_inject: Optional[T] | List[Optional[T]], override_inject: bool = False, **_
     ) -> Optional[str]:
+        from pymilvus.milvus_client import IndexParams
+
+        if to_inject is None:
+            return None
         if not isinstance(to_inject, list):
             to_inject = [to_inject]
-        logger.info(f"Injecting {len(to_inject)} items into the collection '{collection_name}'")
+        if not (seq := [t for t in to_inject if t is not None]):  # filter out None
+            return None
+        logger.info(f"Injecting {len(seq)} items into the collection '{self.collection_name}'")
         if override_inject:
-            self.check_client().client.drop_collection(collection_name)
-        await self.view(collection_name, create=True).consume_string(
-            [
-                t.prepare_vectorization(self.embedding_max_sequence_length)
-                for t in to_inject
-                if isinstance(t, Vectorizable)
-            ],
-        )
-
-        return collection_name
+            self.check_client().client.drop_collection(self.collection_name)
+
+        await self.view(
+            self.collection_name,
+            create=True,
+            schema=seq[0].as_milvus_schema(
+                ok(
+                    self.milvus_dimensions
+                    or configs.rag.milvus_dimensions
+                    or self.embedding_dimensions
+                    or configs.embedding.dimensions
+                ),
+            ),
+            index_params=IndexParams(
+                seq[0].vector_field_name,
+                index_name=seq[0].vector_field_name,
+                index_type=seq[0].index_type,
+                metric_type=seq[0].metric_type,
+            ),
+        ).add_document(seq, flush=True)
+
+        return self.collection_name
 
 
 class RAGTalk(Action, RAG):
@@ -52,6 +72,8 @@ class RAGTalk(Action, RAG):
     output_key: str = "task_output"
 
     async def _execute(self, task_input: Task[str], **kwargs) -> int:
+        from questionary import text
+
         collection_name = kwargs.get("collection_name", "my_collection")
         counter = 0
 
@@ -62,10 +84,10 @@ class RAGTalk(Action, RAG):
                 user_say = await text("User: ").ask_async()
                 if user_say is None:
                     break
-                gpt_say = await self.aask_retrieved(
-                    user_say,
-                    user_say,
-                    extra_system_message=f"You have to answer to user obeying task assigned to you:\n{task_input.briefing}",
+                ret: List[MilvusClassicModel] = await self.aretrieve(user_say, document_model=MilvusClassicModel)
+
+                gpt_say = await self.aask(
+                    user_say, system_message="\n".join(m.text for m in ret) + "\nYou can refer facts provided above."
                 )
                 print(f"GPT: {gpt_say}")  # noqa: T201
                 counter += 1

fabricatio/actions/rules.py

@@ -1,15 +1,16 @@
 """A module containing the DraftRuleSet action."""
 
-from typing import List, Optional
+from typing import Any, List, Mapping, Optional, Self, Tuple
 
 from fabricatio.capabilities.check import Check
 from fabricatio.journal import logger
 from fabricatio.models.action import Action
 from fabricatio.models.extra.rule import RuleSet
+from fabricatio.models.generic import FromMapping
 from fabricatio.utils import ok
 
 
-class DraftRuleSet(Action, Check):
+class DraftRuleSet(Action, Check, FromMapping):
     """Action to draft a ruleset based on a given requirement description."""
 
     output_key: str = "drafted_ruleset"
@@ -45,8 +46,13 @@ class DraftRuleSet(Action, Check):
             logger.warning(f"Drafting Rule Failed for:\n{ruleset_requirement}")
         return ruleset
 
+    @classmethod
+    def from_mapping(cls, mapping: Mapping[str, Tuple[int, str]], **kwargs) -> List[Self]:
+        """Create a list of DraftRuleSet actions from a mapping of output keys to tuples of rule counts and requirements."""
+        return [cls(ruleset_requirement=r, rule_count=c, output_key=k, **kwargs) for k, (c, r) in mapping.items()]
 
-class GatherRuleset(Action):
+
+class GatherRuleset(Action, FromMapping):
     """Action to gather a ruleset from a given requirement description."""
 
     output_key: str = "gathered_ruleset"
@@ -55,6 +61,11 @@ class GatherRuleset(Action):
     to_gather: List[str]
     """the cxt name of RuleSet to gather"""
 
+    @classmethod
+    def from_mapping(cls, mapping: Mapping[str, List[str]], **kwargs: Any) -> List[Self]:
+        """Create a list of GatherRuleset actions from a mapping of output keys to tuples of rule counts and requirements."""
+        return [cls(to_gather=t, output_key=k, **kwargs) for k, t in mapping.items()]
+
     async def _execute(self, **cxt) -> RuleSet:
         logger.info(f"Gathering Ruleset from {self.to_gather}")
         # Fix for not_found

fabricatio/capabilities/extract.py

@@ -0,0 +1,65 @@
+"""A module that provide capabilities for extracting information from a given source to a model."""
+
+from typing import List, Optional, Type, Unpack, overload
+
+from fabricatio import TEMPLATE_MANAGER
+from fabricatio.capabilities.propose import Propose
+from fabricatio.config import configs
+from fabricatio.models.generic import ProposedAble
+from fabricatio.models.kwargs_types import ValidateKwargs
+
+
+class Extract(Propose):
+    """A class that extract information from a given source to a model."""
+
+    @overload
+    async def extract[M: ProposedAble](
+        self,
+        cls: Type[M],
+        source: str,
+        extract_requirement: Optional[str] = None,
+        **kwargs: Unpack[ValidateKwargs[M]],
+    ) -> M: ...
+    @overload
+    async def extract[M: ProposedAble](
+        self,
+        cls: Type[M],
+        source: str,
+        extract_requirement: Optional[str] = None,
+        **kwargs: Unpack[ValidateKwargs[None]],
+    ) -> Optional[M]: ...
+
+    @overload
+    async def extract[M: ProposedAble](
+        self,
+        cls: Type[M],
+        source: List[str],
+        extract_requirement: Optional[str] = None,
+        **kwargs: Unpack[ValidateKwargs[M]],
+    ) -> List[M]: ...
+    @overload
+    async def extract[M: ProposedAble](
+        self,
+        cls: Type[M],
+        source: List[str],
+        extract_requirement: Optional[str] = None,
+        **kwargs: Unpack[ValidateKwargs[None]],
+    ) -> List[Optional[M]]: ...
+    async def extract[M: ProposedAble](
+        self,
+        cls: Type[M],
+        source: List[str] | str,
+        extract_requirement: Optional[str] = None,
+        **kwargs: Unpack[ValidateKwargs[Optional[M]]],
+    ) -> M | List[M] | Optional[M] | List[Optional[M]]:
+        """Extract information from a given source to a model."""
+        return await self.propose(
+            cls,
+            prompt=TEMPLATE_MANAGER.render_template(
+                configs.templates.extract_template,
+                [{"source": s, "extract_requirement": extract_requirement} for s in source]
+                if isinstance(source, list)
+                else {"source": source, "extract_requirement": extract_requirement},
+            ),
+            **kwargs,
+        )

fabricatio/capabilities/rag.py

@@ -130,7 +130,7 @@ class RAG(EmbeddingUsage):
         if isinstance(data, MilvusDataBase):
             data = [data]
 
-        data_vec = await self.vectorize([d.to_vectorize for d in data])
+        data_vec = await self.vectorize([d.prepare_vectorization() for d in data])
         prepared_data = [d.prepare_insertion(vec) for d, vec in zip(data, data_vec, strict=True)]
 
         c_name = collection_name or self.safe_target_collection
@@ -188,13 +188,15 @@ class RAG(EmbeddingUsage):
     async def aretrieve[D: MilvusDataBase](
         self,
         query: List[str] | str,
+        document_model: Type[D],
         final_limit: int = 20,
-        **kwargs: Unpack[FetchKwargs[D]],
+        **kwargs: Unpack[FetchKwargs],
     ) -> List[D]:
         """Retrieve data from the collection.
 
         Args:
             query (List[str] | str): The query to be used for retrieval.
+            document_model (Type[D]): The model class used to convert retrieved data into document objects.
             final_limit (int): The final limit on the number of results to return.
             **kwargs (Unpack[FetchKwargs]): Additional keyword arguments for retrieval.
 
@@ -206,6 +208,7 @@ class RAG(EmbeddingUsage):
         return (
             await self.afetch_document(
                 vecs=(await self.vectorize(query)),
+                document_model=document_model,
                 **kwargs,
             )
         )[:final_limit]

fabricatio/capabilities/rating.py

@@ -14,7 +14,7 @@ from fabricatio.models.generic import Display, ProposedAble
 from fabricatio.models.kwargs_types import CompositeScoreKwargs, ValidateKwargs
 from fabricatio.parser import JsonCapture
 from fabricatio.rust_instances import TEMPLATE_MANAGER
-from fabricatio.utils import fallback_kwargs, ok, override_kwargs
+from fabricatio.utils import ok, override_kwargs
 
 
 class Rating(Propose):
@@ -137,7 +137,7 @@ class Rating(Propose):
             or dict(zip(criteria, criteria, strict=True))
         )
 
-        return await self.rate_fine_grind(to_rate, manual, score_range, **fallback_kwargs(kwargs, co_extractor={}))
+        return await self.rate_fine_grind(to_rate, manual, score_range, **kwargs)
 
     async def draft_rating_manual(
         self, topic: str, criteria: Optional[Set[str]] = None, **kwargs: Unpack[ValidateKwargs[Dict[str, str]]]
@@ -338,6 +338,7 @@ class Rating(Propose):
         criteria: Optional[Set[str]] = None,
         weights: Optional[Dict[str, float]] = None,
         manual: Optional[Dict[str, str]] = None,
+        approx: bool = False,
         **kwargs: Unpack[ValidateKwargs[List[Dict[str, float]]]],
     ) -> List[float]:
         """Calculates the composite scores for a list of items based on a given topic and criteria.
@@ -348,6 +349,7 @@ class Rating(Propose):
             criteria (Optional[Set[str]]): A set of criteria for the rating. Defaults to None.
             weights (Optional[Dict[str, float]]): A dictionary of rating weights for each criterion. Defaults to None.
             manual (Optional[Dict[str, str]]): A dictionary of manual ratings for each item. Defaults to None.
+            approx (bool): Whether to use approximate rating criteria. Defaults to False.
             **kwargs (Unpack[ValidateKwargs]): Additional keyword arguments for the LLM usage.
 
         Returns:
@@ -355,6 +357,7 @@ class Rating(Propose):
         """
         criteria = ok(
             criteria
+            or (await self.draft_rating_criteria(topic, **override_kwargs(kwargs, default=None)) if approx else None)
             or await self.draft_rating_criteria_from_examples(topic, to_rate, **override_kwargs(kwargs, default=None))
         )
         weights = ok(

fabricatio/capabilities/task.py

@@ -3,7 +3,7 @@
 from types import CodeType
 from typing import Any, Dict, List, Optional, Tuple, Unpack
 
-import orjson
+import ujson
 
 from fabricatio.capabilities.propose import Propose
 from fabricatio.config import configs
@@ -20,9 +20,9 @@ class ProposeTask(Propose):
     """A class that proposes a task based on a prompt."""
 
     async def propose_task[T](
-        self,
-        prompt: str,
-        **kwargs: Unpack[ValidateKwargs[Task[T]]],
+            self,
+            prompt: str,
+            **kwargs: Unpack[ValidateKwargs[Task[T]]],
     ) -> Optional[Task[T]]:
         """Asynchronously proposes a task based on a given prompt and parameters.
 
@@ -44,11 +44,11 @@ class HandleTask(ToolBoxUsage):
     """A class that handles a task based on a task object."""
 
     async def draft_tool_usage_code(
-        self,
-        task: Task,
-        tools: List[Tool],
-        data: Dict[str, Any],
-        **kwargs: Unpack[ValidateKwargs],
+            self,
+            task: Task,
+            tools: List[Tool],
+            data: Dict[str, Any],
+            **kwargs: Unpack[ValidateKwargs],
     ) -> Optional[Tuple[CodeType, List[str]]]:
         """Asynchronously drafts the tool usage code for a task based on a given task object and tools."""
         logger.info(f"Drafting tool usage code for task: {task.briefing}")
@@ -60,7 +60,7 @@ class HandleTask(ToolBoxUsage):
 
         def _validator(response: str) -> Tuple[CodeType, List[str]] | None:
             if (source := PythonCapture.convert_with(response, lambda resp: compile(resp, "<string>", "exec"))) and (
-                to_extract := JsonCapture.convert_with(response, orjson.loads)
+                    to_extract := JsonCapture.convert_with(response, ujson.loads)
             ):
                 return source, to_extract
 
@@ -85,12 +85,12 @@ class HandleTask(ToolBoxUsage):
         )
 
     async def handle_fine_grind(
-        self,
-        task: Task,
-        data: Dict[str, Any],
-        box_choose_kwargs: Optional[ChooseKwargs] = None,
-        tool_choose_kwargs: Optional[ChooseKwargs] = None,
-        **kwargs: Unpack[ValidateKwargs],
+            self,
+            task: Task,
+            data: Dict[str, Any],
+            box_choose_kwargs: Optional[ChooseKwargs] = None,
+            tool_choose_kwargs: Optional[ChooseKwargs] = None,
+            **kwargs: Unpack[ValidateKwargs],
     ) -> Optional[Tuple]:
         """Asynchronously handles a task based on a given task object and parameters."""
         logger.info(f"Handling task: \n{task.briefing}")

fabricatio/config.py

@@ -86,8 +86,10 @@ class LLMConfig(BaseModel):
 
     tpm: Optional[PositiveInt] = Field(default=1000000)
     """The rate limit of the LLM model in tokens per minute. None means not checked."""
-
-
+    presence_penalty:Optional[PositiveFloat]=None
+    """The presence penalty of the LLM model."""
+    frequency_penalty:Optional[PositiveFloat]=None
+    """The frequency penalty of the LLM model."""
 class EmbeddingConfig(BaseModel):
     """Embedding configuration class."""
 
@@ -249,6 +251,11 @@ class TemplateConfig(BaseModel):
 
     rule_requirement_template: str = Field(default="rule_requirement")
     """The name of the rule requirement template which will be used to generate a rule requirement."""
+
+
+    extract_template: str = Field(default="extract")
+    """The name of the extract template which will be used to extract model from string."""
+    
 class MagikaConfig(BaseModel):
     """Magika configuration class."""