fabricatio 0.2.12.dev1__tar.gz → 0.2.12.dev3__tar.gz

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/Cargo.lock

@@ -76,6 +76,12 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "anyhow"
+version = "1.0.98"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e16d2d3311acee920a9eb8d33b8cbc1787ce4a264e85f964c2404b969bdcd487"
+
 [[package]]
 name = "arrayref"
 version = "0.3.9"
@@ -462,6 +468,7 @@ dependencies = [
  "rayon",
  "regex",
  "serde_json",
+ "serde_yml",
  "tex2typst-rs",
  "unicode-segmentation",
  "walkdir",
@@ -980,6 +987,16 @@ dependencies = [
  "redox_syscall",
 ]
 
+[[package]]
+name = "libyml"
+version = "0.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3302702afa434ffa30847a83305f0a69d6abd74293b6554c18ec85c7ef30c980"
+dependencies = [
+ "anyhow",
+ "version_check",
+]
+
 [[package]]
 name = "linux-raw-sys"
 version = "0.9.3"
@@ -1670,6 +1687,21 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "serde_yml"
+version = "0.0.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59e2dd588bf1597a252c3b920e0143eb99b0f76e4e082f4c92ce34fbc9e71ddd"
+dependencies = [
+ "indexmap",
+ "itoa",
+ "libyml",
+ "memchr",
+ "ryu",
+ "serde",
+ "version_check",
+]
+
 [[package]]
 name = "sha2"
 version = "0.10.8"

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/Cargo.toml

@@ -15,6 +15,7 @@ pythonize = "0.23.0"
 rayon = "1.10.0"
 regex = "1.11.1"
 serde_json = "1.0.138"
+serde_yml = "0.0.12"
 tex2typst-rs = "0.4.1"
 unicode-segmentation = "1.12.0"
 walkdir = "2.5.0"

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fabricatio
-Version: 0.2.12.dev1
+Version: 0.2.12.dev3
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: 3.12

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/examples/write_article/write_article.py

@@ -7,7 +7,7 @@ import typer
 from fabricatio import Event, Role, WorkFlow, logger
 from fabricatio.actions.article import ExtractOutlineFromRaw, GenerateArticleProposal, GenerateInitialOutline
 from fabricatio.actions.article_rag import ArticleConsultRAG, WriteArticleContentRAG
-from fabricatio.actions.output import DumpFinalizedOutput, PersistentAll
+from fabricatio.actions.output import DumpFinalizedOutput, PersistentAll, RenderedDump
 from fabricatio.models.extra.article_outline import ArticleOutline
 from fabricatio.models.task import Task
 from fabricatio.utils import ok
@@ -18,14 +18,13 @@ from typer import Typer
 Role(
     name="Undergraduate Researcher",
     description="Write an outline for an article in typst format.",
-    llm_model="openai/qwen-plus",
-    llm_temperature=0.35,
+    llm_model="openai/deepseek-v3-250324",
+    llm_temperature=0.45,
     # llm_api_endpoint=HttpUrl("https://dashscope.aliyuncs.com/compatible-mode/v1"),
     llm_top_p=0.95,
     llm_max_tokens=8191,
-    llm_stream=True,
     llm_rpm=600,
-    llm_tpm=1000000,
+    llm_tpm=900000,
     llm_timeout=600,
     registry={
         Event.quick_instantiate(ns := "article"): WorkFlow(
@@ -41,13 +40,13 @@ Role(
                         ref_limit=18,
                         threshold=0.58,
                         result_per_query=2,
-                        llm_model="openai/qwq-plus",
-                        extractor_model="openai/qwen-max",
-                        query_model="openai/qwen-turbo",
+                        extractor_model={"model": "openai/qwen-max"},
+                        query_model={"model": "openai/qwen-turbo", "temperature": 0.3, "top_p": 0.85},
                     )
                 ),
-                DumpFinalizedOutput(output_key="task_output"),
                 PersistentAll,
+                DumpFinalizedOutput(dump_path="median.typ"),
+                RenderedDump(template_name="article").to_task_output(),
             ),
         ),
         Event.quick_instantiate(ns2 := "complete"): WorkFlow(
@@ -57,8 +56,9 @@ Role(
                 ExtractOutlineFromRaw(output_key="article_outline"),
                 PersistentAll,
                 a,
-                DumpFinalizedOutput(output_key="task_output"),
                 PersistentAll,
+                DumpFinalizedOutput(dump_path="median.typ"),
+                RenderedDump(template_name="article").to_task_output(),
             ),
         ),
         Event.quick_instantiate(ns3 := "finish"): WorkFlow(
@@ -66,14 +66,15 @@ Role(
             description="Finish an article with given article outline. dump the outline to the given path. in typst format.",
             steps=(
                 a,
-                DumpFinalizedOutput(output_key="task_output"),
                 PersistentAll,
+                DumpFinalizedOutput(dump_path="median.typ"),
+                RenderedDump(template_name="article").to_task_output(),
             ),
         ),
         Event.quick_instantiate(ns4 := "consult"): WorkFlow(
             name="Consult Article",
             description="Consult an article with given article outline. dump the outline to the given path. in typst format.",
-            steps=(ArticleConsultRAG().to_task_output(),),
+            steps=(ArticleConsultRAG(ref_q_model={"model":"openai/qwen-turbo"}).to_task_output(),),
         ),
     },
 )

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/examples/write_outline/write_outline.py

@@ -3,6 +3,8 @@
 import asyncio
 
 from fabricatio import Event, Role, logger
+from fabricatio.models.task import Task
+from fabricatio.utils import ok
 from fabricatio.workflows.articles import WriteOutlineCorrectedWorkFlow
 
 
@@ -14,10 +16,11 @@ async def main() -> None:
         registry={Event.quick_instantiate(ns := "article"): WriteOutlineCorrectedWorkFlow},
     )
 
-    proposed_task = await role.propose_task(
+    proposed_task = await role.propose(
+        Task,
         "You need to read the `./article_briefing.txt` file and write an outline for the article in typst format. The outline should be saved in the `./out.typ` file.",
     )
-    path = await proposed_task.delegate(ns)
+    path = await ok(proposed_task).delegate(ns)
     logger.success(f"The outline is saved in:\n{path}")
 
 

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fabricatio"
-version = "0.2.12-dev1"
+version = "0.2.12-dev3"
 description = "A LLM multi-agent framework."
 readme = "README.md"
 license = { file = "LICENSE" }

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/python/fabricatio/actions/article_rag.py

@@ -4,9 +4,10 @@ from asyncio import gather
 from pathlib import Path
 from typing import List, Optional
 
-from pydantic import PositiveInt
+from pydantic import Field, PositiveInt
 
 from fabricatio import BibManager
+from fabricatio.capabilities.advanced_rag import AdvancedRAG
 from fabricatio.capabilities.censor import Censor
 from fabricatio.capabilities.extract import Extract
 from fabricatio.capabilities.rag import RAG
@@ -18,6 +19,7 @@ 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.models.kwargs_types import ChooseKwargs, LLMKwargs
 from fabricatio.utils import ask_retain, ok
 
 TYPST_CITE_USAGE = (
@@ -27,17 +29,17 @@ TYPST_CITE_USAGE = (
     "Illegal citing syntax examples(seperated by |): [[1],[2],[3]]|[[1],[1-2]]\n"
     "Those reference mark shall not be omitted during the extraction\n"
     "It's recommended to cite multiple references that supports your conclusion at a time.\n"
-    "Wrapp inline expression using $ $,like '$>5m$' '$89%$' , and wrapp block equation using $$ $$. if you are using '$' as the money unit, you should add a '\\' before it to avoid being interpreted as a inline equation. For example 'The pants worths 5\\$.'\n"
+    "Wrap inline expression with '\\(' and '\\)',like '\\(>5m\\)' '\\(89%\\)', and wrap block equation with '\\[' and '\\]'.\n"
     "In addition to that, you can add a label outside the block equation which can be used as a cross reference identifier, the label is a string wrapped in `<` and `>` like `<energy-release-rate-equation>`.Note that the label string should be a summarizing title for the equation being labeled.\n"
     "you can refer to that label by using the syntax with prefix of `@eqt:`, which indicate that this notation is citing a label from the equations. For example ' @eqt:energy-release-rate-equation ' DO remember that the notation shall have both suffixed and prefixed space char which enable the compiler to distinguish the notation from the plaintext."
-    "Below is a usage example:\n"
+    "Below is two usage example:\n"
     "```typst\n"
     "See @eqt:mass-energy-equation , it's the foundation of physics.\n"
-    "$$\n"
+    "\\[\n"
     "E = m c^2\n"
-    "$$  <mass-energy-equation>\n\n\n"
-    "In @eqt:mass-energy-equation , $m$ stands for mass, $c$ stands for speed of light, and $E$ stands for energy. \n"
-    "```"
+    "\\]  <mass-energy-equation>\n\n\n"
+    "In @eqt:mass-energy-equation , \\(m\\) stands for mass, \\(c\\) stands for speed of light, and \\(E\\) stands for energy. \n"
+    "```\n"
 )
 
 
@@ -50,9 +52,9 @@ class WriteArticleContentRAG(Action, RAG, Extract):
     """The limit of references to be retrieved"""
     threshold: float = 0.62
     """The threshold of relevance"""
-    extractor_model: str
+    extractor_model: LLMKwargs
     """The model to use for extracting the content from the retrieved references."""
-    query_model: str
+    query_model: LLMKwargs
     """The model to use for querying the database"""
     supervisor: bool = False
     """Whether to use supervisor mode"""
@@ -154,12 +156,12 @@ class WriteArticleContentRAG(Action, RAG, Extract):
                 f"Above is the subsection titled `{subsec.title}`.\n"
                 f"I need you to extract the content to update my subsection obj provided below.\n{self.req}"
                 f"{subsec.display()}\n",
-                model=self.extractor_model,
+                **self.extractor_model,
             ),
             "Failed to propose new subsection.",
         )
         for p in new_subsec.paragraphs:
-            p.content = cm.apply(p.content).replace("$$", "\n$$\n")
+            p.content = cm.apply(p.content)
         subsec.update_from(new_subsec)
         logger.debug(f"{subsec.title}:rpl\n{subsec.display()}")
         return subsec
@@ -175,28 +177,16 @@ class WriteArticleContentRAG(Action, RAG, Extract):
         extra_instruction: str = "",
     ) -> str:
         """Write the raw paragraphs of the subsec."""
-        return (
-            (
-                await self.aask(
-                    f"{cm.as_prompt()}\nAbove is some related reference from other auther retrieved for you."
-                    f"{article_outline.finalized_dump()}\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"{self.req}\n"
-                    f"You SHALL use `{article.language}` as writing language.\n{extra_instruction}\n"
-                    f"Do not use numbered list to display the outcome, you should regard you are writing the main text of the thesis.\n"
-                    f"You should not copy others' works from the references directly on to my thesis, we can only harness the conclusion they have drawn.\n"
-                    f"No extra explanation is allowed."
-                )
-            )
-            .replace(r" \( ", "$")
-            .replace(r" \) ", "$")
-            .replace(r"\(", "$")
-            .replace(r"\)", "$")
-            .replace("\\[\n", "$$\n")
-            .replace("\\[ ", "$$\n")
-            .replace("\n\\]", "\n$$")
-            .replace(" \\]", "\n$$")
+        return await self.aask(
+            f"{cm.as_prompt()}\nAbove is some related reference from other auther retrieved for you."
+            f"{article_outline.finalized_dump()}\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"{self.req}\n"
+            f"You SHALL use `{article.language}` as writing language.\n{extra_instruction}\n"
+            f"Do not use numbered list to display the outcome, you should regard you are writing the main text of the thesis.\n"
+            f"You should not copy others' works from the references directly on to my thesis, we can only harness the conclusion they have drawn.\n"
+            f"No extra explanation is allowed."
         )
 
     async def search_database(
@@ -222,7 +212,7 @@ class WriteArticleContentRAG(Action, RAG, Extract):
         ref_q = ok(
             await self.arefined_query(
                 search_req,
-                model=self.query_model,
+                **self.query_model,
             ),
             "Failed to refine query.",
         )
@@ -232,7 +222,7 @@ class WriteArticleContentRAG(Action, RAG, Extract):
         ret = await self.aretrieve(
             ref_q,
             ArticleChunk,
-            final_limit=self.ref_limit,
+            max_accepted=self.ref_limit,
             result_per_query=self.result_per_query,
             similarity_threshold=self.threshold,
         )
@@ -240,7 +230,7 @@ class WriteArticleContentRAG(Action, RAG, Extract):
         cm.add_chunks(ok(ret))
         ref_q = await self.arefined_query(
             f"{cm.as_prompt()}\n\nAbove is the retrieved references in the first RAG, now we need to perform the second RAG.\n\n{search_req}",
-            model=self.query_model,
+            **self.query_model,
         )
 
         if ref_q is None:
@@ -252,7 +242,7 @@ class WriteArticleContentRAG(Action, RAG, Extract):
         ret = await self.aretrieve(
             ref_q,
             ArticleChunk,
-            final_limit=int(self.ref_limit * self.search_increment_multiplier),
+            max_accepted=int(self.ref_limit * self.search_increment_multiplier),
             result_per_query=int(self.result_per_query * self.search_increment_multiplier),
             similarity_threshold=self.threshold,
         )
@@ -262,18 +252,19 @@ class WriteArticleContentRAG(Action, RAG, Extract):
         cm.add_chunks(ret)
 
 
-class ArticleConsultRAG(Action, RAG):
+class ArticleConsultRAG(Action, AdvancedRAG):
     """Write an article based on the provided outline."""
 
     output_key: str = "consult_count"
-
+    search_increment_multiplier: float = 1.6
+    """The multiplier to increase the limit of references to retrieve per query."""
     ref_limit: int = 20
     """The final limit of references."""
     ref_per_q: int = 3
     """The limit of references to retrieve per query."""
     similarity_threshold: float = 0.62
     """The similarity threshold of references to retrieve."""
-    ref_q_model: Optional[str] = None
+    ref_q_model: ChooseKwargs = Field(default_factory=ChooseKwargs)
     """The model to use for refining query."""
     req: str = TYPST_CITE_USAGE
     """The request for the rag model."""
@@ -285,7 +276,7 @@ class ArticleConsultRAG(Action, RAG):
         from questionary import confirm, text
         from rich import print as r_print
 
-        from fabricatio.rust import convert_all_block_tex, convert_all_inline_tex
+        from fabricatio.rust import convert_all_block_tex, convert_all_inline_tex, fix_misplaced_labels
 
         self.target_collection = collection_name or self.safe_target_collection
 
@@ -295,16 +286,19 @@ class ArticleConsultRAG(Action, RAG):
         while (req := await text("User: ").ask_async()) is not None:
             if await confirm("Empty the cm?").ask_async():
                 cm.empty()
-            ref_q = await self.arefined_query(req, model=self.ref_q_model)
-            refs = await self.aretrieve(
-                ok(ref_q, "Failed to refine query."),
-                ArticleChunk,
-                final_limit=self.ref_limit,
+            await self.clued_search(
+                req,
+                cm,
+                refinery_kwargs=self.ref_q_model,
+                expand_multiplier=self.search_increment_multiplier,
+                base_accepted=self.ref_limit,
                 result_per_query=self.ref_per_q,
                 similarity_threshold=self.similarity_threshold,
             )
 
-            ret = await self.aask(f"{cm.add_chunks(refs).as_prompt()}\n{self.req}\n{req}")
+            ret = await self.aask(f"{cm.as_prompt()}\n{self.req}\n{req}")
+
+            ret = fix_misplaced_labels(ret)
             ret = convert_all_inline_tex(ret)
             ret = convert_all_block_tex(ret)
             ret = cm.apply(ret)
@@ -396,7 +390,7 @@ class TweakArticleRAG(Action, RAG, Censor):
         await self.censor_obj_inplace(
             subsec,
             ruleset=ruleset,
-            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"
+            reference=f"{'\n\n'.join(d.display() for d in await self.aretrieve(refind_q, document_model=ArticleEssence, max_accepted=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}`. "

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/python/fabricatio/actions/output.py

@@ -1,9 +1,10 @@
 """Dump the finalized output to a file."""
 
 from pathlib import Path
-from typing import Any, Dict, Iterable, List, Mapping, Optional, Type
+from typing import Any, Iterable, List, Mapping, Optional, Type
 
 from fabricatio import TEMPLATE_MANAGER
+from fabricatio.fs import dump_text
 from fabricatio.journal import logger
 from fabricatio.models.action import Action
 from fabricatio.models.generic import FinalizedDumpAble, FromMapping, PersistentAble
@@ -51,7 +52,7 @@ class RenderedDump(Action, LLMUsage):
 
     async def _execute(
         self,
-        to_dump: Dict[str, Any],
+        to_dump: FinalizedDumpAble,
         task_input: Optional[Task] = None,
         dump_path: Optional[str | Path] = None,
         **_,
@@ -66,8 +67,16 @@ class RenderedDump(Action, LLMUsage):
                 "Could not find the path of file to dump the data.",
             )
         )
+
         logger.info(f"Saving output to {dump_path.as_posix()}")
-        return TEMPLATE_MANAGER.render_template(self.template_name, to_dump)
+        dump_text(
+            dump_path,
+            TEMPLATE_MANAGER.render_template(
+                self.template_name, {to_dump.__class__.__name__: to_dump.finalized_dump()}
+            ),
+        )
+
+        return dump_path.as_posix()
 
 
 class PersistentAll(Action, LLMUsage):

fabricatio-0.2.12.dev3/python/fabricatio/capabilities/advanced_rag.py

@@ -0,0 +1,56 @@
+"""Advanced RAG (Retrieval Augmented Generation) model."""
+
+from typing import Optional, Unpack
+
+from fabricatio.capabilities.rag import RAG
+from fabricatio.journal import logger
+from fabricatio.models.adv_kwargs_types import FetchKwargs
+from fabricatio.models.extra.aricle_rag import ArticleChunk, CitationManager
+from fabricatio.models.kwargs_types import ChooseKwargs
+
+
+class AdvancedRAG(RAG):
+    """A class representing the Advanced RAG (Retrieval Augmented Generation) model."""
+
+    async def clued_search(
+        self,
+        requirement: str,
+        cm: CitationManager,
+        max_capacity: int = 40,
+        max_round: int = 3,
+        expand_multiplier: float = 1.4,
+        base_accepted: int = 12,
+        refinery_kwargs: Optional[ChooseKwargs] = None,
+        **kwargs: Unpack[FetchKwargs],
+    ) -> CitationManager:
+        """Asynchronously performs a clued search based on a given requirement and citation manager."""
+        if max_round<=0:
+            raise ValueError("max_round should be greater than 0")
+        if max_round == 1:
+            logger.warning(
+                "max_round should be greater than 1, otherwise it behaves nothing different from the `self.aretrieve`"
+            )
+
+        refinery_kwargs = refinery_kwargs or {}
+
+        for i in range(max_round + 1, 1):
+            logger.info(f"Round [{i + 1}/{max_round}] search started.")
+            ref_q = await self.arefined_query(
+                f"{cm.as_prompt()}\n\nAbove is the retrieved references in the {i - 1}th RAG, now we need to perform the {i}th RAG."
+                f"\n\n{requirement}",
+                **refinery_kwargs,
+            )
+            if ref_q is None:
+                logger.error(f"At round [{i + 1}/{max_round}] search, failed to refine the query, exit.")
+                return cm
+            refs = await self.aretrieve(ref_q, ArticleChunk, base_accepted, **kwargs)
+
+            if (max_capacity := max_capacity - len(refs)) < 0:
+                cm.add_chunks(refs[0:max_capacity])
+                logger.debug(f"At round [{i + 1}/{max_round}] search, the capacity is not enough, exit.")
+                return cm
+
+            cm.add_chunks(refs)
+            base_accepted = int(base_accepted * expand_multiplier)
+        logger.debug(f"Exceeded max_round: {max_round}, exit.")
+        return cm

{fabricatio-0.2.12.dev1 → fabricatio-0.2.12.dev3}/python/fabricatio/capabilities/rag.py

@@ -189,7 +189,7 @@ class RAG(EmbeddingUsage):
         self,
         query: List[str] | str,
         document_model: Type[D],
-        final_limit: int = 20,
+        max_accepted: int = 20,
         **kwargs: Unpack[FetchKwargs],
     ) -> List[D]:
         """Retrieve data from the collection.
@@ -197,7 +197,7 @@ class RAG(EmbeddingUsage):
         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.
+            max_accepted (int): The final limit on the number of results to return.
             **kwargs (Unpack[FetchKwargs]): Additional keyword arguments for retrieval.
 
         Returns:
@@ -211,9 +211,9 @@ class RAG(EmbeddingUsage):
                 document_model=document_model,
                 **kwargs,
             )
-        )[:final_limit]
+        )[:max_accepted]
 
-    async def arefined_query(self, question: List[str] | str, **kwargs: Unpack[ChooseKwargs]) -> Optional[List[str]]:
+    async def arefined_query(self, question: List[str] | str, **kwargs: Unpack[ChooseKwargs[Optional[List[str]]]]) -> Optional[List[str]]:
         """Refines the given question using a template.
 
         Args: