flowtask 5.8.4__cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl

flowtask/components/LangchainLoader/loader.py

@@ -0,0 +1,194 @@
+import asyncio
+from typing import List
+from collections.abc import Callable
+import importlib
+from pathlib import Path, PurePath
+from parrot.loaders import AbstractLoader, Document
+from parrot.llms.vertex import VertexLLM
+from ..flow import FlowComponent
+from ...exceptions import ConfigError, ComponentError
+from ...conf import (
+    DEFAULT_LLM_MODEL,
+    DEFAULT_LLM_TEMPERATURE
+)
+
+class LangchainLoader(FlowComponent):
+    """LangchainLoader.
+
+    Overview:
+
+    Getting a list of documents and convert into Langchain Documents.
+
+
+        Example:
+
+        ```yaml
+        LangchainLoader:
+          path: /home/ubuntu/symbits/lg/bot/products_positive
+          source_type: Product-Top-Reviews
+          loader: HTMLLoader
+          chunk_size: 2048
+          elements:
+          - div: .product
+        ```
+
+    """
+
+    def __init__(
+        self,
+        loop: asyncio.AbstractEventLoop = None,
+        job: Callable = None,
+        stat: Callable = None,
+        **kwargs,
+    ):
+        self.extensions: list = kwargs.pop('extensions', [])
+        self.encoding: str = kwargs.get('encoding', 'utf-8')
+        self.path: str = kwargs.pop('path', None)
+        self.skip_directories: List[str] = kwargs.pop('skip_directories', [])
+        self._chunk_size = kwargs.get('chunk_size', 2048)
+        self._embed_size: int = kwargs.pop('embed_size', 768)
+        self.source_type: str = kwargs.pop('source_type', 'document')
+        self.doctype: str = kwargs.pop('doctype', 'document')
+        # LLM (if required)
+        self._llm = kwargs.pop('llm', None)
+        super().__init__(
+            loop=loop, job=job, stat=stat, **kwargs
+        )
+        self._device: str = kwargs.get('device', 'cpu')
+        self._cuda_number: int = kwargs.get('cuda_device', 0)
+        # Use caching to avoid instanciate several times same loader
+        self._caching_loaders: dict = {}
+
+    async def close(self):
+        # Destroy effectively all Models.
+        pass
+
+    def get_default_llm(self):
+        """Return a VertexLLM instance."""
+        return VertexLLM(
+            model=DEFAULT_LLM_MODEL,
+            temperature=DEFAULT_LLM_TEMPERATURE,
+            top_k=30,
+            top_p=0.5,
+        )
+
+    async def start(self, **kwargs):
+        await super().start(**kwargs)
+        if self.path:
+            if isinstance(self.path, str):
+                self.path = self.mask_replacement_recursively(self.path)
+                self.path = Path(self.path).resolve()
+                if not self.path.exists():
+                    raise ComponentError(
+                        f"Langchain: {self.path} doesn't exists."
+                    )
+        else:
+            raise ConfigError(
+                "Provide at least one directory or filename in *path* attribute."
+            )
+
+    def _get_loader(self, suffix: str, **kwargs):
+        """
+        Get a Document Loader based on Prefix.
+        TODO: a more automated way using importlib.
+        """
+        # Common Arguments
+        args = {
+            "markdown_splitter": self._md_splitter,
+            "summarization_model": self.summarization_model,
+            "device": self._device,
+            "cuda_number": self._cuda_number,
+            "source_type": self.source_type,
+            "encoding": self.encoding,
+            "llm": self._llm
+        }
+
+    def _load_loader(self, name: str, **kwargs):
+        """Dynamically imports a loader class from the loaders module.
+
+        Args:
+            loader_name: The name of the loader class to import (e.g., 'QALoader').
+
+        Returns:
+            The imported loader class.
+        """
+        try:
+            module_path = ".loaders"
+            module = importlib.import_module(module_path, package=__package__)
+            cls = getattr(module, name)
+            if cls:
+                args = {
+                    "markdown_splitter": self._md_splitter,
+                    "summarization_model": self.summarization_model,
+                    "device": self._device,
+                    "cuda_number": self._cuda_number,
+                    "source_type": self.source_type,
+                    "encoding": self.encoding,
+                    "llm": self._llm,
+                    **kwargs
+                }
+                loader = cls(**args)
+                self._caching_loaders[name] = loader
+            return loader
+        except (ModuleNotFoundError, AttributeError) as e:
+            raise ImportError(
+                f"Unable to load the loader '{name}': {e}"
+            ) from e
+
+    async def _load_document(self, path: PurePath) -> List[Document]:
+        documents = []
+        suffix = path.suffix
+        if suffix in self._caching_loaders:
+            loader = self._caching_loaders[suffix]
+        else:
+            loader = self._get_loader(suffix)
+            self._caching_loaders[suffix] = loader
+        async with loader as ld:
+            documents = await ld.load(path)
+        # split or not split?
+        return documents
+
+    async def run(self):
+        documents = []
+        if hasattr(self, 'loader'):
+            print('PARAMS >> ', self._attrs)
+            loader = self._load_loader(self.loader, **self._attrs)
+            async with loader as ld:
+                ext = loader.supported_extensions()
+                if self.path.is_dir():
+                    if self.extensions:
+                        # iterate over the files in the directory
+                        for ext in self.extensions:
+                            for item in self.path.glob(f'*{ext}'):
+                                if item.is_file() and set(item.parts).isdisjoint(self.skip_directories):
+                                    documents.extend(await ld.load(item))
+                    else:
+                        for item in self.path.glob('*.*'):
+                            if item.is_file() and set(item.parts).isdisjoint(self.skip_directories):
+                                documents.extend(await ld.load(item))
+                else:
+                    documents = await ld.load(self.path)
+        else:
+            if self.path.is_dir():
+                # iterate over the files in the directory
+                if self.extensions:
+                    for ext in self.extensions:
+                        for item in self.path.glob(f'*{ext}'):
+                            if item.is_file() and set(item.parts).isdisjoint(self.skip_directories):
+                                documents.extend(await self._load_document(item))
+                else:
+                    for item in self.path.glob('*.*'):
+                        if item.is_file() and set(item.parts).isdisjoint(self.skip_directories):
+                            documents.extend(await self._load_document(item))
+            elif self.path.is_file():
+                if self.path.suffix in self.extensions:
+                    if set(self.path.parts).isdisjoint(self.skip_directories):
+                        documents = await self._load_document(self.path)
+            else:
+                raise ValueError(
+                    f"Langchain Loader: Invalid path: {self.path}"
+                )
+        self._result = documents
+        self.add_metric('NUM_DOCUMENTS', len(documents))
+        # return self._result
+        return True

flowtask/components/LangchainLoader/loaders/__init__.py

@@ -0,0 +1,22 @@
+# """
+# Langchain Loaders.
+
+# Basic Documents Loaders, adapted to be used in Flowtask Tasks.
+# """
+# from .docx import MSWordLoader
+# from .qa import QAFileLoader
+# from .pdfmark import PDFMarkdown
+# from .pdftables import PDFTables
+# from .pdfblocks import PDFBlocks
+# from .txt import TXTLoader
+# from .html import HTMLLoader
+
+# __all__ = (
+#     "MSWordLoader",
+#     "QAFileLoader",
+#     "PDFMarkdown",
+#     "PDFTables",
+#     "TXTLoader",
+#     "PDFBlocks",
+#     "HTMLLoader",
+# )

flowtask/components/LangchainLoader/loaders/abstract.py

@@ -0,0 +1,362 @@
+from abc import ABC, abstractmethod
+from typing import Union, List, Optional
+from collections.abc import Callable
+from datetime import datetime
+from pathlib import Path, PurePath
+import torch
+from langchain.docstore.document import Document
+from langchain.chains.summarize import load_summarize_chain
+from langchain.text_splitter import (
+    TokenTextSplitter
+)
+from langchain_huggingface import HuggingFacePipeline
+from transformers import (
+    AutoModelForSeq2SeqLM,
+    AutoTokenizer,
+    pipeline
+)
+from langchain_core.prompts import PromptTemplate
+from navconfig.logging import logging
+from navigator.libs.json import JSONContent  # pylint: disable=E0611
+from parrot.llms.vertex import VertexLLM
+from ....conf import (
+    EMBEDDING_DEVICE,
+    DEFAULT_LLM_MODEL,
+    DEFAULT_LLM_TEMPERATURE,
+)
+
+class AbstractLoader(ABC):
+    """
+    Abstract class for Document loaders.
+    """
+    _extension: List[str] = []
+
+    def __init__(
+        self,
+        tokenizer: Union[str, Callable] = None,
+        text_splitter: Union[str, Callable] = None,
+        summarizer: Union[str, Callable] = None,
+        markdown_splitter: Union[str, Callable] = None,
+        source_type: str = 'file',
+        doctype: Optional[str] = 'document',
+        device: str = None,
+        cuda_number: int = 0,
+        llm: Callable = None,
+        **kwargs
+    ):
+        self.tokenizer = tokenizer
+        self._summary_model = summarizer
+        self.text_splitter = text_splitter
+        self.markdown_splitter = markdown_splitter
+        self.doctype = doctype
+        self.logger = logging.getLogger(
+            f"Loader.{self.__class__.__name__}"
+        )
+        self.path = kwargs.pop('path', None)
+        self._source_type = source_type
+        # LLM (if required)
+        self._llm = llm
+        # JSON encoder:
+        self._encoder = JSONContent()
+        self.device_name = device
+        self.cuda_number = cuda_number
+        self._device = None
+        self.encoding: str = kwargs.get('encoding', 'utf-8')
+        self.summarization_model = kwargs.get(
+            'summarization_model',
+            "facebook/bart-large-cnn"
+        )
+        self._no_summarization = kwargs.get('no_summarization', False)
+        self._translation = kwargs.get('translation', False)
+        self.category: str = kwargs.get('category', 'document')
+
+    async def __aenter__(self):
+        # Cuda Device:
+        self._device = self._get_device(
+            self.device_name,
+            self.cuda_number
+        )
+        return self
+
+    def supported_extensions(self):
+        return self._extension
+
+    async def __aexit__(self, *exc_info):
+        self.post_load()
+
+    def post_load(self):
+        self.tokenizer = None  # Reset the tokenizer
+        self.text_splitter = None  # Reset the text splitter
+        torch.cuda.synchronize()  # Wait for all kernels to finish
+        torch.cuda.empty_cache()  # Clear unused memory
+
+    def _get_device(self, device_type: str = None, cuda_number: int = 0):
+        """Get Default device for Torch and transformers.
+
+        """
+        if device_type == 'cpu':
+            return torch.device('cpu')
+        elif device_type == 'cuda':
+            return torch.device(f'cuda:{cuda_number}')
+        else:
+            if torch.cuda.is_available():
+                # Use CUDA GPU if available
+                return torch.device(f'cuda:{cuda_number}')
+            if torch.backends.mps.is_available():
+                # Use CUDA Multi-Processing Service if available
+                return torch.device("mps")
+            if EMBEDDING_DEVICE == 'cuda':
+                return torch.device(f'cuda:{cuda_number}')
+            else:
+                return torch.device(EMBEDDING_DEVICE)
+
+    def _check_path(
+        self,
+        path: PurePath,
+        suffix: Optional[List[str]] = None
+    ) -> bool:
+        """Check if the file path exists.
+        Args:
+            path (PurePath): The path to the file.
+        Returns:
+            bool: True if the file exists, False otherwise.
+        """
+        if isinstance(path, str):
+            path = Path(path).resolve()
+        if not suffix:
+            suffix = self._extension
+        return path.exists() and path.is_file() and path.suffix in suffix
+
+    def create_metadata(
+        self,
+        path: Union[str, PurePath],
+        doctype: str = 'document',
+        source_type: str = 'source',
+        doc_metadata: Optional[dict] = None,
+        summary: Optional[str] = '',
+        **kwargs
+    ):
+        if not doc_metadata:
+            doc_metadata = {}
+        if isinstance(path, PurePath):
+            origin = path.name
+            url = f'file://{path.name}'
+            filename = path
+        else:
+            origin = path
+            url = path
+            filename = f'file://{path}'
+        metadata = {
+            "url": url,
+            "source": origin,
+            "filename": str(filename),
+            "type": doctype,
+            "question": '',
+            "answer": '',
+            "summary": summary,
+            "source_type": source_type,
+            "created_at": datetime.now().strftime("%Y-%m-%d, %H:%M:%S"),
+            "category": self.category,
+            "document_meta": {
+                **doc_metadata
+            },
+            **kwargs
+        }
+        return metadata
+
+    def get_default_llm(self):
+        """Return a VertexLLM instance."""
+        return VertexLLM(
+            model=DEFAULT_LLM_MODEL,
+            temperature=DEFAULT_LLM_TEMPERATURE,
+            top_k=30,
+            top_p=0.5,
+        )
+
+    def get_summary_from_text(self, text: str, use_gpu: bool = False) -> str:
+        """
+        Get a summary of a text.
+        """
+        if not text:
+            # NO data to be summarized
+            return ''
+        # splitter = TokenTextSplitter(
+        #     chunk_size=2048,
+        #     chunk_overlap=100,
+        # )
+        prompt_template = """Write a summary of the following, please also identify the main theme:
+        {text}
+        SUMMARY:"""
+        prompt = PromptTemplate.from_template(prompt_template)
+        refine_template = (
+            "Your job is to produce a final summary\n"
+            "We have provided an existing summary up to a certain point: {existing_answer}\n"
+            "We have the opportunity to refine the existing summary"
+            "(only if needed) with some more context below.\n"
+            "------------\n"
+            "{text}\n"
+            "------------\n"
+            "Given the new context, refine the original summary adding more explanation."
+            "If the context isn't useful, return the original summary."
+        )
+        refine_prompt = PromptTemplate.from_template(refine_template)
+        # if self._llm:
+        #     llm = self._llm
+        # else:
+        #     llm = self.get_summarization_model(
+        #         self.summarization_model,
+        #         use_gpu=use_gpu
+        #     )
+        # if not llm:
+        #     return ''
+        llm = self.get_default_llm()
+        llm = llm.get_llm()
+        summarize_chain = load_summarize_chain(
+            llm=llm,
+            chain_type="refine",
+            question_prompt=prompt,
+            refine_prompt=refine_prompt,
+            return_intermediate_steps=False,
+            input_key="input_documents",
+            output_key="output_text",
+        )
+        doc = Document(page_content=text)
+        try:
+            summary = summarize_chain.invoke(
+                {"input_documents": [doc]}, return_only_outputs=True
+            )
+            return summary.get('output_text', '')
+        except Exception as e:
+            print('ERROR in get_summary_from_text:', e)
+            return ""
+
+    def get_translator(self, model_name: str = 'Helsinki-NLP/opus-mt-en-es'):
+        if not self._translation:
+            return None
+        trans_model = AutoModelForSeq2SeqLM.from_pretrained(
+            model_name,
+            # device_map="auto",
+            # torch_dtype=torch.bfloat16,
+            trust_remote_code=True
+        )
+        trans_tokenizer = AutoTokenizer.from_pretrained(model_name)
+        translator = pipeline(
+            "translation",
+            model=trans_model,
+            tokenizer=trans_tokenizer,
+            batch_size=True,
+            max_new_tokens=500,
+            min_new_tokens=300,
+            use_fast=True
+        )
+        return translator
+
+    def get_summarization_model(
+        self,
+        model_name: str = 'facebook/bart-large-cnn',
+        use_gpu: bool = False
+    ):
+        if self._no_summarization is True:
+            return None
+        if not self._summary_model:
+            summarize_model = AutoModelForSeq2SeqLM.from_pretrained(
+                model_name,
+                # torch_dtype=torch.float32,
+                torch_dtype=torch.bfloat16,
+                trust_remote_code=True
+            )
+            if use_gpu:
+                # summarize_model.to(0)
+                summarize_model.cuda()
+            summarize_tokenizer = AutoTokenizer.from_pretrained(
+                model_name,
+                padding_side="left"
+            )
+            pipe_summary = pipeline(
+                "summarization",
+                model=summarize_model,
+                tokenizer=summarize_tokenizer,
+                # device='cuda:0',
+                # batch_size=True,
+                max_new_tokens=256,
+                # min_new_tokens=300,
+                use_fast=True
+            )
+            self._summary_model = HuggingFacePipeline(
+                model_id=model_name,
+                pipeline=pipe_summary,
+                verbose=True
+            )
+        return self._summary_model
+
+    def resolve_paths(self, path: Union[str, PurePath, List[PurePath]]) -> List[Path]:
+        """
+        Resolve the input path into a list of file paths.
+        Handles lists, directories, glob patterns, and single file paths.
+
+        Args:
+            path (Union[str, PurePath, List[PurePath]]): Input path(s).
+
+        Returns:
+            List[Path]: A list of resolved file paths.
+        """
+        resolved_paths = []
+
+        if isinstance(path, str):
+            if "*" in path:
+                # Glob pattern
+                resolved_paths = list(Path().glob(path))
+            else:
+                # Single path as string
+                resolved_paths = [Path(path)]
+        elif isinstance(path, PurePath):
+            # Single Path
+            resolved_paths = [Path(path)]
+        elif isinstance(path, list):
+            # List of paths
+            resolved_paths = [Path(p) for p in path]
+
+        final_paths = []
+        for p in resolved_paths:
+            if p.is_dir():
+                # Add all matching files in the directory
+                if self._extension:
+                    for ext in self._extension:
+                        final_paths.extend(p.glob(f"*{ext}"))
+                else:
+                    final_paths.extend(p.glob("*"))
+            elif p.is_file():
+                final_paths.append(p)
+
+        return final_paths
+
+    async def load(self, path: Union[str, PurePath, List[PurePath]]) -> List[Document]:
+        """Load data from a source and return it as a Langchain Document.
+
+        Args:
+            path (Union[str, PurePath, List[PurePath]]): The source of the data.
+
+        Returns:
+            List[Document]: A list of Langchain Documents.
+        """
+        self.logger.info(
+            f"Loading file: {path}"
+        )
+        paths = self.resolve_paths(path)
+        docs = []
+        for p in paths:
+            if p.exists():
+                docs.extend(await self._load_document(p))
+        return docs
+
+    async def _load_document(self, path: Path) -> List:
+        """
+        Abstract method for loading a document.
+
+        Args:
+            path (Path): The path to the file.
+
+        Returns:
+            List: A list of Langchain documents.
+        """
+        pass

flowtask/components/LangchainLoader/loaders/basepdf.py

@@ -0,0 +1,50 @@
+from abc import abstractmethod
+from typing import List, Union
+from pathlib import Path, PurePath
+from markdownify import markdownify as md
+from langchain.docstore.document import Document
+from .abstract import AbstractLoader
+
+
+class BasePDF(AbstractLoader):
+    """
+    Base Abstract loader for all PDF-file Loaders.
+    """
+    _extension = ['.pdf']
+    chunk_size = 768
+
+    def __init__(self, **kwargs):
+        self._lang = 'eng'
+        super().__init__(**kwargs)
+
+    @abstractmethod
+    def _load_pdf(self, path: Path) -> list:
+        """
+        Load a PDF file using Fitz.
+
+        Args:
+            path (Path): The path to the PDF file.
+
+        Returns:
+            list: A list of Langchain Documents.
+        """
+        pass
+
+    async def load(self, path: Union[str, PurePath, List[PurePath]]) -> List[Document]:
+        """Load data from a source and return it as a Langchain Document.
+
+        Args:
+            path (Union[str, PurePath, List[PurePath]]): The source of the data.
+
+        Returns:
+            List[Document]: A list of Langchain Documents.
+        """
+        self.logger.info(
+            f"Loading file: {path}"
+        )
+        paths = self.resolve_paths(path)
+        docs = []
+        for p in paths:
+            if p.exists():
+                docs.extend(self._load_pdf(p))
+        return docs