lollms/lollms/utilities.py

######
# Project       : lollms
# File          : utilities.py
# Author        : ParisNeo with the help of the community
# license       : Apache 2.0
# Description   : 
# This file contains utilities functions that can be used by any
# module.
######
from ascii_colors import ASCIIColors, trace_exception
from sklearn.feature_extraction.text import TfidfVectorizer
import numpy as np
from pathlib import Path
import json
import re
import subprocess
import gc

from typing import List

from PIL import Image
import requests
from io import BytesIO
import base64


def load_image(image_file):
    s_image_file = str(image_file)
    if s_image_file.startswith('http://') or s_image_file.startswith('https://'):
        response = requests.get(s_image_file)
        image = Image.open(BytesIO(response.content)).convert('RGB')
    else:
        image = Image.open(s_image_file).convert('RGB')
    return image

def load_image_from_base64(image):
    return Image.open(BytesIO(base64.b64decode(image)))


def expand2square(pil_img, background_color):
    width, height = pil_img.size
    if width == height:
        return pil_img
    elif width > height:
        result = Image.new(pil_img.mode, (width, width), background_color)
        result.paste(pil_img, (0, (width - height) // 2))
        return result
    else:
        result = Image.new(pil_img.mode, (height, height), background_color)
        result.paste(pil_img, ((height - width) // 2, 0))
        return result




def find_first_available_file_index(folder_path, prefix, extension=""):
    """
    Finds the first available file index in a folder with files that have a prefix and an optional extension.
    
    Args:
        folder_path (str): The path to the folder.
        prefix (str): The file prefix.
        extension (str, optional): The file extension (including the dot). Defaults to "".
    
    Returns:
        int: The first available file index.
    """
    # Create a Path object for the folder
    folder = Path(folder_path)
    
    # Get a list of all files in the folder
    files = folder.glob(f'{prefix}*'+extension)
    
    # Initialize the first available number
    available_number = 1
    
    # Iterate through the files
    for file in files:
        # Extract the number from the file name
        file_number = int(file.stem[len(prefix):])
        
        # If the file number is equal to the available number, increment the available number
        if file_number == available_number:
            available_number += 1
        # If the file number is greater than the available number, break the loop
        elif file_number > available_number:
            break
    
    return available_number




# Prompting tools
def detect_antiprompt(text:str, anti_prompts=["!@>"]) -> bool:
    """
    Detects if any of the antiprompts in self.anti_prompts are present in the given text.
    Used for the Hallucination suppression system

    Args:
        text (str): The text to check for antiprompts.

    Returns:
        bool: True if any antiprompt is found in the text (ignoring case), False otherwise.
    """
    for prompt in anti_prompts:
        if prompt.lower() in text.lower():
            return prompt.lower()
    return None


def remove_text_from_string(string, text_to_find):
    """
    Removes everything from the first occurrence of the specified text in the string (case-insensitive).

    Parameters:
    string (str): The original string.
    text_to_find (str): The text to find in the string.

    Returns:
    str: The updated string.
    """
    index = string.lower().find(text_to_find.lower())

    if index != -1:
        string = string[:index]

    return string


# Pytorch and cuda tools
def check_torch_version(min_version, min_cuda_versio=12):
    import torch

    if "+" in torch.__version__ and int(torch.__version__.split("+")[-1][2:4])<min_cuda_versio:
        return False

    # Extract torch version from __version__ attribute with regular expression
    current_version_float = float('.'.join(torch.__version__.split(".")[:2]))
    # Check if the current version meets or exceeds the minimum required version
    return current_version_float >= min_version


def reinstall_pytorch_with_cuda():
    try:
        ASCIIColors.info("Installing cuda 12.1.1")
        result = subprocess.run(["conda", "install", "-c", "nvidia/label/cuda-12.1.1", "cuda-toolkit", "-y"])
    except Exception as ex:
        ASCIIColors.error(ex)
    try:
        ASCIIColors.info("Installing pytorch 2.1.1")
        result = subprocess.run(["pip", "install", "--upgrade", "torch", "torchvision", "torchaudio", "--no-cache-dir", "--index-url", "https://download.pytorch.org/whl/cu121"])
    except Exception as ex:
        ASCIIColors.error(ex)
    if result.returncode != 0:
        ASCIIColors.warning("Couldn't find Cuda build tools on your PC. Reverting to CPU.")
        result = subprocess.run(["pip", "install", "--upgrade", "torch", "torchvision", "torchaudio", "--no-cache-dir"])
        if result.returncode != 0:
            ASCIIColors.error("Couldn't install pytorch !!")
        else:
            ASCIIColors.error("Pytorch installed successfully!!")


def reinstall_pytorch_with_rocm():
    result = subprocess.run(["pip", "install", "--upgrade", "torch", "torchvision", "torchaudio", "--no-cache-dir", "--index-url", "https://download.pytorch.org/whl/rocm5.6"])
    if result.returncode != 0:
        ASCIIColors.warning("Couldn't find Cuda build tools on your PC. Reverting to CPU.")
        result = subprocess.run(["pip", "install", "--upgrade", "torch", "torchvision", "torchaudio", "--no-cache-dir"])
        if result.returncode != 0:
            ASCIIColors.error("Couldn't install pytorch !!")
        else:
            ASCIIColors.error("Pytorch installed successfully!!")
            
            

def reinstall_pytorch_with_cpu():
    result = subprocess.run(["pip", "install", "--upgrade", "torch", "torchvision", "torchaudio", "--no-cache-dir"])
    if result.returncode != 0:
        ASCIIColors.warning("Couldn't find Cuda build tools on your PC. Reverting to CPU.")
        result = subprocess.run(["pip", "install", "--upgrade", "torch", "torchvision", "torchaudio", "--no-cache-dir"])
        if result.returncode != 0:
            ASCIIColors.error("Couldn't install pytorch !!")
        else:
            ASCIIColors.error("Pytorch installed successfully!!")     


def check_and_install_torch(enable_gpu:bool, version:float=2.1):
    if enable_gpu:
        ASCIIColors.yellow("This installation has enabled GPU support. Trying to install with GPU support")
        ASCIIColors.info("Checking pytorch")
        try:
            import torch
            import torchvision
            if torch.cuda.is_available():
                ASCIIColors.success(f"CUDA is supported.\nCurrent version is {torch.__version__}.")
                if not check_torch_version(version):
                    ASCIIColors.yellow("Torch version is old. Installing new version")
                    reinstall_pytorch_with_cuda()
                else:
                    ASCIIColors.yellow("Torch OK")
            else:
                ASCIIColors.warning("CUDA is not supported. Trying to reinstall PyTorch with CUDA support.")
                reinstall_pytorch_with_cuda()
        except Exception as ex:
            ASCIIColors.info("Pytorch not installed. Reinstalling ...")
            reinstall_pytorch_with_cuda()    
    else:
        try:
            import torch
            import torchvision
            if check_torch_version(version):
                ASCIIColors.warning("Torch version is too old. Trying to reinstall PyTorch with CUDA support.")
                reinstall_pytorch_with_cpu()
        except Exception as ex:
            ASCIIColors.info("Pytorch not installed. Reinstalling ...")
            reinstall_pytorch_with_cpu() 
    

class NumpyEncoderDecoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.ndarray):
            return {'__numpy_array__': True, 'data': obj.tolist()}
        return super(NumpyEncoderDecoder, self).default(obj)

    @staticmethod
    def as_numpy_array(dct):
        if '__numpy_array__' in dct:
            return np.array(dct['data'])
        return dct


def git_pull(folder_path):
    try:
        # Change the current working directory to the desired folder
        subprocess.run(["git", "checkout", folder_path], check=True, cwd=folder_path)
        # Run 'git pull' in the specified folder
        subprocess.run(["git", "pull"], check=True, cwd=folder_path)
        print("Git pull successful in", folder_path)
    except subprocess.CalledProcessError as e:
        print("Error occurred while executing Git pull:", e)
        # Handle any specific error handling here if required
class AdvancedGarbageCollector:
    @staticmethod
    def hardCollect(obj):
        """
        Remove a reference to the specified object and attempt to collect it.

        Parameters:
        - obj: The object to be collected.

        This method first identifies all the referrers (objects referencing the 'obj')
        using Python's garbage collector (gc.get_referrers). It then iterates through
        the referrers and attempts to break their reference to 'obj' by setting them
        to None. Finally, it deletes the 'obj' reference.

        Note: This method is designed to handle circular references and can be used
        to forcefully collect objects that might not be collected automatically.

        """
        if obj is None:
            return
        all_referrers = gc.get_referrers(obj)
        for referrer in all_referrers:
            try:
                if isinstance(referrer, (list, tuple, dict, set)):
                    if isinstance(referrer, list):
                        if obj in referrer:
                            referrer.remove(obj)
                    elif isinstance(referrer, dict):
                        new_dict = {}
                        for key, value in referrer.items():
                            if value != obj:
                                new_dict[key] = value
                        referrer.clear()
                        referrer.update(new_dict)
                    elif isinstance(referrer, set):
                        if obj in referrer:
                            referrer.remove(obj)
            except:
                ASCIIColors.warning("Couldn't remove object from referrer")
        del obj

    @staticmethod
    def safeHardCollect(variable_name, instance=None):
        """
        Safely remove a reference to a variable and attempt to collect its object.

        Parameters:
        - variable_name: The name of the variable to be collected.
        - instance: An optional instance (object) to search for the variable if it
          belongs to an object.

        This method provides a way to safely break references to a variable by name.
        It first checks if the variable exists either in the local or global namespace
        or within the provided instance. If found, it calls the 'hardCollect' method
        to remove the reference and attempt to collect the associated object.

        """
        if instance is not None:
            if hasattr(instance, variable_name):
                obj = getattr(instance, variable_name)
                AdvancedGarbageCollector.hardCollect(obj)
            else:
                print(f"The variable '{variable_name}' does not exist in the instance.")
        else:
            if variable_name in locals():
                obj = locals()[variable_name]
                AdvancedGarbageCollector.hardCollect(obj)
            elif variable_name in globals():
                obj = globals()[variable_name]
                AdvancedGarbageCollector.hardCollect(obj)
            else:
                print(f"The variable '{variable_name}' does not exist in the local or global namespace.")

    @staticmethod
    def safeHardCollectMultiple(variable_names, instance=None):
        """
        Safely remove references to multiple variables and attempt to collect their objects.

        Parameters:
        - variable_names: A list of variable names to be collected.
        - instance: An optional instance (object) to search for the variables if they
          belong to an object.

        This method iterates through a list of variable names and calls 'safeHardCollect'
        for each variable, effectively removing references and attempting to collect
        their associated objects.

        """
        for variable_name in variable_names:
            AdvancedGarbageCollector.safeHardCollect(variable_name, instance)

    @staticmethod
    def collect():
        """
        Perform a manual garbage collection using Python's built-in 'gc.collect' method.

        This method triggers a manual garbage collection, attempting to clean up
        any unreferenced objects in memory. It can be used to free up memory and
        resources that are no longer in use.

        """
        gc.collect()


class PackageManager:
    @staticmethod
    def install_package(package_name):
        import subprocess
        import sys
        subprocess.check_call([sys.executable, "-m", "pip", "install", package_name])

class GitManager:
    @staticmethod
    def git_pull(folder_path):
        try:
            # Change the current working directory to the desired folder
            subprocess.run(["git", "checkout", folder_path], check=True, cwd=folder_path)
            # Run 'git pull' in the specified folder
            subprocess.run(["git", "pull"], check=True, cwd=folder_path)
            print("Git pull successful in", folder_path)
        except subprocess.CalledProcessError as e:
            print("Error occurred while executing Git pull:", e)
            # Handle any specific error handling here if required

class File64BitsManager:

    @staticmethod
    def raw_b64_img(image) -> str:
        try:
            from PIL import Image, PngImagePlugin
            import io
            import base64
        except:
            PackageManager.install_package("pillow")
            from PIL import Image
            import io
            import base64

        # XXX controlnet only accepts RAW base64 without headers
        with io.BytesIO() as output_bytes:
            metadata = None
            for key, value in image.info.items():
                if isinstance(key, str) and isinstance(value, str):
                    if metadata is None:
                        metadata = PngImagePlugin.PngInfo()
                    metadata.add_text(key, value)
            image.save(output_bytes, format="PNG", pnginfo=metadata)

            bytes_data = output_bytes.getvalue()

        return str(base64.b64encode(bytes_data), "utf-8")


    @staticmethod
    def img2b64(image) -> str:
        return "data:image/png;base64," + File64BitsManager.raw_b64_img(image)    

    @staticmethod
    def b642img(b64img) -> str:
        try:
            from PIL import Image, PngImagePlugin
            import io
            import base64
        except:
            PackageManager.install_package("pillow")
            from PIL import Image
            import io
            import base64        
        image_data = re.sub('^data:image/.+;base64,', '', b64img)
        return Image.open(io.BytesIO(base64.b64decode(image_data)))  

    @staticmethod
    def get_supported_file_extensions_from_base64(b64data):
        # Extract the file extension from the base64 data
        data_match = re.match(r'^data:(.*?);base64,', b64data)
        if data_match:
            mime_type = data_match.group(1)
            extension = mime_type.split('/')[-1]
            return extension
        else:
            raise ValueError("Invalid base64 data format.")
        
    @staticmethod
    def extract_content_from_base64(b64data):
        # Split the base64 data at the comma separator
        header, content = b64data.split(',', 1)

        # Extract only the content part and remove any white spaces and newlines
        content = content.strip()

        return content

    @staticmethod
    def b642file(b64data, filename):
        import base64   
        # Extract the file extension from the base64 data
        
        
        # Save the file with the determined extension
        with open(filename, 'wb') as file:
            file.write(base64.b64decode(File64BitsManager.extract_content_from_base64(b64data)))

        return filename
class TFIDFLoader:
    @staticmethod
    def create_vectorizer_from_dict(tfidf_info):
        vectorizer = TfidfVectorizer(**tfidf_info['params'])
        vectorizer.vocabulary_ = tfidf_info['vocabulary']
        vectorizer.idf_ = [tfidf_info['idf_values'][feature] for feature in vectorizer.get_feature_names()]
        return vectorizer

    @staticmethod
    def create_dict_from_vectorizer(vectorizer):
        tfidf_info = {
            "vocabulary": vectorizer.vocabulary_,
            "idf_values": dict(zip(vectorizer.get_feature_names(), vectorizer.idf_)),
            "params": vectorizer.get_params()
        }
        return tfidf_info
    
class DocumentDecomposer:
    @staticmethod
    def clean_text(text):
        # Remove extra returns and leading/trailing spaces
        text = text.replace('\r', '').strip()
        return text

    @staticmethod
    def split_into_paragraphs(text):
        # Split the text into paragraphs using two or more consecutive newlines
        paragraphs = [p+"\n" for p in re.split(r'\n{2,}', text)]
        return paragraphs

    @staticmethod
    def tokenize_sentences(paragraph):
        # Custom sentence tokenizer using simple regex-based approach
        sentences = [s+"." for s in paragraph.split(".")]
        sentences = [sentence.strip() for sentence in sentences if sentence.strip()]
        return sentences

    @staticmethod
    def decompose_document(text, max_chunk_size, overlap_size, tokenize, detokenize):
        cleaned_text = DocumentDecomposer.clean_text(text)
        paragraphs = DocumentDecomposer.split_into_paragraphs(cleaned_text)

        # List to store the final clean chunks
        clean_chunks = []

        current_chunk = []  # To store the current chunk being built
        l=0
        for paragraph in paragraphs:
            # Tokenize the paragraph into sentences
            sentences = DocumentDecomposer.tokenize_sentences(paragraph)

            for sentence in sentences:
                # If adding the current sentence to the chunk exceeds the max_chunk_size,
                # we add the current chunk to the list of clean chunks and start a new chunk
                tokens = tokenize(sentence)
                nb_tokens = len(tokens)
                if nb_tokens>max_chunk_size:
                    while nb_tokens>max_chunk_size:
                        current_chunk += tokens[:max_chunk_size-l-1]
                        clean_chunks.append(current_chunk)
                        tokens = tokens[max_chunk_size-l-1-overlap_size:]
                        nb_tokens -= max_chunk_size-l-1-overlap_size
                        l=0
                        current_chunk = current_chunk[-overlap_size:]
                else:
                    if l + nb_tokens + 1 > max_chunk_size:

                        clean_chunks.append(current_chunk)
                        if overlap_size==0:
                            current_chunk = []
                        else:
                            current_chunk = current_chunk[-overlap_size:]
                        l=0

                    # Add the current sentence to the chunk
                    current_chunk += tokens
                    l += nb_tokens

        # Add the remaining chunk from the paragraph to the clean_chunks
        if current_chunk:
            clean_chunks.append(current_chunk)
            current_chunk = ""

        return clean_chunks
    
class TextVectorizer:
    def __init__(
                    self, 
                    vectorization_method, # supported "model_embedding" or "ftidf_vectorizer"
                    model=None, #needed in case of using model_embedding
                    database_path=None,
                    save_db=False,
                    visualize_data_at_startup=False,
                    visualize_data_at_add_file=False,
                    visualize_data_at_generate=False,
                    data_visualization_method="PCA",
                    database_dict=None
                    ):
        
        self.vectorization_method = vectorization_method
        self.save_db = save_db
        self.model = model
        self.database_file = database_path

        self.visualize_data_at_startup=visualize_data_at_startup
        self.visualize_data_at_add_file=visualize_data_at_add_file
        self.visualize_data_at_generate=visualize_data_at_generate
        
        self.data_visualization_method = data_visualization_method
        
        if database_dict is not None:
            self.chunks =  database_dict["chunks"]
            self.vectorizer = database_dict["vectorizer"]
            self.infos =   database_dict["infos"]
            self.ready = True
        else:
            self.chunks = {}
            self.ready = False
            self.vectorizer = None
        
            if vectorization_method=="model_embedding":
                try:
                    if not self.model or self.model.embed("hi")==None: # test
                        self.vectorization_method="ftidf_vectorizer"
                        self.infos={
                            "vectorization_method":"ftidf_vectorizer"
                        }
                    else:
                        self.infos={
                            "vectorization_method":"model_embedding"
                        }
                except Exception as ex:
                    ASCIIColors.error("Couldn't embed the text, so trying to use tfidf instead.")
                    trace_exception(ex)
                    self.infos={
                        "vectorization_method":"ftidf_vectorizer"
                    }
        # Load previous state from the JSON file
        if self.save_db:
            if Path(self.database_file).exists():
                ASCIIColors.success(f"Database file found : {self.database_file}")
                self.load_from_json()
                if self.visualize_data_at_startup:
                    self.show_document()
                self.ready = True
            else:
                ASCIIColors.info(f"No database file found : {self.database_file}")

                
    def show_document(self, query_text=None, save_fig_path =None, show_interactive_form=False):
        import textwrap
        import seaborn as sns
        import matplotlib.pyplot as plt
        import mplcursors
        from tkinter import Tk, Text, Scrollbar, Frame, Label, TOP, BOTH, RIGHT, LEFT, Y, N, END

        
        from sklearn.manifold import TSNE
        from sklearn.decomposition import PCA

        if self.data_visualization_method=="PCA":
            use_pca =  True
        else:
            use_pca =  False
        
        if use_pca:
            print("Showing pca representation :")
        else:
            print("Showing t-sne representation :")

        embeddings = {key:chunk["embeddings"] for key, chunk in self.chunks.items()}
        emb = list(embeddings.values())
        ref = list(embeddings.keys())
        if len(emb)>=2:
            # Normalize embeddings
            emb = np.vstack(emb)
            norms = np.linalg.norm(emb, axis=1)
            normalized_embeddings = emb / norms[:, np.newaxis]

            # Embed the query text
            if query_text is not None:
                query_embedding = self.embed_query(query_text)
                query_embedding = query_embedding.detach().squeeze().numpy()
                query_normalized_embedding = query_embedding / np.linalg.norm(query_embedding)

                # Combine the query embeddings with the document embeddings
                combined_embeddings = np.vstack((normalized_embeddings, query_normalized_embedding))
                ref.append("Quey_chunk_0")
            else:
                # Combine the query embeddings with the document embeddings
                combined_embeddings = normalized_embeddings

            if use_pca:
                # Use PCA for dimensionality reduction
                pca = PCA(n_components=2)
                try:
                    embeddings_2d = pca.fit_transform(combined_embeddings)
                except Exception as ex:
                    
                    embeddings_2d = []
            else:
                # Use t-SNE for dimensionality reduction
                # Adjust the perplexity value
                perplexity = min(30, combined_embeddings.shape[0] - 1)
                tsne = TSNE(n_components=2, perplexity=perplexity)
                embeddings_2d = tsne.fit_transform(combined_embeddings)

            # Create a dictionary to map document paths to colors
            document_path_colors = {}
            for i, path in enumerate(ref):
                document_path = "_".join(path.split("_")[:-1])  # Extract the document path (excluding chunk and chunk number)
                if document_path not in document_path_colors:
                    # Assign a new color to the document path if it's not in the dictionary
                    document_path_colors[document_path] = sns.color_palette("hls", len(document_path_colors) + 1)[-1]

            # Generate a list of colors for each data point based on the document path
            point_colors = [document_path_colors["_".join(path.split("_")[:-1])] for path in ref]
            

            # Create a scatter plot using Seaborn
            sns.scatterplot(x=embeddings_2d[:, 0], y=embeddings_2d[:, 1], hue=point_colors)  # Plot document embeddings
            # Add labels to the scatter plot
            for i, (x, y) in enumerate(embeddings_2d[:-1]):
                plt.text(x, y, str(i), fontsize=8)

            plt.xlabel('Dimension 1')
            plt.ylabel('Dimension 2')
            if use_pca:      
                plt.title('Embeddings Scatter Plot based on PCA')
            else:
                plt.title('Embeddings Scatter Plot based on t-SNE')
            # Enable mplcursors to show tooltips on hover
            cursor = mplcursors.cursor(hover=True)

            # Define the hover event handler
            @cursor.connect("add")
            def on_hover(sel):
                index = sel.target.index
                if index > 0:
                    text = self.chunks[index]["chunk_text"]
                    wrapped_text = textwrap.fill(text, width=50)  # Wrap the text into multiple lines
                    sel.annotation.set_text(f"Index: {index}\nText:\n{wrapped_text}")
                else:
                    sel.annotation.set_text("Query")

            # Define the click event handler using matplotlib event handling mechanism
            def on_click(event):
                if event.xdata is not None and event.ydata is not None:
                    x, y = event.xdata, event.ydata
                    distances = ((embeddings_2d[:, 0] - x) ** 2 + (embeddings_2d[:, 1] - y) ** 2)
                    index = distances.argmin()
                    text = self.chunks[index]["chunk_text"] if index < len(self.chunks) else query_text

                    # Open a new Tkinter window with the content of the text
                    root = Tk()
                    root.title(f"Text for Index {index}")
                    frame = Frame(root)
                    frame.pack(fill=BOTH, expand=True)

                    label = Label(frame, text="Text:")
                    label.pack(side=TOP, padx=5, pady=5)

                    text_box = Text(frame)
                    text_box.pack(side=TOP, padx=5, pady=5, fill=BOTH, expand=True)
                    text_box.insert(END, text)

                    scrollbar = Scrollbar(frame)
                    scrollbar.pack(side=RIGHT, fill=Y)
                    scrollbar.config(command=text_box.yview)
                    text_box.config(yscrollcommand=scrollbar.set)

                    text_box.config(state="disabled")

                    root.mainloop()

            # Connect the click event handler to the figure
            plt.gcf().canvas.mpl_connect("button_press_event", on_click)
            if save_fig_path:
                try:
                    plt.savefig(save_fig_path)
                except Exception as ex:
                    trace_exception(ex)
            if show_interactive_form:
                plt.show()
    
    def file_exists(self, document_name:str)->bool:
        # Loop through the list of dictionaries
        for dictionary in self.chunks:
            if 'document_name' in dictionary and dictionary['document_name'] == document_name:
                # If the document_name is found in the current dictionary, set the flag to True and break the loop
                document_name_found = True
                return True
        return False
    
    def remove_document(self, document_name:str):
        for dictionary in self.chunks:
            if 'document_name' in dictionary and dictionary['document_name'] == document_name:
                # If the document_name is found in the current dictionary, set the flag to True and break the loop
                self.chunks.remove(dictionary)
                return True
        return False



    def add_document(self, document_name:Path, text:str, chunk_size: int, overlap_size:int, force_vectorize=False,add_as_a_bloc=False):
        if self.file_exists(document_name) and not force_vectorize:
            print(f"Document {document_name} already exists. Skipping vectorization.")
            return
        if add_as_a_bloc:
            chunks_text = [self.model.tokenize(text)]
            for i, chunk in enumerate(chunks_text):
                chunk_id = f"{document_name}_chunk_{i + 1}"
                chunk_dict = {
                    "document_name": document_name,
                    "chunk_index": i+1,
                    "chunk_text":self.model.detokenize(chunk),
                    "chunk_tokens": chunk,
                    "embeddings":[]
                }
                self.chunks[chunk_id] = chunk_dict
        else:
            chunks_text = DocumentDecomposer.decompose_document(text, chunk_size, overlap_size, self.model.tokenize, self.model.detokenize)
            for i, chunk in enumerate(chunks_text):
                chunk_id = f"{document_name}_chunk_{i + 1}"
                chunk_dict = {
                    "document_name": document_name,
                    "chunk_index": i+1,
                    "chunk_text":self.model.detokenize(chunk),
                    "chunk_tokens": chunk,
                    "embeddings":[]
                }
                self.chunks[chunk_id] = chunk_dict
        
    def index(self):
        if self.vectorization_method=="ftidf_vectorizer":
            self.vectorizer = TfidfVectorizer()
            #if self.debug:
            #    ASCIIColors.yellow(','.join([len(chunk) for chunk in chunks]))
            data=[]
            for k,chunk in self.chunks.items():
                try:
                    data.append(chunk["chunk_text"]) 
                except Exception as ex:
                    print("oups")
            self.vectorizer.fit(data)

        # Generate embeddings for each chunk
        for chunk_id, chunk in self.chunks.items():
            # Store chunk ID, embeddings, and original text
            try:
                if self.vectorization_method=="ftidf_vectorizer":
                    chunk["embeddings"] = self.vectorizer.transform([chunk["chunk_text"]]).toarray()
                else:
                    chunk["embeddings"] = self.model.embed(chunk["chunk_text"])
            except Exception as ex:
                print("oups")

        if self.save_db:
            self.save_to_json()
            
        self.ready = True
        if self.visualize_data_at_add_file:
            self.show_document()


    def embed_query(self, query_text):
        # Generate query embeddings
        if self.vectorization_method=="ftidf_vectorizer":
            query_embedding = self.vectorizer.transform([query_text]).toarray()
        else:
            query_embedding = self.model.embed(query_text)
            if query_embedding is None:
                ASCIIColors.warning("The model doesn't implement embeddings extraction")
                self.vectorization_method="ftidf_vectorizer"
                query_embedding = self.vectorizer.transform([query_text]).toarray()

        return query_embedding

    def recover_text(self, query_embedding, top_k=1):
        from sklearn.metrics.pairwise import cosine_similarity
        similarities = {}
        for chunk_id, chunk in self.chunks.items():
            similarity = cosine_similarity(query_embedding, chunk["embeddings"])
            similarities[chunk_id] = similarity

        # Sort the similarities and retrieve the top-k most similar embeddings
        sorted_similarities = sorted(similarities.items(), key=lambda x: x[1], reverse=True)[:top_k]

        # Retrieve the original text associated with the most similar embeddings
        texts = [self.chunks[chunk_id]["chunk_text"] for chunk_id, _ in sorted_similarities]

        if self.visualize_data_at_generate:
            self.show_document()

        return texts, sorted_similarities

    def toJson(self):
        state = {
            "chunks": self.chunks,
            "infos": self.infos,
            "vectorizer": TFIDFLoader.create_vectorizer_from_dict(self.vectorizer) if self.vectorization_method=="ftidf_vectorizer" else None
        }
        return state
    
    def setVectorizer(self, vectorizer_dict:dict):
        self.vectorizer=TFIDFLoader.create_vectorizer_from_dict(vectorizer_dict)

    def save_to_json(self):
        state = {
            "chunks": self.chunks,
            "infos": self.infos,
            "vectorizer": TFIDFLoader.create_vectorizer_from_dict(self.vectorizer) if self.vectorization_method=="ftidf_vectorizer" else None
        }
        with open(self.database_file, "w") as f:
            json.dump(state, f, cls=NumpyEncoderDecoder, indent=4)

    def load_from_json(self):

        ASCIIColors.info("Loading vectorized documents")
        with open(self.database_file, "r") as f:
            database = json.load(f, object_hook=NumpyEncoderDecoder.as_numpy_array)
            self.chunks = database["chunks"]
            self.infos= database["infos"]
            self.ready = True
        if self.vectorization_method=="ftidf_vectorizer":
            from sklearn.feature_extraction.text import TfidfVectorizer
            data = [c["chunk_text"] for k,c in self.chunks.items()]
            if len(data)>0:
                self.vectorizer = TfidfVectorizer()
                self.vectorizer.fit(data)
                self.embeddings={}
                for k,chunk in self.chunks.items():
                    chunk["embeddings"][k]= self.vectorizer.transform([chunk["embeddings"]]).toarray()
                    
                    
    def clear_database(self):
        self.ready = False
        self.vectorizer=None
        self.chunks = {}
        self.infos={}
        if self.save_db:
            self.save_to_json()
            
      




class PromptReshaper:
    def __init__(self, template:str):
        self.template = template
    def replace(self, placeholders:dict)->str:
        template = self.template
        # Calculate the number of tokens for each placeholder
        for placeholder, text in placeholders.items():
            template = template.replace(placeholder, text)
        return template
    def build(self, placeholders:dict, tokenize, detokenize, max_nb_tokens:int, place_holders_to_sacrifice:list=[])->str:
        # Tokenize the template without placeholders
        template_text = self.template
        template_tokens = tokenize(template_text)
        
        # Calculate the number of tokens in the template without placeholders
        template_tokens_count = len(template_tokens)
        
        # Calculate the number of tokens for each placeholder
        placeholder_tokens_count = {}
        all_count = template_tokens_count
        for placeholder, text in placeholders.items():
            text_tokens = tokenize(text)
            placeholder_tokens_count[placeholder] = len(text_tokens)
            all_count += placeholder_tokens_count[placeholder]

        def fill_template(template, data):
            for key, value in data.items():
                placeholder = "{{" + key + "}}"
                template = template.replace(placeholder, value)
            return template
        
        if max_nb_tokens-all_count>0 or len(place_holders_to_sacrifice)==0:
            return fill_template(self.template, placeholders)
        else:
            to_remove = -int((max_nb_tokens - all_count)/len(place_holders_to_sacrifice))
            for placeholder, text in placeholders.items():
                if placeholder in place_holders_to_sacrifice:
                    text_tokens = tokenize(text)[to_remove:]
                    placeholders[placeholder]=detokenize(text_tokens)
            return fill_template(self.template, placeholders)



class LOLLMSLocalizer:
    def __init__(self, dictionary):
        self.dictionary = dictionary

    def localize(self, input_string):
        def replace(match):
            key = match.group(1)
            return self.dictionary.get(key, match.group(0))
        
        import re
        pattern = r'@<([^>]+)>@'
        localized_string = re.sub(pattern, replace, input_string)
        return localized_string


class File_Path_Generator:
    @staticmethod
    def generate_unique_file_path(folder_path, file_base_name, file_extension):
        folder_path = Path(folder_path)
        index = 0
        while True:
            # Construct the full file path with the current index
            file_name = f"{file_base_name}_{index}.{file_extension}"
            full_file_path = folder_path / file_name
            
            # Check if the file already exists in the folder
            if not full_file_path.exists():
                return full_file_path
            
            # If the file exists, increment the index and try again
            index += 1