Duplicate from OrganizedProgrammers/PDFAISS-2.3.1

Co-authored-by: Alma atla <Almaatla@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,34 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,13 @@
+---
+title: PDFAISS 2.3
+emoji: 🏃
+colorFrom: red
+colorTo: green
+sdk: gradio
+sdk_version: 3.29.0
+app_file: app.py
+pinned: false
+duplicated_from: OrganizedProgrammers/PDFAISS-2.3.1
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,355 @@
+import langchain
+from langchain.embeddings import SentenceTransformerEmbeddings
+from langchain.chains.question_answering import load_qa_chain
+from langchain.document_loaders import UnstructuredPDFLoader,UnstructuredWordDocumentLoader
+from langchain.indexes import VectorstoreIndexCreator
+from langchain.vectorstores import FAISS
+from langchain import HuggingFaceHub
+from langchain import PromptTemplate
+from langchain.chat_models import ChatOpenAI
+from zipfile import ZipFile
+import gradio as gr
+import openpyxl
+import os
+import shutil
+from langchain.schema import Document
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+import tiktoken
+import secrets
+import openai
+import time
+
+tokenizer = tiktoken.encoding_for_model("gpt-3.5-turbo")
+
+# create the length function
+def tiktoken_len(text):
+    tokens = tokenizer.encode(
+        text,
+        disallowed_special=()
+    )
+    return len(tokens)
+
+text_splitter = RecursiveCharacterTextSplitter(
+    chunk_size=600,
+    chunk_overlap=200,
+    length_function=tiktoken_len,
+    separators=["\n\n", "\n", " ", ""]
+)
+
+embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
+foo = Document(page_content='foo is fou!',metadata={"source":'foo source'})
+
+def reset_database(ui_session_id):
+  session_id = f"PDFAISS-{ui_session_id}"
+  if 'drive' in session_id:
+    print("RESET DATABASE: session_id contains 'drive' !!")
+    return None
+
+  try:
+    shutil.rmtree(session_id)
+  except:
+    print(f'no {session_id} directory present')
+  
+  try:
+    os.remove(f"{session_id}.zip")
+  except:
+    print("no {session_id}.zip present")
+
+  return None
+
+def is_duplicate(split_docs,db):
+  epsilon=0.0
+  print(f"DUPLICATE: Treating: {split_docs[0].metadata['source'].split('/')[-1]}")
+  for i in range(min(3,len(split_docs))):
+    query = split_docs[i].page_content
+    docs = db.similarity_search_with_score(query,k=1)
+    _ , score = docs[0]
+    epsilon += score
+  print(f"DUPLICATE: epsilon: {epsilon}")
+  return epsilon < 0.1
+
+def merge_split_docs_to_db(split_docs,db,progress,progress_step=0.1):
+  progress(progress_step,desc="merging docs")
+  if len(split_docs)==0:
+    print("MERGE to db: NO docs!!")
+    return
+
+  filename = split_docs[0].metadata['source']
+  if is_duplicate(split_docs,db):
+    print(f"MERGE: Document is duplicated: {filename}")
+    return
+  print(f"MERGE: number of split docs: {len(split_docs)}")
+  batch = 10
+  for i in range(0, len(split_docs), batch):
+    progress(i/len(split_docs),desc=f"added {i} chunks of {len(split_docs)} chunks")
+    db1 = FAISS.from_documents(split_docs[i:i+batch], embeddings)
+    db.merge_from(db1)
+  return db
+
+def merge_pdf_to_db(filename,db,progress,progress_step=0.1):
+  progress_step+=0.05
+  progress(progress_step,'unpacking pdf')
+  doc = UnstructuredPDFLoader(filename).load()
+  doc[0].metadata['source'] = filename.split('/')[-1]
+  split_docs = text_splitter.split_documents(doc)
+  progress_step+=0.3
+  progress(progress_step,'docx unpacked')
+  return merge_split_docs_to_db(split_docs,db,progress,progress_step)
+
+def merge_docx_to_db(filename,db,progress,progress_step=0.1):
+  progress_step+=0.05
+  progress(progress_step,'unpacking docx')
+  doc = UnstructuredWordDocumentLoader(filename).load()
+  doc[0].metadata['source'] = filename.split('/')[-1]
+  split_docs = text_splitter.split_documents(doc)
+  progress_step+=0.3
+  progress(progress_step,'docx unpacked')
+  return merge_split_docs_to_db(split_docs,db,progress,progress_step)
+
+def merge_txt_to_db(filename,db,progress,progress_step=0.1):
+  progress_step+=0.05
+  progress(progress_step,'unpacking txt')
+  with open(filename) as f:
+      docs = text_splitter.split_text(f.read())
+      split_docs = [Document(page_content=doc,metadata={'source':filename.split('/')[-1]}) for doc in docs]
+  progress_step+=0.3
+  progress(progress_step,'txt unpacked')
+  return merge_split_docs_to_db(split_docs,db,progress,progress_step)
+
+def unpack_zip_file(filename,db,progress):
+    with ZipFile(filename, 'r') as zipObj:
+        contents = zipObj.namelist()
+        print(f"unpack zip: contents: {contents}")
+        tmp_directory = filename.split('/')[-1].split('.')[-2]
+        shutil.unpack_archive(filename, tmp_directory)
+
+        if 'index.faiss' in [item.lower() for item in contents]:
+            db2 = FAISS.load_local(tmp_directory, embeddings)
+            db.merge_from(db2)
+            return db
+        
+        for file in contents:
+            if file.lower().endswith('.docx'):
+              db = merge_docx_to_db(f"{tmp_directory}/{file}",db,progress)
+            if file.lower().endswith('.pdf'):
+              db = merge_pdf_to_db(f"{tmp_directory}/{file}",db,progress)
+            if file.lower().endswith('.txt'):
+              db = merge_txt_to_db(f"{tmp_directory}/{file}",db,progress)
+        return db
+
+def add_files_to_zip(session_id):
+    zip_file_name = f"{session_id}.zip"
+    with ZipFile(zip_file_name, "w") as zipObj:
+        for root, dirs, files in os.walk(session_id):
+            for file_name in files:
+                file_path = os.path.join(root, file_name)
+                arcname = os.path.relpath(file_path, session_id)
+                zipObj.write(file_path, arcname)
+
+## Summary functions ##
+
+## Load each doc from the vector store
+def load_docs(ui_session_id):
+    session_id_global_db = f"PDFAISS-{ui_session_id}"
+    try:
+        db = FAISS.load_local(session_id_global_db,embeddings)
+        print("load_docs after loading global db:",session_id_global_db,len(db.index_to_docstore_id))
+    except:
+        return f"SESSION: {session_id_global_db} database does not exist","",""
+    docs = []
+    for i in range(1,len(db.index_to_docstore_id)):
+        docs.append(db.docstore.search(db.index_to_docstore_id[i]))
+    return docs
+
+                
+# summarize with gpt 3.5 turbo
+def summarize_gpt(doc,system='provide a summary of the following document: ', first_tokens=600):
+    doc = doc.replace('\n\n\n', '').replace('---', '').replace('...', '').replace('___', '')
+    encoded = tokenizer.encode(doc)
+    print("/n TOKENIZED : ", encoded)
+    decoded = tokenizer.decode(encoded[:min(first_tokens, len(encoded))])
+    print("/n DOC SHORTEN", min(first_tokens, len(encoded)), " : ", decoded)
+    completion = openai.ChatCompletion.create(
+    model="gpt-3.5-turbo",
+    messages=[
+        {"role": "system", "content": system},
+        {"role": "user", "content": decoded}
+        ]
+    )
+    return completion.choices[0].message["content"]
+
+
+def summarize_docs_generator(apikey_input, session_id):
+    openai.api_key = apikey_input
+    docs=load_docs(session_id)
+    print("################# DOCS LOADED ##################", "docs type : ", type(docs[0]))
+    
+    try:
+        fail = docs[0].page_content
+    except:
+        return docs[0]
+
+    source = ""
+    summaries = ""
+    i = 0
+    while i<len(docs):
+        doc = docs[i]
+        unique_doc = ""
+        if source != doc.metadata:
+            unique_doc = ''.join([doc.page_content for doc in docs[i:i+3]])
+            print("\n\n****Open AI API called****\n\n")
+            if i == 0:
+                try:
+                    summary = summarize_gpt(unique_doc)
+                except:
+                    return f"ERROR : Try checking the validity of the provided OpenAI API Key"
+            else:
+                try:
+                    summary = summarize_gpt(unique_doc)
+                except:
+                    print(f"ERROR : There was an error but it is not linked with the validity of api key, taking a 20s nap")
+                    yield summaries + f"\n\n °°° OpenAI error, please wait 20 sec of cooldown. °°°"
+                    time.sleep(20)
+                    summary = summarize_gpt(unique_doc)
+
+            print("SUMMARY : ", summary)
+            summaries += f"Source : {doc.metadata['source'].split('/')[-1]}\n{summary} \n\n"
+            source = doc.metadata
+            yield summaries
+        i+=1
+    yield summaries
+
+
+def summarize_docs(apikey_input, session_id):
+    gen = summarize_docs_generator(apikey_input, session_id)
+    while True:
+        try:
+            yield str(next(gen))
+        except StopIteration:
+            return
+
+#### UI Functions ####
+
+def embed_files(files,ui_session_id,progress=gr.Progress(),progress_step=0.05):
+    print(files)
+    progress(progress_step,desc="Starting...")
+    split_docs=[]
+    if len(ui_session_id)==0:
+      ui_session_id = secrets.token_urlsafe(16)
+    session_id = f"PDFAISS-{ui_session_id}"
+
+    try:
+      db = FAISS.load_local(session_id,embeddings)
+    except:
+      print(f"SESSION: {session_id} database does not exist, create a FAISS db")
+      db =  FAISS.from_documents([foo], embeddings)
+      db.save_local(session_id)
+      print(f"SESSION: {session_id} database created")
+    
+    print("EMBEDDED, before embeddeding: ",session_id,len(db.index_to_docstore_id))
+    for file_id,file in enumerate(files):
+        print("ID : ", file_id,"FILE : ", file)
+        file_type = file.name.split('.')[-1].lower()
+        source = file.name.split('/')[-1]
+        print(f"current file: {source}")
+        progress(file_id/len(files),desc=f"Treating {source}")
+
+        if file_type == 'pdf':
+            db2 = merge_pdf_to_db(file.name,db,progress)
+        
+        if file_type == 'txt':
+            db2 = merge_txt_to_db(file.name,db,progress)
+        
+        if file_type == 'docx':
+            db2 = merge_docx_to_db(file.name,db,progress)
+
+        if file_type == 'zip':
+            db2 = unpack_zip_file(file.name,db,progress)
+
+        if db2 != None:
+            db = db2
+            db.save_local(session_id)
+            ### move file to store ###
+            progress(progress_step, desc = 'moving file to store')
+            directory_path = f"{session_id}/store/"
+            if not os.path.exists(directory_path):
+                os.makedirs(directory_path)
+            try:
+                shutil.move(file.name, directory_path)
+            except:
+                pass
+
+    ### load the updated db and zip it ###
+    progress(progress_step, desc = 'loading db')
+    db = FAISS.load_local(session_id,embeddings)
+    print("EMBEDDED, after embeddeding: ",session_id,len(db.index_to_docstore_id))
+    progress(progress_step, desc = 'zipping db for download')
+    add_files_to_zip(session_id)
+    print(f"EMBEDDED: db zipped")
+    progress(progress_step, desc = 'db zipped')
+    return f"{session_id}.zip",ui_session_id
+
+def display_docs(docs):
+  output_str = ''
+  for i, doc in enumerate(docs):
+      source = doc.metadata['source'].split('/')[-1]
+      output_str += f"Ref: {i+1}\n{repr(doc.page_content)}\nSource: {source}\n\n"
+  return output_str
+
+def ask_gpt(query, apikey,history,ui_session_id):
+    session_id = f"PDFAISS-{ui_session_id}"
+    try:
+      db = FAISS.load_local(session_id,embeddings)
+      print("ASKGPT after loading",session_id,len(db.index_to_docstore_id))
+    except:
+      print(f"SESSION: {session_id} database does not exist")
+      return f"SESSION: {session_id} database does not exist","",""
+
+    docs = db.similarity_search(query)
+    history += f"[query]\n{query}\n[answer]\n"
+    if(apikey==""):
+        history += f"None\n[references]\n{display_docs(docs)}\n\n"
+        return "No answer from GPT", display_docs(docs),history
+    else:
+        llm = ChatOpenAI(temperature=0, model_name = 'gpt-3.5-turbo', openai_api_key=apikey)
+        chain = load_qa_chain(llm, chain_type="stuff")
+        answer = chain.run(input_documents=docs, question=query, verbose=True)
+        history += f"{answer}\n[references]\n{display_docs(docs)}\n\n"
+        return answer,display_docs(docs),history
+
+with gr.Blocks() as demo:
+    gr.Markdown("Upload your documents and question them.")
+    with gr.Accordion("Open to enter your API key", open=False):
+        apikey_input = gr.Textbox(placeholder="Type here your OpenAI API key to use Summarization and Q&A", label="OpenAI API Key",type='password')
+    with gr.Tab("Upload PDF & TXT"): 
+        tb_session_id = gr.Textbox(label='session id')
+        docs_input = gr.File(file_count="multiple", file_types=[".txt", ".pdf",".zip",".docx"])
+        db_output = gr.outputs.File(label="Download zipped database")
+        btn_generate_db = gr.Button("Generate database")
+        btn_reset_db = gr.Button("Reset database")
+
+    with gr.Tab("Summarize PDF"):
+        with gr.Column():
+            summary_output = gr.Textbox(label='Summarized files')
+            btn_summary = gr.Button("Summarize")
+            summary_output.style(show_copy_button=True)
+
+    
+    with gr.Tab("Ask PDF"):
+        with gr.Column():
+            query_input = gr.Textbox(placeholder="Type your question", label="Question")
+            btn_askGPT = gr.Button("Answer")
+            answer_output = gr.Textbox(label='GPT 3.5 answer')
+            answer_output.style(show_copy_button=True)
+            sources = gr.Textbox(label='Sources')
+            sources.style(show_copy_button=True)
+            history = gr.Textbox(label='History')
+            history.style(show_copy_button=True)
+
+    btn_generate_db.click(embed_files, inputs=[docs_input,tb_session_id], outputs=[db_output,tb_session_id])
+    btn_reset_db.click(reset_database,inputs=[tb_session_id],outputs=[db_output])
+    btn_summary.click(summarize_docs, inputs=[apikey_input,tb_session_id], outputs=summary_output)
+    btn_askGPT.click(ask_gpt, inputs=[query_input,apikey_input,history,tb_session_id], outputs=[answer_output,sources,history])
+
+demo.queue(concurrency_count=10)
+demo.launch(debug=False,share=False)

requirements.txt

@@ -0,0 +1,8 @@
+langchain
+huggingface_hub
+sentence_transformers
+openai
+unstructured==0.6.3
+tiktoken
+faiss-cpu
+gradio