new updates

app.py

@@ -1,6 +1,14 @@
 import gradio as gr
 import os
 
+
+import string
+import random
+import requests
+from bs4 import BeautifulSoup
+from datetime import datetime
+
+
 from langchain.document_loaders import PyPDFLoader
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain.vectorstores import Chroma
@@ -22,8 +30,8 @@ import accelerate
 
 
 # default_persist_directory = './chroma_HF/'
-list_llm = ["mistralai/Mixtral-8x7B-Instruct-v0.1", "mistralai/Mistral-7B-Instruct-v0.2", "mistralai/Mistral-7B-Instruct-v0.1", \
-    "HuggingFaceH4/zephyr-7b-beta", "meta-llama/Llama-2-7b-chat-hf", "microsoft/phi-2", \
+list_llm = ["mistralai/Mistral-7B-Instruct-v0.2", "mistralai/Mistral-7B-Instruct-v0.1", \
+    "HuggingFaceH4/zephyr-7b-beta", "NousResearch/Llama-2-7b-chat-hf",  \
     "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "mosaicml/mpt-7b-instruct", "tiiuae/falcon-7b-instruct", \
     "google/flan-t5-xxl"
 ]
@@ -31,20 +39,45 @@ list_llm_simple = [os.path.basename(llm) for llm in list_llm]
 
 # Load PDF document and create doc splits
 def load_doc(list_file_path, chunk_size, chunk_overlap):
-    # Processing for one document only
-    # loader = PyPDFLoader(file_path)
-    # pages = loader.load()
     loaders = [PyPDFLoader(x) for x in list_file_path]
     pages = []
     for loader in loaders:
         pages.extend(loader.load())
-    # text_splitter = RecursiveCharacterTextSplitter(chunk_size = 600, chunk_overlap = 50)
     text_splitter = RecursiveCharacterTextSplitter(
         chunk_size = chunk_size, 
         chunk_overlap = chunk_overlap)
     doc_splits = text_splitter.split_documents(pages)
     return doc_splits
 
+def convert_github_url_to_raw(url):
+    # Ensure the URL is a GitHub blob URL
+    if "github.com" in url and "/blob/" in url:
+        raw_url = url.replace("github.com", "raw.githubusercontent.com").replace("/blob", "")
+        response = requests.get(raw_url)
+        html_content = response.text
+        # Step 2: Find the GitHub Icon and Extract the Link
+        soup = BeautifulSoup(html_content, "html.parser")
+        github_icon_link = None
+        for a in soup.find_all('a', href=True):
+            if "github.com" in a['href']:  # Assuming the GitHub link contains "github.com"
+                github_icon_link = a['href']
+                break
+        markdown_url = convert_github_url_to_raw(github_icon_link)
+        response = requests.get(markdown_url)
+        return response
+    else:
+        return ''
+
+def load_url(list_url_path, chunk_size, chunk_overlap):
+    texts = [convert_github_url_to_raw(x) for x in list_url_path]
+    pages = []
+    for text in texts:
+        pages.extend(text)
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size = chunk_size, 
+        chunk_overlap = chunk_overlap)
+    doc_splits = text_splitter.split_documents(pages)
+    return doc_splits
 
 # Create vector database
 def create_db(splits, collection_name):
@@ -59,7 +92,6 @@ def create_db(splits, collection_name):
     )
     return vectordb
 
-
 # Load vector database
 def load_db():
     embedding = HuggingFaceEmbeddings()
@@ -68,57 +100,18 @@ def load_db():
         embedding_function=embedding)
     return vectordb
 
-
 # Initialize langchain LLM chain
 def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
     progress(0.1, desc="Initializing HF tokenizer...")
-    # HuggingFacePipeline uses local model
-    # Note: it will download model locally...
-    # tokenizer=AutoTokenizer.from_pretrained(llm_model)
-    # progress(0.5, desc="Initializing HF pipeline...")
-    # pipeline=transformers.pipeline(
-    #     "text-generation",
-    #     model=llm_model,
-    #     tokenizer=tokenizer,
-    #     torch_dtype=torch.bfloat16,
-    #     trust_remote_code=True,
-    #     device_map="auto",
-    #     # max_length=1024,
-    #     max_new_tokens=max_tokens,
-    #     do_sample=True,
-    #     top_k=top_k,
-    #     num_return_sequences=1,
-    #     eos_token_id=tokenizer.eos_token_id
-    #     )
-    # llm = HuggingFacePipeline(pipeline=pipeline, model_kwargs={'temperature': temperature})
-    
     # HuggingFaceHub uses HF inference endpoints
     progress(0.5, desc="Initializing HF Hub...")
     # Use of trust_remote_code as model_kwargs
-    # Warning: langchain issue
     # URL: https://github.com/langchain-ai/langchain/issues/6080
-    if llm_model == "mistralai/Mixtral-8x7B-Instruct-v0.1":
-        llm = HuggingFaceHub(
-            repo_id=llm_model, 
-            model_kwargs={"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k, "load_in_8bit": True}
-        )
-    elif llm_model == "microsoft/phi-2":
-        raise gr.Error("phi-2 model requires 'trust_remote_code=True', currently not supported by langchain HuggingFaceHub...")
-        llm = HuggingFaceHub(
-            repo_id=llm_model, 
-            model_kwargs={"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k, "trust_remote_code": True, "torch_dtype": "auto"}
-        )
-    elif llm_model == "TinyLlama/TinyLlama-1.1B-Chat-v1.0":
+    if llm_model == "TinyLlama/TinyLlama-1.1B-Chat-v1.0":
         llm = HuggingFaceHub(
             repo_id=llm_model, 
             model_kwargs={"temperature": temperature, "max_new_tokens": 250, "top_k": top_k}
         )
-    elif llm_model == "meta-llama/Llama-2-7b-chat-hf":
-        raise gr.Error("Llama-2-7b-chat-hf model requires a Pro subscription...")
-        llm = HuggingFaceHub(
-            repo_id=llm_model, 
-            model_kwargs={"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k}
-        )
     else:
         llm = HuggingFaceHub(
             repo_id=llm_model, 
@@ -148,30 +141,36 @@ def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, pr
     progress(0.9, desc="Done!")
     return qa_chain
 
-
 # Initialize database
-def initialize_database(list_file_obj, chunk_size, chunk_overlap, progress=gr.Progress()):
+def initialize_database(list_file_obj, input_urls, chunk_size, chunk_overlap, progress=gr.Progress()):
     # Create list of documents (when valid)
     list_file_path = [x.name for x in list_file_obj if x is not None]
+    list_url = [x for x in input_urls if x is not None]
+
     # Create collection_name for vector database
-    progress(0.1, desc="Creating collection name...")
-    collection_name = Path(list_file_path[0]).stem
-    # Fix potential issues from naming convention
-    collection_name = collection_name.replace(" ","-") 
-    collection_name = collection_name[:50]
-    # print('list_file_path: ', list_file_path)
+    progress(0.1, desc="Creating collection...")
+    # collection_name = Path(list_file_path[0]).stem
+
+    # # Fix potential issues from naming convention
+    # collection_name = collection_name.replace(" ","-") 
+    # collection_name = collection_name[:50]
+    res = ''.join(random.choices(string.ascii_letters, k=10)) 
+    collection_name = f"HuggingFace101_{res}"
     print('Collection name: ', collection_name)
     progress(0.25, desc="Loading document...")
+    
     # Load document and create splits
     doc_splits = load_doc(list_file_path, chunk_size, chunk_overlap)
+    print(type(doc_splits))
+    
     # Create or load vector database
     progress(0.5, desc="Generating vector database...")
+    
     # global vector_db
     vector_db = create_db(doc_splits, collection_name)
     progress(0.9, desc="Done!")
     return vector_db, collection_name, "Complete!"
 
-
 def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
     # print("llm_option",llm_option)
     llm_name = list_llm[llm_option]
@@ -179,7 +178,6 @@ def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, pr
     qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db, progress)
     return qa_chain, "Complete!"
 
-
 def format_chat_history(message, chat_history):
     formatted_chat_history = []
     for user_message, bot_message in chat_history:
@@ -187,7 +185,6 @@ def format_chat_history(message, chat_history):
         formatted_chat_history.append(f"Assistant: {bot_message}")
     return formatted_chat_history
     
-
 def conversation(qa_chain, message, history):
     formatted_chat_history = format_chat_history(message, history)
     #print("formatted_chat_history",formatted_chat_history)
@@ -209,7 +206,6 @@ def conversation(qa_chain, message, history):
     # return gr.update(value=""), new_history, response_sources[0], response_sources[1] 
     return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page
     
-
 def upload_file(file_obj):
     list_file_path = []
     for idx, file in enumerate(file_obj):
@@ -227,16 +223,16 @@ def demo():
         collection_name = gr.State()
         
         gr.Markdown(
-        """<center><h2>PDF-based chatbot (powered by LangChain and open-source LLMs)</center></h2>
-        <h3>Ask any questions about your PDF documents, along with follow-ups</h3>
-        <b>Note:</b> This AI assistant performs retrieval-augmented generation from your PDF documents. \
+        """<center><h2>HugginFace Articles URL-based chatbot (powered by LangChain and open-source LLMs)</center></h2>
+        <h3>Ask any questions about your Huggingface Articles, along with follow-ups</h3>
+        <b>Note:</b> This AI assistant performs retrieval-augmented generation from Huggingface Articles. \
         When generating answers, it takes past questions into account (via conversational memory), and includes document references for clarity purposes.</i>
         <br><b>Warning:</b> This space uses the free CPU Basic hardware from Hugging Face. Some steps and LLM models used below (free inference endpoints) can take some time to generate an output.<br>
         """)
         with gr.Tab("Step 1 - Document pre-processing"):
             with gr.Row():
                 document = gr.Files(height=100, file_count="multiple", file_types=["pdf"], interactive=True, label="Upload your PDF documents (single or multiple)")
-                # upload_btn = gr.UploadButton("Loading document...", height=100, file_count="multiple", file_types=["pdf"], scale=1)
+                input_url = gr.Textbox(label="Or enter a URL", placeholder="https://example.com")
             with gr.Row():
                 db_btn = gr.Radio(["ChromaDB"], label="Vector database type", value = "ChromaDB", type="index", info="Choose your vector database")
             with gr.Accordion("Advanced options - Document text splitter", open=False):
@@ -247,7 +243,7 @@ def demo():
             with gr.Row():
                 db_progress = gr.Textbox(label="Vector database initialization", value="None")
             with gr.Row():
-                db_btn = gr.Button("Generate vector database...")
+                db_btn = gr.Button("Generating vector database...")
             
         with gr.Tab("Step 2 - QA chain initialization"):
             with gr.Row():
@@ -283,7 +279,7 @@ def demo():
         # Preprocessing events
         #upload_btn.upload(upload_file, inputs=[upload_btn], outputs=[document])
         db_btn.click(initialize_database, \
-            inputs=[document, slider_chunk_size, slider_chunk_overlap], \
+            inputs=[document, input_url, slider_chunk_size, slider_chunk_overlap], \
             outputs=[vector_db, collection_name, db_progress])
         qachain_btn.click(initialize_LLM, \
             inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], \

requirements.txt

@@ -6,4 +6,4 @@ tqdm
 accelerate
 pypdf
 chromadb
-
+bs4