a

README.md

@@ -11,3 +11,5 @@ license: apache-2.0
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+this project is used in huggingface space "teslaalerts"

pages/Tesla_Alerts.py

@@ -1,153 +1,153 @@
-from langchain.prompts import PromptTemplate
-from langchain.output_parsers import PydanticOutputParser
-from langchain.chat_models import ChatOpenAI
-
-from llama_index import VectorStoreIndex, ServiceContext, StorageContext,  download_loader, SimpleDirectoryReader
-from llama_index.vector_stores import FaissVectorStore
-from llama_index.tools import QueryEngineTool, ToolMetadata
-from llama_index.query_engine import SubQuestionQueryEngine
-from llama_index.embeddings import OpenAIEmbedding
-from llama_index.schema import Document
-from llama_index.node_parser import UnstructuredElementNodeParser
-from llama_index.llms import OpenAI
-
-import streamlit as st
 import os
-import faiss
+import sys
+
+from llama_index import SimpleDirectoryReader, ServiceContext, StorageContext, VectorStoreIndex, download_loader,load_index_from_storage
+from llama_index.llms import HuggingFaceLLM
+from llama_index.embeddings import HuggingFaceEmbedding
+from llama_index.vector_stores import ChromaVectorStore
+from llama_index.storage.index_store import SimpleIndexStore
+from llama_index.indices.query.schema import QueryBundle, QueryType
+import chromadb
+import streamlit as st
 import time
 
 
-st.set_page_config(page_title="Yield Case Analyzer", page_icon=":card_index_dividers:", initial_sidebar_state="expanded", layout="wide")
+st.set_page_config(page_title="Tesla Alert Analyzer", page_icon=":card_index_dividers:", initial_sidebar_state="expanded", layout="wide")
 
-st.title(":card_index_dividers: Yield Case Analyzer")
+st.title(":card_index_dividers: Tesla Alert Analyzer")
 st.info("""
-Begin by uploading the case report in PDF format. Afterward, click on 'Process Document'. Once the document has been processed. You can enter question and click send, system will answer your question.
+Begin by uploading the case report in pptx format. Afterward, click on 'Process Document'. Once the document has been processed. You can enter question and click send, system will answer your question.
 """)
 
 
-def get_model(model_name):
-    OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
-    if model_name == "openai":
-        model = ChatOpenAI(openai_api_key=OPENAI_API_KEY, model_name="gpt-3.5-turbo")
-    return model
+if "process_doc" not in st.session_state:
+    st.session_state.process_doc = False
 
-def get_vector_index(docs, vector_store):
-    print(docs)
-    llm = get_model("openai")
-    if vector_store == "faiss":
-        d = 1536
-        faiss_index = faiss.IndexFlatL2(d)
-        vector_store = FaissVectorStore(faiss_index=faiss_index)
-        storage_context = StorageContext.from_defaults(vector_store=vector_store)
-        # embed_model = OpenAIEmbedding()
-        # service_context = ServiceContext.from_defaults(embed_model=embed_model)
-        service_context = ServiceContext.from_defaults(llm=llm)
-        index = VectorStoreIndex(docs,
-            service_context=service_context,
-            storage_context=storage_context
-        )
-    elif vector_store == "simple":
-        index = VectorStoreIndex.from_documents(docs)
 
 
-    return index
 
+def fmetadata(dummy: str): return {"file_path": ""}
 
+def load_files(file_dir):
 
-def generate_insight(engine, search_string):
+    PptxReader = download_loader("PptxReader")
+    loader = SimpleDirectoryReader(input_dir=file_dir, file_extractor={".pptx": PptxReader(),}, file_metadata=fmetadata)
+    documents  = loader.load_data()
 
-    with open("prompts/main.prompt", "r") as f:
-        template = f.read()
+    for doc in documents:
+        doc.metadata["file_path"]=""
 
-    prompt_template = PromptTemplate(
-        template=template,
-        input_variables=['search_string']
-    )
+    return documents 
 
-    formatted_input = prompt_template.format(search_string=search_string)
-    print(formatted_input)
-    response = engine.query(formatted_input)
-    return response.response
+system_prompt = "You are a Q&A assistant. "
+system_prompt += "Your goal is to answer questions as accurately as possible based on the instructions and context provided."
+system_prompt += "Please say you do not know if you do not find answer."
 
+# This will wrap the default prompts that are internal to llama-index
+query_wrapper_prompt = "<|USER|>{query_str}<|ASSISTANT|>"
 
-def get_query_engine(engine):
-    llm = get_model("openai")
-    service_context = ServiceContext.from_defaults(llm=llm)
 
-    query_engine_tools = [
-        QueryEngineTool(
-            query_engine=engine,
-            metadata=ToolMetadata(
-                name="Alert Report",
-                description=f"Provides information about the alerts from alerts files uploaded.",
-            ),
-        ),
-    ]
 
+import torch
+#torch.set_default_device('cuda')
 
-    s_engine = SubQuestionQueryEngine.from_defaults(
-        query_engine_tools=query_engine_tools,
-        service_context=service_context
+@st.cache_resource
+def llm_loading():
+    print("before huggingfacellm")
+    llm = HuggingFaceLLM(
+        context_window=8000,
+        max_new_tokens=500,
+        generate_kwargs={"temperature": 0.1, "do_sample": True},
+        system_prompt=system_prompt,
+        query_wrapper_prompt=query_wrapper_prompt,
+        tokenizer_name="mistralai/Mistral-7B-Instruct-v0.1",
+        model_name="mistralai/Mistral-7B-Instruct-v0.1",
+        device_map="auto",
+        tokenizer_kwargs={"max_length": 8000},
+        model_kwargs={"torch_dtype": torch.float16}
     )
-    return s_engine
-
-
-if "process_doc" not in st.session_state:
-        st.session_state.process_doc = False
 
+    print("after huggingfacellm")
+    embed_model =  HuggingFaceEmbedding(model_name="thenlper/gte-base")
+    print("after embed_model")
+    return llm,embed_model
 
-OPENAI_API_KEY = "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"
-os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY
+llm, embed_model = llm_loading()
 
+files_uploaded = st.sidebar.file_uploader("Upload the case report in PPT format", type="pptx", accept_multiple_files=True)
 
-if OPENAI_API_KEY:
-    files_uploaded = st.sidebar.file_uploader("Upload the case report in PDF format", type="pptx")
-    st.sidebar.info("""
-    Example pdf reports you can upload here:
-    """)
-
-    if st.sidebar.button("Process Document"):
-        with st.spinner("Processing Document..."):
-
-            data_dir = "./data"
-            if not os.path.exists(data_dir):
-                os.makedirs(data_dir)
-
-            for file in files_uploaded:
-                print(f'file named {file.name}')
-                fname=f'{data_dir}/{file.name}'
-                with open(fname, 'wb') as f:
-                    f.write(file.read())
-
-            def fmetadata(dummy: str): return {"file_path": ""}
-
-            PptxReader = download_loader("PptxReader")
-            loader =  SimpleDirectoryReader(input_dir=data_dir, file_extractor={".pptx": PptxReader(),}, file_metadata=fmetadata)
-
-            documents = loader.load_data()
-            for doc in documents:
-                doc.metadata["file_path"]=""
-
-            st.session_state.index = get_vector_index(documents, vector_store="faiss")
-            #st.session_state.index = get_vector_index(documents, vector_store="simple")
-            st.session_state.process_doc = True
-            st.toast("Document Processsed!")
+st.sidebar.info("""
+Example pptx reports you can upload here:
+""")
 
-    #st.session_state.process_doc = True
+if st.sidebar.button("Process Document"):
 
-    if st.session_state.process_doc:
-        search_text = st.text_input("Enter your question")
-        if st.button("Submit"):
-            engine = get_query_engine(st.session_state.index.as_query_engine(similarity_top_k=3))
-            start_time = time.time()
+    with st.spinner("Processing Document..."):
 
-            st.write("Alert search result...")
-            response = generate_insight(engine, search_text)
-            st.write(response)
-            #st.session_state["end_time"] = "{:.2f}".format((time.time() - start_time))
+        data_dir = "data"
+        if not os.path.exists(data_dir):
+            os.makedirs(data_dir)
 
-            st.toast("Report Analysis Complete!")
+        for uploaded_file in files_uploaded:
+            print(f'file named {uploaded_file.name}')
+            fname=f'{data_dir}/{uploaded_file.name}'
+            with open(fname, 'wb') as f:
+                f.write(uploaded_file.read())
+        
+        documents=load_files(data_dir)
 
-        #if st.session_state.end_time:
-        #    st.write("Report Analysis Time: ", st.session_state.end_time, "s")
+        collection_name = "tesla_report"
+        chroma_client = chromadb.PersistentClient()
+        chroma_collection = chroma_client.get_or_create_collection(collection_name)
+        vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
+        storage_context = StorageContext.from_defaults(vector_store=vector_store)
+        service_context = ServiceContext.from_defaults(
+            chunk_size=8000,
+            llm=llm,
+            embed_model=embed_model
+        )
+        index = VectorStoreIndex.from_documents(documents, service_context=service_context, storage_context=storage_context)
+        index.storage_context.persist()
+
+        #chroma_collection.peek()
+        
+        #st.session_state.index = index
+        st.session_state.process_doc = True
+
+        st.toast("Document Processsed!")
+
+        #st.session_state.process_doc = True
+
+def clear_form():
+    st.session_state.query_text = st.session_state["question"] 
+    st.session_state["question"] = ""
+    st.session_state["response"] = ""
+
+@st.cache_resource
+def reload_index(_llm,_embed_model, col ) :
+    chroma_client = chromadb.PersistentClient()
+    chroma_collection = chroma_client.get_or_create_collection(col)
+    vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
+    service_context = ServiceContext.from_defaults(llm=llm,embed_model=embed_model)
+    load_index = VectorStoreIndex.from_vector_store(service_context=service_context, 
+                                     vector_store=vector_store)
+    return load_index
+
+if st.session_state.process_doc:
+    #alert number looks like APP_wnnn where nnn is a number. Please list out all the alerts uploaded in these files!
+    search_text = st.text_input("Enter your question", key='question' )
+    if st.button(label="Submit", on_click=clear_form):
+        index = reload_index(llm,embed_model,"tesla_report" )
+        query_engine =  index.as_query_engine()
+        start_time = time.time()
+        #qry = QueryBundle(search_text)
+        #alert number looks like APP_wnnn where nnn is a number. Please list out all the alerts uploaded in these files!"
+        st.write("Processing....")
+        search_text = st.session_state.query_text
+        print(search_text)
+        response = query_engine.query(search_text)
+        st.write(response.response)
+        #st.session_state["end_time"] = "{:.2f}".format((time.time() - start_time))
+
+        st.toast("Report Analysis Complete!")
 

prompts/main.prompt

@@ -1,3 +1,3 @@
 You are a tesla alert analyzer.
 Your job is the is to provide a detailed analysis of the alert and follow-up
-
+Question: {search_string}

requirements.txt

@@ -5,3 +5,4 @@ transformers
 accelerate
 openai
 streamlit
+faiss