GPT4All contribution #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open

AleCola01 wants to merge 2 commits into eurotech:main from AleCola01:gpt4all

Indexes/EUROTECH_DSS/default__vector_store.json

This file was deleted.

Indexes/EUROTECH_DSS/docstore.json

This file was deleted.

Indexes/EUROTECH_DSS/graph_store.json

This file was deleted.

Indexes/EUROTECH_DSS/image__vector_store.json

This file was deleted.

Indexes/EUROTECH_DSS/index_store.json

This file was deleted.

0 .gitignore → gpt4all-project/.gitignore

File renamed without changes.

0 Documents/README.md → gpt4all-project/Documents/README.md

File renamed without changes.

0 ECP.png → gpt4all-project/ECP.png

File renamed without changes

0 Indexes/README.md → gpt4all-project/Indexes/README.md

File renamed without changes.

0 LICENSE → gpt4all-project/LICENSE

File renamed without changes.

0 README.md → gpt4all-project/README.md

File renamed without changes.

0 streamlit_app/images/ecp_agent.png → gpt4all-project/images/ecp_agent.png

File renamed without changes

0 streamlit_app/images/ecp_logo.png → gpt4all-project/images/ecp_logo.png

File renamed without changes

0 streamlit_app/images/ecp_user.png → gpt4all-project/images/ecp_user.png

File renamed without changes

0 streamlit_app/images/jetson-soc.png → gpt4all-project/images/jetson-soc.png

File renamed without changes

0 streamlit_app/images/logo.png → gpt4all-project/images/logo.png

File renamed without changes

0 streamlit_app/images/user-purple.png → gpt4all-project/images/user-purple.png

File renamed without changes

0 index_name_checked.png → gpt4all-project/index_name_checked.png

File renamed without changes

0 launch_dev.sh → gpt4all-project/launch_dev.sh

100755 → 100644

File renamed without changes.

0 launch_everyware_copilot.sh → gpt4all-project/launch_everyware_copilot.sh

File renamed without changes.

0 local_documents_selected.png → gpt4all-project/local_documents_selected.png

File renamed without changes

0 logs/README.md → gpt4all-project/logs/README.md

File renamed without changes.

gpt4all-project/record_manager_cache.sql

Binary file not shown.

0 setup_environment.sh → gpt4all-project/setup_environment.sh

100755 → 100644

File renamed without changes.

0 streamlit_app/.streamlit/config.toml → ...ject/streamlit_app/.streamlit/config.toml

File renamed without changes.

gpt4all-project/streamlit_app/app.py

-Original file line number
+Diff line change
@@ -0,0 +1,235 @@
+    import os
+    import streamlit as st
+    import chromadb
+    from langchain_community.llms import GPT4All
+    from langchain_community.embeddings import GPT4AllEmbeddings
+    from langchain.schema import SystemMessage, HumanMessage
+    from langchain_chroma.vectorstores import Chroma
+    from langchain_milvus.vectorstores import Milvus
+    from langchain_core.output_parsers import StrOutputParser
+    from langchain_core.runnables import RunnablePassthrough
+    from langchain_core.prompts.prompt import PromptTemplate
+    from PIL import Image
+    from typing import List
+    import time
+    import logging
+    import sys
+    logging.basicConfig(stream=sys.stdout, level=logging.INFO)
+    import utils.func
+    from utils.func import vector_db_index
+    import utils.constants as const
+    chroma_db = chromadb.PersistentClient(path="./chromadb")
+    # App title
+    st.set_page_config(page_title="everyware copilot", menu_items=None)
+    AVATAR_AI   = Image.open('./images/ecp_agent.png')
+    AVATAR_USER = Image.open('./images/ecp_user.png')
+    ETH_LOGO = Image.open('./images/logo.png')
+    ECP_LOGO = Image.open('./images/ecp_logo.png')
+    SYSTEM_PROMPT=(
+    """You are an expert Q&A assistant, working as an employee in Eurotech, answering questions on Eurotech products.
+    Never offend or attack or use bad words against Eurotech.
+    Always answer the query using the provided context and not prior knowledge.
+    Some rules to follow:
+. Provide specific answers and enumerate in bullet points when appropriate
+. Never directly reference the given context in your answer.
+. DO NOT start the response with the statement 'Based on the provided context' or something similar.
+    If you don't know the answer, just say that you don't know the answer, don't try to make up an answer.
+    Always answer the user question below based on provided context.
+    Context: {context}
+    Question: {question}
+    """
+    )
+    def find_saved_indexes() -> List[vector_db_index]:
+        chroma_collections = chroma_db.list_collections()
+        milvus_collections = utils.func.list_files(const.INDEX_ROOT_PATH, ".db")
+        result = []
+        for coll in chroma_collections:
+            result.append(vector_db_index(engine= 0, name= coll.name))
+        for coll in milvus_collections:
+            result.append(vector_db_index(engine= 1, name= coll))
+        return result
+    def load_index(index: vector_db_index):
+        use_gpu = os.environ.get("USE_GPU", "").lower() in ("yes", "true", "t", "1")
+        device = "gpu" if use_gpu else "cpu"
+        st.session_state.embed_model = GPT4AllEmbeddings(model_name='all-MiniLM-L6-v2.gguf2.f16.gguf', device=device, gpt4all_kwargs={'allow_download': 'True'})
+        if(index.engine == 0):
+            index = Chroma(
+                client=chroma_db,
+                collection_name=index.name,
+                embedding_function=st.session_state.embed_model,
+            )
+            return index
+        if(index.engine == 1):
+            index = Milvus(
+                st.session_state.embed_model,
+                connection_args={"uri": os.path.join(const.INDEX_ROOT_PATH, index.name+".mvdb")},
+            )
+            return index
+        return None
+    def format_model_name(model):
+        return model[0]
+    def reload_index():
+        logging.info(f"> selected_index = {st.session_state.selected_index}")
+        logging.info(f"> old_selected_index = {st.session_state.old_selected_index}")
+        if st.session_state.old_selected_index != st.session_state.selected_index:
+            st.session_state.old_selected_index = st.session_state.selected_index
+            logging.info(f"> replaced old_selected_index = {st.session_state.old_selected_index}")
+            if st.session_state.selected_index != None:
+                with st.spinner('Loading Index...'):
+                    st.session_state.index = load_index(st.session_state.selected_index)
+                    logging.info(f" ### Loading Index '{st.session_state.selected_index}' completed.")
+    def clear_history():
+        st.session_state.messages = [
+            {"role": "assistant", "content": "Ask me any question about Eurotech!", "avatar": AVATAR_AI}
+        ]
+        if st.session_state.index != None:
+            logging.info("> clearing chat context")
+            #st.session_state.chat_engine.reset()
+    def index_name_format(value: vector_db_index):
+        if value.engine == 0:
+            return "ChromaDB - "+value.name
+        else:
+            return "Milvus - "+value.name
+    if "old_selected_index" not in st.session_state.keys():
+        st.session_state.old_selected_index = ''
+    # Side bar
+    with st.sidebar:
+        st.logo(ETH_LOGO)
+        st.image(ECP_LOGO, width=300)
+        st.subheader('Your local AI assistant')
+        st.html(body="<hr style='border: none; height: 4px; background-color: #410099;'\>")
+        engine = st.radio("LLM engine:", ["GPT4All"])
+        if engine == "GPT4All":
+            models_dir = os.path.join("./gpt4all_models")
+            models = os.listdir(models_dir)
+            check_list = [model for model in models]
+            if not check_list:
+                with st.spinner('Downloaing Nous-Hermes 2 model ...'):
+                    GPT4All(model=os.path.join(models_dir, 'Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf'), allow_download=True)
+                    logging.info(" ### Downloaing Nous-Hermes 2 completed.")
+            st.session_state["model"] = st.selectbox("Choose your LLM", models, index=models.index(check_list[0]))
+            logging.info(f"> GPT4All model = {st.session_state.model}")
+            st.page_link("pages/download_model.py", label=" Download a new LLM", icon="➕")
+            use_gpu = os.environ.get("USE_GPU", "").lower() in ("yes", "true", "t", "1")
+            device = "gpu" if use_gpu else "cpu"
+            logging.info(f"> currently using: {device}")
+            num_threads = os.cpu_count()
+            llm = GPT4All(
+                model=os.path.join(models_dir, st.session_state.model),
+                device=device,
+                n_predict=512,
+                n_threads=num_threads,
+                allow_download=False,
+            )
+        use_index = st.toggle("Use RAG", value=True)
+        if use_index:
+            saved_index_list = find_saved_indexes()
+            index = next((i for i, item in enumerate(saved_index_list) if item.name.startswith('_')), None)
+            st.session_state.selected_index = st.selectbox("Index", saved_index_list, index, format_func=index_name_format)
+            reload_index()
+            logging.info(f"> selected_index = {st.session_state.selected_index}")
+            st.page_link("pages/build_index.py", label=" Build a new index", icon="➕")
+            st.session_state.top_k = 2
+            use_custom_params = st.toggle("Customize retrieval options", value=False)
+            if use_custom_params:
+                st.session_state.top_k = st.slider("top k", 1, 10, 2, key='my_top_k_size')
+                logging.info(f"> Top_K    = {st.session_state.top_k}")
+            if st.session_state.selected_index != None and st.session_state.index != None:
+                prompt = PromptTemplate.from_template(template=SYSTEM_PROMPT)
+                st.session_state.llm_chain = (
+                    {
+                        "context": RunnablePassthrough(),
+                        "question": RunnablePassthrough(),
+                    }
+                    | prompt
+                    | llm
+                    | StrOutputParser()
+                )
+    # initialize history
+    if "index" not in st.session_state.keys():
+        st.session_state.index = None
+    if "messages" not in st.session_state.keys():
+        clear_history()
+    if "old_selected_index" not in st.session_state.keys():
+        st.session_state.old_selected_index = ''
+    def model_res_generator(question=""):
+        start = time.time()
+        if use_index:
+            if st.session_state.selected_index == None:
+                st.warning('No index selected!', icon="⚠️")
+                return
+            logging.info(f">>> RAG enabled:")
+            context = st.session_state.index.similarity_search(query=question, k=st.session_state.top_k)
+            for doc in context:
+                if len(doc.page_content) > 512:
+                    doc.page_content = doc.page_content[:512] + "..."
+            response = st.session_state.llm_chain.invoke({"context": context, "question": question})
+            yield(response)
+            end = time.time()
+            logging.info(f"Time to generate the response: {end-start}")
+        else:
+            logging.info(f">>> Just LLM (no RAG):")
+            messages_only_role_and_content = []
+            for message in st.session_state.messages:
+                if message["role"] == "assistant":
+                    messages_only_role_and_content.append(SystemMessage(content=message["content"]))
+                else:
+                    messages_only_role_and_content.append(HumanMessage(content=message["content"]))
+            chat_response = llm.invoke(
+                messages_only_role_and_content,
+            )
+            yield chat_response
+            end = time.time()
+            logging.info(f"Time to generate the response: {end-start}")
+    # Display chat messages from history on app rerun
+    for message in st.session_state.messages:
+        with st.chat_message(message["role"], avatar=message["avatar"]):
+            st.markdown(message["content"])
+    if prompt := st.chat_input("Enter prompt here..."):
+        # add latest message to history in format {role, content}
+        st.session_state.messages.append({"role": "user", "content": prompt, "avatar": AVATAR_USER})
+        with st.chat_message("user", avatar=AVATAR_USER):
+            st.markdown(prompt)
+    # if st.session_state.messages[-1]["role"] != "assistant":
+        with st.chat_message("assistant", avatar=AVATAR_AI):
+            with st.spinner("Thinking..."):
+                time.sleep(1)
+                message = st.write(model_res_generator(prompt))
+                st.button("clear conversation context", on_click=clear_history)
+                st.session_state.messages.append({"role": "assistant", "content": message, "avatar": AVATAR_AI})

gpt4all-project/streamlit_app/images/ecp_agent.png

Sorry, something went wrong. Reload?