functioning version

langchain-streamlit-demo/app.py

@@ -6,7 +6,6 @@ import langsmith.utils
 import openai
 import streamlit as st
 from defaults import default_values
-from langchain.agents import load_tools
 from langchain.agents.tools import tool
 from langchain.callbacks.base import BaseCallbackHandler
 from langchain.callbacks.manager import Callbacks
@@ -17,11 +16,9 @@ from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
 from langchain.schema.document import Document
 from langchain.schema.retriever import BaseRetriever
 from langchain_community.callbacks import StreamlitCallbackHandler
-from langchain_community.tools import DuckDuckGoSearchRun, WikipediaQueryRun
-from langchain_community.utilities import WikipediaAPIWrapper
 from langsmith.client import Client
 from llm_resources import (
-    get_agent,
+    # get_agent,
     get_doc_agent,
     get_llm,
     get_runnable,
@@ -168,7 +165,7 @@ with sidebar:
         index=default_values.SUPPORTED_MODELS.index(default_values.DEFAULT_MODEL),
     )
 
-    st.session_state.provider = default_values.MODEL_DICT[model]
+    st.session_state.provider = "OpenAI"  # default_values.MODEL_DICT[model]
 
     provider_api_key = (
         default_values.PROVIDER_KEY_DICT.get(
@@ -198,11 +195,12 @@ with sidebar:
             or st.sidebar.text_input("OpenAI API Key: ", type="password")
         )
 
-        document_chat = st.checkbox(
-            "Document Chat",
-            value=True if uploaded_file else False,
-            help="Uploaded document will provide context for the chat.",
-        )
+        document_chat = True
+        # st.checkbox(
+        #     "Document Chat",
+        #     value=True if uploaded_file else False,
+        #     help="Uploaded document will provide context for the chat.",
+        # )
 
         k = st.slider(
             label="Number of Chunks",
@@ -285,7 +283,7 @@ with sidebar:
     # --- Advanced Settings ---
     with st.expander("Advanced Settings", expanded=False):
         st.markdown("## Feedback Scale")
-        use_faces = st.toggle(label="`Thumbs` ⇄ `Faces`", value=False)
+        use_faces = False  # st.toggle(label="`Thumbs` ⇄ `Faces`", value=False)
         feedback_option = "faces" if use_faces else "thumbs"
 
         system_prompt = (
@@ -447,23 +445,23 @@ if st.session_state.llm:
                     config["max_concurrency"] = 5
                 return config
 
-            use_document_chat = all(
-                [
-                    document_chat,
-                    st.session_state.retriever,
-                ],
-            )
+            # use_document_chat = all(
+            #     [
+            #         document_chat,
+            #         st.session_state.retriever,
+            #     ],
+            # )
 
             full_response: Union[str, None] = None
             # stream_handler = StreamHandler(message_placeholder)
             # callbacks.append(stream_handler)
             message_placeholder = st.empty()
             default_tools = [
-                DuckDuckGoSearchRun(),
-                WikipediaQueryRun(api_wrapper=WikipediaAPIWrapper()),
+                # DuckDuckGoSearchRun(),
+                # WikipediaQueryRun(api_wrapper=WikipediaAPIWrapper()),
             ]
-            default_tools += load_tools(["requests_get"])
-            default_tools += load_tools(["llm-math"], llm=st.session_state.llm)
+            # default_tools += load_tools(["requests_get"])
+            # default_tools += load_tools(["llm-math"], llm=st.session_state.llm)
             if st.session_state.provider in ("Azure OpenAI", "OpenAI"):
                 # research_assistant_chain = get_research_assistant_chain(
                 #     search_llm=get_llm(**get_llm_args_temp_zero),  # type: ignore
@@ -499,52 +497,52 @@ if st.session_state.llm:
                 TOOLS = default_tools
                 # [research_assistant_tool, python_coder_tool] + default_tools
 
-                if use_document_chat:
-                    st.session_state.doc_chain = get_runnable(
-                        use_document_chat,
-                        document_chat_chain_type,
-                        st.session_state.llm,
-                        st.session_state.retriever,
-                        MEMORY,
-                        chat_prompt,
-                        prompt,
+                # if use_document_chat:
+                st.session_state.doc_chain = get_runnable(
+                    True,  # use_document_chat,
+                    document_chat_chain_type,
+                    st.session_state.llm,
+                    st.session_state.retriever,
+                    MEMORY,
+                    chat_prompt,
+                    prompt,
+                )
+
+                @tool("user-document-chat")
+                def doc_chain_tool(input_str: str, callbacks: Callbacks = None):
+                    """Always use this tool at least once. Input should be a question."""
+                    return st.session_state.doc_chain.invoke(
+                        input_str,
+                        config=get_config(callbacks),
                     )
 
-                    @tool("user-document-chat")
-                    def doc_chain_tool(input_str: str, callbacks: Callbacks = None):
-                        """Always use this tool at least once. Input should be a question."""
-                        return st.session_state.doc_chain.invoke(
-                            input_str,
-                            config=get_config(callbacks),
-                        )
+                doc_chain_agent = get_doc_agent(
+                    [doc_chain_tool],
+                )
 
-                    doc_chain_agent = get_doc_agent(
-                        [doc_chain_tool],
+                @tool("document-question-tool")
+                def doc_question_tool(input_str: str, callbacks: Callbacks = None):
+                    """This tool is an AI assistant with access to the user's uploaded document.
+                    Input should be one or more questions, requests, instructions, etc.
+                    If the user's meaning is unclear, perhaps the answer is here.
+                    Generally speaking, try this tool before conducting web research.
+                    """
+                    return doc_chain_agent.invoke(
+                        input_str,
+                        config=get_config(callbacks),
                     )
 
-                    @tool("document-question-tool")
-                    def doc_question_tool(input_str: str, callbacks: Callbacks = None):
-                        """This tool is an AI assistant with access to the user's uploaded document.
-                        Input should be one or more questions, requests, instructions, etc.
-                        If the user's meaning is unclear, perhaps the answer is here.
-                        Generally speaking, try this tool before conducting web research.
-                        """
-                        return doc_chain_agent.invoke(
-                            input_str,
-                            config=get_config(callbacks),
-                        )
-
-                    TOOLS = [doc_question_tool] + TOOLS
-
-                st.session_state.chain = get_agent(
-                    TOOLS,
-                    STMEMORY,
-                    st.session_state.llm,
-                    callbacks,
-                )
-            else:
+                TOOLS = [doc_question_tool] + TOOLS
+
+                # st.session_state.chain = get_agent(
+                #     TOOLS,
+                #     STMEMORY,
+                #     st.session_state.llm,
+                #     callbacks,
+                # )
+                # else:
                 st.session_state.chain = get_runnable(
-                    use_document_chat,
+                    True,  # use_document_chat,
                     document_chat_chain_type,
                     st.session_state.llm,
                     st.session_state.retriever,

langchain-streamlit-demo/defaults.py

@@ -6,17 +6,17 @@ MODEL_DICT = {
     "gpt-4": "OpenAI",
     "gpt-4-turbo-preview": "OpenAI",
     "gpt-4-1106-preview": "OpenAI",
-    "claude-instant-1": "Anthropic",
-    "claude-instant-1.2": "Anthropic",
-    "claude-2": "Anthropic",
-    "claude-2.1": "Anthropic",
-    "meta-llama/Llama-2-7b-chat-hf": "Anyscale Endpoints",
-    "meta-llama/Llama-2-13b-chat-hf": "Anyscale Endpoints",
-    "meta-llama/Llama-2-70b-chat-hf": "Anyscale Endpoints",
-    "codellama/CodeLlama-34b-Instruct-hf": "Anyscale Endpoints",
-    "mistralai/Mistral-7B-Instruct-v0.1": "Anyscale Endpoints",
-    "HuggingFaceH4/zephyr-7b-beta": "Anyscale Endpoints",
-    "Azure OpenAI": "Azure OpenAI",
+    # "claude-instant-1": "Anthropic",
+    # "claude-instant-1.2": "Anthropic",
+    # "claude-2": "Anthropic",
+    # "claude-2.1": "Anthropic",
+    # "meta-llama/Llama-2-7b-chat-hf": "Anyscale Endpoints",
+    # "meta-llama/Llama-2-13b-chat-hf": "Anyscale Endpoints",
+    # "meta-llama/Llama-2-70b-chat-hf": "Anyscale Endpoints",
+    # "codellama/CodeLlama-34b-Instruct-hf": "Anyscale Endpoints",
+    # "mistralai/Mistral-7B-Instruct-v0.1": "Anyscale Endpoints",
+    # "HuggingFaceH4/zephyr-7b-beta": "Anyscale Endpoints",
+    # "Azure OpenAI": "Azure OpenAI",
 }
 
 SUPPORTED_MODELS = list(MODEL_DICT.keys())
@@ -28,7 +28,7 @@ DEFAULT_SYSTEM_PROMPT = os.environ.get(
     # "You are a helpful chatbot. Do not rush. Always plan, think, and act in a step-by-step manner.",
     """
     Comportati come un operatore di call center. 
-    Hai accesso a tutte le informazioni necessarie per rispondere alle domande dei clienti tramite il vectorstore.
+    Hai accesso a tutte le informazioni necessarie per rispondere alle domande dei clienti tramite il tool 'document-question-tool'.
     Se non sai rispondere ad una domanda, rispondi che non hai la risposta e offri di connettere il cliente con un operatore umano.
     Rispondi in italiano, usando uno stile amichevole ma formale.
     """,
@@ -36,11 +36,11 @@ DEFAULT_SYSTEM_PROMPT = os.environ.get(
 
 MIN_TEMP = float(os.environ.get("MIN_TEMPERATURE", 0.0))
 MAX_TEMP = float(os.environ.get("MAX_TEMPERATURE", 1.0))
-DEFAULT_TEMP = float(os.environ.get("DEFAULT_TEMPERATURE", 0.7))
+DEFAULT_TEMP = float(os.environ.get("DEFAULT_TEMPERATURE", 0.0))
 
 MIN_MAX_TOKENS = int(os.environ.get("MIN_MAX_TOKENS", 1))
-MAX_MAX_TOKENS = int(os.environ.get("MAX_MAX_TOKENS", 100000))
-DEFAULT_MAX_TOKENS = int(os.environ.get("DEFAULT_MAX_TOKENS", 1000))
+MAX_MAX_TOKENS = int(os.environ.get("MAX_MAX_TOKENS", 1000))
+DEFAULT_MAX_TOKENS = int(os.environ.get("DEFAULT_MAX_TOKENS", 300))
 
 DEFAULT_LANGSMITH_PROJECT = os.environ.get("LANGCHAIN_PROJECT")
 

langchain-streamlit-demo/llm_resources.py

@@ -6,7 +6,7 @@ from defaults import DEFAULT_CHUNK_OVERLAP, DEFAULT_CHUNK_SIZE, DEFAULT_RETRIEVE
 from langchain.agents import AgentExecutor, AgentType, initialize_agent
 from langchain.agents.openai_functions_agent.base import OpenAIFunctionsAgent
 from langchain.callbacks.base import BaseCallbackHandler
-from langchain.chains import LLMChain, RetrievalQA
+from langchain.chains import RetrievalQA
 from langchain.chat_models import (
     AzureChatOpenAI,
     ChatAnthropic,
@@ -14,7 +14,7 @@ from langchain.chat_models import (
     ChatOpenAI,
 )
 from langchain.document_loaders import PyPDFLoader
-from langchain.embeddings import AzureOpenAIEmbeddings, OpenAIEmbeddings
+from langchain.embeddings import OpenAIEmbeddings
 from langchain.llms.base import BaseLLM
 from langchain.memory import ConversationBufferMemory
 from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
@@ -132,31 +132,38 @@ def get_runnable(
     chat_prompt,
     summarization_prompt,
 ):
-    if not use_document_chat:
-        return LLMChain(
-            prompt=chat_prompt,
-            llm=llm,
-            memory=memory,
-        ) | (lambda output: output["text"])
+    # if not use_document_chat:
+    #     return LLMChain(
+    #         prompt=chat_prompt,
+    #         llm=llm,
+    #         memory=memory,
+    #     ) | (lambda output: output["text"])
 
-    # if document_chat_chain_type == "Q&A Generation":
-    #     return get_rag_qa_gen_chain(
-    #         retriever,
-    #         llm,
-    #     )
-    # elif document_chat_chain_type == "Summarization":
-    #     return get_rag_summarization_chain(
-    #         summarization_prompt,
-    #         retriever,
-    #         llm,
-    #     )
-    else:
-        return RetrievalQA.from_chain_type(
-            llm=llm,
-            chain_type=document_chat_chain_type,
-            retriever=retriever,
-            output_key="output_text",
-        ) | (lambda output: output["output_text"])
+    # # if document_chat_chain_type == "Q&A Generation":
+    # #     return get_rag_qa_gen_chain(
+    # #         retriever,
+    # #         llm,
+    # #     )
+    # # elif document_chat_chain_type == "Summarization":
+    # #     return get_rag_summarization_chain(
+    # #         summarization_prompt,
+    # #         retriever,
+    # #         llm,
+    # #     )
+    # else:
+    return RetrievalQA.from_chain_type(
+        llm=llm,
+        chain_type=document_chat_chain_type,
+        retriever=retriever,
+        output_key="output_text",
+        return_source_documents=True,
+    ) | (
+        lambda output: output["output_text"]
+        + "\n\n"
+        + "Source Documents:"
+        + "\n"
+        + "\n".join([x for x in output["source_documents"].page_content])
+    )
 
 
 def get_llm(
@@ -246,12 +253,12 @@ def get_texts_and_multiretriever(
             sub_texts.extend(_sub_texts)
 
         embeddings_kwargs = {"openai_api_key": openai_api_key}
-        if use_azure and azure_kwargs:
-            azure_kwargs["azure_endpoint"] = azure_kwargs.pop("openai_api_base")
-            embeddings_kwargs.update(azure_kwargs)
-            embeddings = AzureOpenAIEmbeddings(**embeddings_kwargs)
-        else:
-            embeddings = OpenAIEmbeddings(**embeddings_kwargs)
+        # if use_azure and azure_kwargs:
+        #     azure_kwargs["azure_endpoint"] = azure_kwargs.pop("openai_api_base")
+        #     embeddings_kwargs.update(azure_kwargs)
+        #     embeddings = AzureOpenAIEmbeddings(**embeddings_kwargs)
+        # else:
+        embeddings = OpenAIEmbeddings(**embeddings_kwargs)
         store = InMemoryStore()
 
         # MultiVectorRetriever