init

README.md

@@ -1,8 +1,8 @@
 ---
-title: Examin8
-emoji: 💻
-colorFrom: indigo
-colorTo: pink
+title: Examinate
+emoji: 🌖
+colorFrom: blue
+colorTo: purple
 sdk: gradio
 sdk_version: 3.16.2
 app_file: app.py

app.py

@@ -0,0 +1,99 @@
+import datetime
+import os
+
+import gradio as gr
+import langchain
+import weaviate
+from langchain.vectorstores import Weaviate
+
+from chain import get_new_chain1
+
+WEAVIATE_URL = os.environ["WEAVIATE_URL"]
+
+
+def get_weaviate_store():
+    client = weaviate.Client(
+        url=WEAVIATE_URL,
+        additional_headers={"X-OpenAI-Api-Key": os.environ["OPENAI_API_KEY"]},
+    )
+    return Weaviate(client, "Paragraph", "content", attributes=["source"])
+
+
+def set_openai_api_key(api_key, agent):
+    if api_key:
+        os.environ["OPENAI_API_KEY"] = api_key
+        vectorstore = get_weaviate_store()
+        qa_chain = get_new_chain1(vectorstore)
+        os.environ["OPENAI_API_KEY"] = ""
+        return qa_chain
+
+
+def chat(inp, history, agent):
+    history = history or []
+    if agent is None:
+        history.append((inp, "Please paste your OpenAI key to use"))
+        return history, history
+    print("\n==== date/time: " + str(datetime.datetime.now()) + " ====")
+    print("inp: " + inp)
+    history = history or []
+    output = agent({"question": inp, "chat_history": history})
+    answer = output["answer"]
+    history.append((inp, answer))
+    print(history)
+    return history, history
+
+
+block = gr.Blocks(css=".gradio-container {background-color: lightblue}")
+
+with block:
+    with gr.Row():
+        gr.Markdown("<h3><center>Elenchos AI</center></h3>")
+
+        openai_api_key_textbox = gr.Textbox(
+            placeholder="Paste your OpenAI API key (sk-...)",
+            show_label=False,
+            lines=1,
+            type="password",
+        )
+
+    chatbot = gr.Chatbot()
+
+    with gr.Row():
+        message = gr.Textbox(
+            label="What would you like to learn about marine biology?",
+            placeholder="What is an estuary?",
+            lines=1,
+        )
+        submit = gr.Button(value="Send", variant="secondary").style(full_width=False)
+
+    gr.Examples(
+        examples=[
+            "What are phytoplankton?",
+            "How does microplastic pollution affect the oceans?",
+            "What are artificial reefs?",
+        ],
+        inputs=message,
+    )
+
+    gr.HTML(
+        """
+    This simple application is an implementation of ChatGPT but over an external dataset (in this case, Wikipedia pages on Marine biology)."""
+    )
+
+    gr.HTML(
+        "<center>Powered by <a href='https://github.com/hwchase17/langchain'>LangChain 🦜️🔗</a></center>"
+    )
+
+    state = gr.State()
+    agent_state = gr.State()
+
+    submit.click(chat, inputs=[message, state, agent_state], outputs=[chatbot, state])
+    message.submit(chat, inputs=[message, state, agent_state], outputs=[chatbot, state])
+
+    openai_api_key_textbox.change(
+        set_openai_api_key,
+        inputs=[openai_api_key_textbox, agent_state],
+        outputs=[agent_state],
+    )
+
+block.launch(debug=True)

chain.py

@@ -0,0 +1,127 @@
+import json
+import os
+import pathlib
+from typing import Dict, List, Tuple
+
+import weaviate
+from langchain import OpenAI, PromptTemplate
+from langchain.chains import LLMChain
+from langchain.chains.base import Chain
+from langchain.chains.combine_documents.base import BaseCombineDocumentsChain
+from langchain.chains.conversation.memory import ConversationBufferMemory
+from langchain.chains.question_answering import load_qa_chain
+from langchain.embeddings import OpenAIEmbeddings
+from langchain.prompts import FewShotPromptTemplate, PromptTemplate
+from langchain.prompts.example_selector import \
+    SemanticSimilarityExampleSelector
+from langchain.vectorstores import FAISS, Weaviate
+from pydantic import BaseModel
+
+
+class CustomChain(Chain, BaseModel):
+
+    vstore: Weaviate
+    chain: BaseCombineDocumentsChain
+    key_word_extractor: Chain
+
+    @property
+    def input_keys(self) -> List[str]:
+        return ["question"]
+
+    @property
+    def output_keys(self) -> List[str]:
+        return ["answer"]
+
+    def _call(self, inputs: Dict[str, str]) -> Dict[str, str]:
+        question = inputs["question"]
+        chat_history_str = _get_chat_history(inputs["chat_history"])
+        if chat_history_str:
+            new_question = self.key_word_extractor.run(
+                question=question, chat_history=chat_history_str
+            )
+        else:
+            new_question = question
+        print(new_question)
+        docs = self.vstore.similarity_search(new_question, k=4)
+        new_inputs = inputs.copy()
+        new_inputs["question"] = new_question
+        new_inputs["chat_history"] = chat_history_str
+        answer, _ = self.chain.combine_docs(docs, **new_inputs)
+        return {"answer": answer}
+
+
+def get_new_chain1(vectorstore) -> Chain:
+    WEAVIATE_URL = os.environ["WEAVIATE_URL"]
+    client = weaviate.Client(
+        url=WEAVIATE_URL,
+        additional_headers={"X-OpenAI-Api-Key": os.environ["OPENAI_API_KEY"]},
+    )
+
+    _eg_template = """## Example:
+
+    Chat History:
+    {chat_history}
+    Follow Up Input: {question}
+    Standalone question: {answer}"""
+    _eg_prompt = PromptTemplate(
+        template=_eg_template,
+        input_variables=["chat_history", "question", "answer"],
+    )
+
+    _prefix = """Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question. You should assume that the question is related to marine biology."""
+    _suffix = """## Example:
+
+    Chat History:
+    {chat_history}
+    Follow Up Input: {question}
+    Standalone question:"""
+    eg_store = Weaviate(
+        client,
+        "Rephrase",
+        "content",
+        attributes=["question", "answer", "chat_history"],
+    )
+    example_selector = SemanticSimilarityExampleSelector(vectorstore=eg_store, k=4)
+    prompt = FewShotPromptTemplate(
+        prefix=_prefix,
+        suffix=_suffix,
+        example_selector=example_selector,
+        example_prompt=_eg_prompt,
+        input_variables=["question", "chat_history"],
+    )
+    llm = OpenAI(temperature=0, model_name="text-davinci-003")
+    key_word_extractor = LLMChain(llm=llm, prompt=prompt)
+
+    EXAMPLE_PROMPT = PromptTemplate(
+        template=">Example:\nContent:\n---------\n{page_content}\n----------\nSource: {source}",
+        input_variables=["page_content", "source"],
+    )
+    template = """You are an AI assistant for Wikipedia information about marine biology.
+You are given the following extracted parts of a long document and a question. Provide a conversational answer with a hyperlink to the wikipedia page.
+You should only use hyperlinks that are explicitly listed as a source in the context. Do NOT make up a hyperlink that is not listed.
+If you don't know the answer, just say "Hmm, I'm not sure." Don't try to make up an answer.
+If the question is not about marine biology, the oceans, or biology, politely inform them that you are tuned to only answer questions about marine biology.
+Question: {question}
+=========
+{context}
+=========
+Answer in Markdown:"""
+    PROMPT = PromptTemplate(template=template, input_variables=["question", "context"])
+    doc_chain = load_qa_chain(
+        OpenAI(temperature=0, model_name="text-davinci-003", max_tokens=-1),
+        chain_type="stuff",
+        prompt=PROMPT,
+        document_prompt=EXAMPLE_PROMPT,
+    )
+    return CustomChain(
+        chain=doc_chain, vstore=vectorstore, key_word_extractor=key_word_extractor
+    )
+
+
+def _get_chat_history(chat_history: List[Tuple[str, str]]):
+    buffer = ""
+    for human_s, ai_s in chat_history:
+        human = f"Human: " + human_s
+        ai = f"Assistant: " + ai_s
+        buffer += "\n" + "\n".join([human, ai])
+    return buffer

requirements.txt

@@ -0,0 +1,15 @@
+langchain==0.0.64
+beautifulsoup4
+weaviate-client
+openai
+black
+isort
+Flask
+transformers
+gradio
+wikipedia
+gpt-index
+requests==2.28.2
+boto3
+pygit2
+better_profanity