Upload 8 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+indexs/text-embedding-ada-002/index.faiss filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,99 @@
+import os
+import time
+
+import gradio as gr
+import pandas as pd
+from model import Model
+from tqdm import tqdm
+
+tqdm.pandas()
+
+OUTPUT_FILE = "./results_qa.csv"
+
+def new_vote(data: gr.LikeData, question, model_name, **kwargs):
+    feedback = "Good" if data.liked else "Bad"
+    df = pd.read_csv(OUTPUT_FILE)
+    df['Feedback'] = df.apply(lambda x: feedback if (x.Model == model_name and x.Question == question) else None, axis = 1)
+    df.to_csv(OUTPUT_FILE, index=False)
+    
+# def answer_question(question: str, model_name: str, system_prompt: str):
+#     start_time = time.time()
+#     qa_model = Model(model_name=model_name)
+#     response, sources = qa_model.run(system_prompt=system_prompt, query=question)
+#     time_taken = time.time() - start_time
+#     words = len(question) + len(response)
+#     efficiency = words / time_taken
+#     final_response = f"{response} \n\nTime Taken: {time_taken}"
+#     new_row = {'Model': model_name, 'Question': question, 'Answer': response, "Sources": sources, "Time": time_taken, "Words": words, "Efficiency": efficiency, "Feedback": None, "final_response": final_response}
+#     if os.path.isfile(OUTPUT_FILE):
+#         df = pd.read_csv(OUTPUT_FILE)
+#         rows = df.values.tolist()
+#         # print("df.values.tolist(): ", df.values.tolist())
+#         # df = df.append(new_row, ignore_index=True)
+#         rows.append(new_row)
+#     else:
+#         rows = [new_row]
+#     df = pd.DataFrame(rows)
+#     df.to_csv(OUTPUT_FILE, index=False)
+#     yield [(question, final_response)]
+
+def answer_question(question: str, model_name: str, system_prompt: str):
+    start_time = time.time()
+    qa_model = Model(model_name=model_name)
+    gen_response = qa_model.run(system_prompt=system_prompt, query=question)
+    response = ""
+    for resp in gen_response:
+        if isinstance(resp, list):
+            sources = resp
+            break
+        resp = resp.replace("$", "₹") 
+        response += resp
+        yield [(question, response)], OUTPUT_FILE
+        
+    time_taken = time.time() - start_time
+    words = len(question) + len(response)
+    efficiency = words / time_taken
+    temp_sources = "\n".join([f"{i + 1}. {d}" for i, d in enumerate(sources)])
+    final_response = f"{response} \n\nSources: \n{temp_sources} \n\nTime Taken: {time_taken}"
+    new_row = {'Model': model_name, 'Question': question, 'Answer': response, "Sources": sources, "Time": time_taken, "Words": words, "Efficiency": efficiency, "Feedback": None, "final_response": final_response}
+    if os.path.isfile(OUTPUT_FILE):
+        try:
+            df = pd.read_csv(OUTPUT_FILE)
+            rows = df.to_dict(orient="records")
+            rows.append(new_row)
+        except Exception:
+            rows = [new_row]
+    else:
+        rows = [new_row]
+    
+    df = pd.DataFrame(rows)
+    df.to_csv(OUTPUT_FILE, index=False)
+    final_response = final_response.strip("Question").strip("\n")
+    final_response = final_response.strip("\n").strip(" ").strip("Answer:").strip("Question").strip("\n").replace("Answer:", "")
+    yield [(question, final_response)], OUTPUT_FILE
+    
+    
+
+if __name__ == "__main__":
+    with gr.Blocks() as demo:
+        chatbot = gr.Chatbot()
+        
+        # with gr.Row():
+        
+        textbox = gr.Textbox(label="Query")
+        # system_prompt = """Answer the question using the context. Provide examples only from the context and use only Rupees (₹) in examples. If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+        system_prompt = """"Answer the question using the context. Provide examples only from the context and use only Rupees (₹) in examples. If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+        system_prompt = "Use the following pieces of book to answer the question at the end. \nIf you don't know the answer, please think rationally and answer from the book"
+        system_prompt = """Answer the question using the context. Provide examples only from the context and use only Rupees (₹) in examples. If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+        system_prompt = """Answer the question from the book. Provide examples only from the book. If you don't know the answer, just say 'Please rephrase the question'"""
+        
+        choices=["gpt4", "gpt-3.5-turbo"]
+        
+        system_prompt = gr.Textbox(value=system_prompt, label="System Prompt")
+        model_name = gr.Dropdown(choices=choices, value="gpt-3.5-turbo", label="Model")
+        file = gr.File(value = OUTPUT_FILE, file_types=["csv"], label="Output")
+        textbox.submit(answer_question, [textbox, model_name, system_prompt], [chatbot, file])
+        chatbot.like(new_vote, [textbox, model_name], None)    
+    
+    demo.queue()
+    demo.launch(share=True)

embedder.py

@@ -0,0 +1,80 @@
+from typing import List
+
+import requests
+from langchain.pydantic_v1 import BaseModel
+from langchain.schema.embeddings import Embeddings
+from retry import retry
+from tqdm import tqdm
+
+
+# @dataclass    
+class CustomEmbeddings(BaseModel, Embeddings):
+    """Wrapper around OpenAI embedding models.
+
+    To use, you should have the ``openai`` python package installed, and the
+    environment variable ``OPENAI_API_KEY`` set with your API key or pass it
+    as a named parameter to the constructor.
+
+    Example:
+        .. code-block:: python
+
+            from langchain.embeddings import OpenAIEmbeddings
+            openai = OpenAIEmbeddings(model_name="davinci", openai_api_key="my-api-key")
+    """
+    model: str = ""
+    model_url: str = ""
+    api_key: str = "EMPTY"
+    # engine: str = None
+    # api_type: str = None
+
+    def _embedding_func(self, text: str) -> List[float]:
+        """Call out to OpenAI's embedding endpoint."""
+        # replace newlines, which can negatively affect performance.
+        text = text.replace("\n", " ")
+        result = self.api_call(input_text=text)
+        return result['data'][0]['embedding']
+    
+    @retry(tries=3, delay=2, backoff=2, exceptions=(requests.RequestException,))
+    def api_call(self, input_text: str):
+        data = {
+                "input": input_text,
+                "model": self.model
+            }
+            
+        response = requests.post(
+            self.model_url, 
+            headers={
+                "Content-Type": "application/json",
+                # "Authorization": f"Bearer {self.api_key}",
+                "api-key": self.api_key
+            }, 
+            json=data
+        )
+
+        if response.status_code == 200:
+            return response.json()
+        else:
+            response.raise_for_status()
+
+    def embed_documents(self, texts: List[str]) -> List[List[float]]:
+        """Call out to OpenAI's embedding endpoint for embedding search docs.
+
+        Args:
+            texts: The list of texts to embed.
+
+        Returns:
+            List of embeddings, one for each text.
+        """
+        return [self._embedding_func(text) for text in tqdm(texts)]
+
+    def embed_query(self, text: str) -> List[float]:
+        """Call out to OpenAI's embedding endpoint for embedding query text.
+
+        Args:
+            text: The text to embed.
+
+        Returns:
+            Embeddings for the text.
+        """
+        return self._embedding_func(text)
+

indexs/text-embedding-ada-002/index.faiss

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0d252dde59cab3da1aa892c4c430aadd9ac0bc16b3e27595d6806997690580f
+size 4497453

indexs/text-embedding-ada-002/index.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:586a91e52cce6dd1750160eec565a24222617a8187ae8145899e7abba5b44daf
+size 2602597

model.py

@@ -0,0 +1,205 @@
+import os
+
+os.environ["GOOGLE_API_KEY"] = "AIzaSyAGoYnNPu__70AId7EJS7F_61i69Qmn-wM"
+os.environ["OPENAI_API_TYPE"] = "azure"
+# os.environ["OPENAI_API_VERSION"] = "2023-07-01-preview"
+# # os.environ["OPENAI_API_KEY"] = "5b624f6b71884a488560a86b1fffbf42"
+# os.environ["OPENAI_API_KEY"] = "9e337d6696ce4a22a9a1b901e2ebb5fb"
+
+
+from embedder import CustomEmbeddings
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
+from langchain.prompts.chat import (ChatPromptTemplate,
+                                    HumanMessagePromptTemplate,
+                                    SystemMessagePromptTemplate)
+from langchain_google_genai import ChatGoogleGenerativeAI
+from search import SimilaritySearch
+
+embeddings = CustomEmbeddings(
+    model="text-embedding-ada-002", 
+    model_url="https://year-embedding-ada-002-aiservices-2136192926.openai.azure.com//openai/deployments/fresh-embedding-ada-002/embeddings?api-version=2023-10-01-preview",
+    api_key="6eed3006cdd3445cb3f422a7358ce461"
+)
+vector_store = SimilaritySearch.load_from_disk(
+    embedding_function=embeddings, 
+    data_dir="../indexs/text-embedding-ada-002/"
+    # data_dir="../indexs/basic-fno-text-embedding-ada-002/"
+)
+
+class Model:
+    def __init__(self, model_name: str, **kwargs) -> None:
+        self.model_name = model_name
+        self.llm = self.load_llm(model_name=model_name, **kwargs)
+    
+    def load_llm(self, model_name: str, **kwargs):
+        if self.model_name == "gemini-pro":
+            self.retriever = vector_store.as_retriever(search_kwargs={"k": 2}, search_type="similarity")
+            return ChatGoogleGenerativeAI(model=model_name, temperature=0, max_tokens=4096)
+        elif self.model_name == "gpt-3.5-turbo":
+            self.retriever = vector_store.as_retriever(search_kwargs={"k": 2}, search_type="similarity")
+            return AzureChatOpenAI(
+                deployment_name="latest-gpt-35-turbo-16k", 
+                temperature=0, 
+                max_tokens=4096, 
+                # azure_endpoint="https://high-gpt4-32k-0613-aiservices336365459.openai.azure.com/",
+                openai_api_key="9e337d6696ce4a22a9a1b901e2ebb5fb",
+                # openai_api_base="https://jan-2024-gpt35-turbo16k-aiservices800630185.openai.azure.com/",
+                openai_api_base = "https://fresh-gpt35-turbo-aiservices-2112150452.openai.azure.com/",
+                openai_api_version="2023-07-01-preview"
+            )
+        elif self.model_name == "gpt4":
+            self.retriever = vector_store.as_retriever(search_kwargs={"k": kwargs.get("k", 2)}, search_type="similarity")
+            return AzureChatOpenAI(
+                deployment_name="gpt-4-32k", 
+                temperature=0, 
+                max_tokens=4096, 
+                # azure_endpoint="https://high-gpt4-32k-0613-aiservices336365459.openai.azure.com/",
+                openai_api_key="e91a341abb2f4646ab7b0acd3b9d461e",
+                openai_api_base="https://jan-2024-gpt4-ai-aiservices-1959882301.openai.azure.com/",
+                openai_api_version="2023-07-01-preview"
+            )
+
+        self.retriever = vector_store.as_retriever(search_kwargs={"k": kwargs.get("k", 1)}, search_type="similarity")
+        return ChatOpenAI(
+            model=model_name,
+            openai_api_key="EMPTY",
+            openai_api_base="http://localhost:8000/v1",
+            max_tokens=1024,
+            temperature=0,
+            model_kwargs={"stop": ["<|im_end|>", "Query:", "Question:"], "top_p": 0.95}
+        )
+        
+        
+    def run_qa_result(self, query: str):
+        support_docs = self.retriever.get_relevant_documents(query)
+        sources = list({d.metadata['source'] for d in support_docs})
+        context = "\n\n".join([f"{i + 1}. {d.page_content}" for i, d in enumerate(support_docs)])
+        return context, sources
+
+    def return_prompt(self, system_prompt: str, query: str, context: str):
+        
+        # human_template = "Context:\n\n{context}\n\nQuery: {query}"
+        # human_template = "E-Book:\n\n{context}\n\nQuestion: {query}"
+
+        human_template = "\n\nContext:\n\n{context}\n\nQuestion: {query}"
+        # human_template = "\n\nBook:\n\n{context}\n\nQuestion: {query}"
+
+        messages = []
+        if self.model_name in [
+            "gemini-pro",
+            "TheBloke/Mistral-7B-Instruct-v0.2-AWQ",
+        ]:
+            human_template = system_prompt + "\n\n" + human_template
+            human_message_prompt = HumanMessagePromptTemplate.from_template(human_template)
+            messages.append(human_message_prompt)
+        else:
+            system_message_prompt = SystemMessagePromptTemplate.from_template(system_prompt)
+            human_message_prompt = HumanMessagePromptTemplate.from_template(human_template)
+            messages.extend([system_message_prompt, human_message_prompt])
+
+        chat_prompt = ChatPromptTemplate.from_messages(messages)
+        return chat_prompt.format_prompt(context=context, query=query).to_messages()
+        
+    def run(self, system_prompt: str, query: str):
+        context, sources = self.run_qa_result(query=query)
+        chat_prompt = self.return_prompt(system_prompt=system_prompt, query=query, context=context)
+        # text = "".join(resp.content for resp in self.llm.stream(chat_prompt))
+        # text += "\nSources: \n" + "\n".join([f"{i + 1}. {d}" for i, d in enumerate(sources)])
+        # return text, sources
+        for resp in self.llm.stream(chat_prompt):
+            yield resp.content.replace("$", "₹")
+        
+        yield sources
+        # text = "".join(resp.content for resp in self.llm.stream(chat_prompt))
+        # text += "\nSources: \n" + "\n".join([f"{i + 1}. {d}" for i, d in enumerate(sources)])
+        # return text, sources
+   
+def get_sources(query):
+    results = vector_store.similarity_search_with_relevance_scores(query, k=2)
+    return [
+        {
+            "score": r[-1],
+            "source": r[0].metadata['source']
+        }
+        for r in results
+    ]     
+
+if __name__ == "__main__":
+    # model = Model(model_name="phi2")
+    # model = Model(model_name="gpt-3.5-turbo")       
+    # model = Model(model_name="gemini-pro")    
+    # model = Model(model_name="TheBloke/zephyr-7B-beta-AWQ")
+    # model = Model(model_name="TheBloke/neural-chat-7B-v3-3-AWQ")
+    model = Model(model_name="TheBloke/Mistral-7B-Instruct-v0.2-AWQ")
+    model = Model(model_name="gpt4")
+    model = Model(model_name="gpt-3.5-turbo") 
+    
+    # query = "what is reliance?"
+    # print("results: ", get_sources(query))
+    
+    # query = "explain FNO trading?"
+    # print("results: ", get_sources(query))
+      
+#     query="What is FNO trading?"
+#     query = "Describe ITM, ATM and OTM"
+#     query = "give formula to calculate intrinsic value in Put and provide an example"
+#     query = "what is the order of delta, theta, gamma and vega amongst options in a given options chain"
+#     query = "Explain apple stock and nasdaq"
+    
+#     query = "generate a table with long and short in F&O instruments"
+#     query = "how can we calculate intrinsic value and time value"
+#     query = "give formula to calculate intrinsic value in Put"
+    
+    query = "explain exit from a put trade"
+    # 
+    # query = "what will be buying cost if I long tesla CE"
+    
+    
+#     system_prompt="""Use the following pieces of context to answer the question in detail. Provide example only if it is in provided context and make sure to use them in rupees.""", 
+    
+#     system_prompt = """Use the following pieces of context to answer the question in detail. Provide example only if it is in context and make sure to use them in ₹.
+# If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+
+#     system_prompt = """Answer the question using the context. Provide examples only from the context and use only Rupees (₹) in examples. If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+    
+#     system_prompt = """Your task is to answer the question using the given context. 
+
+# Follow the below rules while answering the question:
+# - Only create example using the context
+# - Use only Rupees '₹' to represent currency. 
+# - If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+
+#     system_prompt = """You are an Indian Stock Market Assistant. Your task is to answer the question using the given context. Only create example from the given context and don't use '$'."""
+    
+    # query = "what is reliance?"    
+    # query = "what is python?"    
+    query = "what is an apple stock and nasdq"
+    query = "Generate a tabular format on playing long and short through options"
+    query = "What is FNO Trading?"
+    
+    system_prompt = """Answer the question only from context. 
+Provide examples only from the context. 
+If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+    
+    system_prompt = "Answer the question only from the e-book. If it is not sufficient then respond as \"Unknown\""
+    system_prompt = """Use the following pieces of book to answer the question at the end. \nIf you don't know the answer, please think rationally and answer from the book"""
+    # system_prompt = """Answer the question using the context. Provide examples only from the context and use only Rupees (₹) in examples. If you don't know the answer, just say 'Please rephrase the question I am unable to answer'"""
+    
+    # system_prompt = """Answer the question from the context. Provide examples only from the context. If you don't know the answer, just say 'Please rephrase the question'"""
+    # system_prompt = """Answer the question from the book. Provide examples only from the book. If you don't know the answer, just say 'Please rephrase the question'"""
+    
+    response = model.run(
+        system_prompt=system_prompt,
+        query=query
+    )
+    text = ""
+    for resp in response:
+        if isinstance(resp, list):
+            sources = resp
+            break
+        text += resp
+    
+    text = text.split("Question")[0].strip("\n")
+        
+    print("text: ", text)
+    open("./text.txt", "w").write(text)

requirements.txt

@@ -0,0 +1,10 @@
+langchain==0.0.353
+langchain-community==0.0.7
+langchain-core==0.1.4
+langchain-google-genai==0.0.5
+gradio==4.12.0
+gradio_client==0.8.0
+tqdm==4.66.1
+faiss-cpu==1.7.4
+pandas==2.2.0
+numpy==1.26.4

search.py

@@ -0,0 +1,113 @@
+import pickle
+import uuid
+from typing import Any, Callable, List, Optional
+
+import faiss
+import numpy as np
+from langchain.docstore.document import Document
+from langchain.docstore.in_memory import InMemoryDocstore
+from langchain.embeddings.base import Embeddings
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.vectorstores import FAISS
+from tqdm import tqdm
+
+
+def return_on_failure(value):
+  def decorate(f):
+    def applicator(*args, **kwargs):
+      try:
+        return f(*args,**kwargs)
+      except Exception as e:
+        print(f'Error "{e}" in {f.__name__}')
+        return value
+
+    return applicator
+
+  return decorate
+
+
+class SimilaritySearch(FAISS):
+     
+    @classmethod
+    @return_on_failure(None)
+    def load_from_disk(cls, embedding_function: Callable, data_dir: str = None):
+        docstore, index_to_docstore_id = pickle.load(open(f"{data_dir}/index.pkl", "rb"))
+        index_cpu = faiss.read_index(f"{data_dir}/index.faiss")
+        
+        # index_gpu = faiss.index_cpu_to_gpu(GPU_RESOURCE, 0, index_cpu)
+        # vector_store = FAISS(embedding_function, index_gpu, docstore, index_to_docstore_id)
+        
+        return FAISS(embedding_function, index_cpu, docstore, index_to_docstore_id)
+        
+    @classmethod
+    def __from(
+        cls,
+        texts: List[str],
+        embeddings: List[List[float]],
+        embedding: Embeddings,
+        metadatas: Optional[List[dict]] = None,
+        **kwargs: Any,
+    ) -> FAISS:
+        print("embeddings: ", len(embeddings), len(texts), len(metadatas))
+        index = faiss.IndexFlatIP(len(embeddings[0]))
+        index.add(np.array(embeddings, dtype=np.float32))
+        documents = []
+        for i, text in tqdm(enumerate(texts), total=len(texts)):
+            metadata = metadatas[i] if metadatas else {}
+            documents.append(Document(page_content=text, metadata=metadata))
+        index_to_id = {i: str(uuid.uuid4()) for i in range(len(documents))}
+        docstore = InMemoryDocstore(
+            {index_to_id[i]: doc for i, doc in enumerate(documents)}
+        )
+        return cls(embedding.embed_query, index, docstore, index_to_id, **kwargs)
+    
+    @classmethod
+    def from_texts(
+        cls,
+        texts: List[str],
+        embedding: Embeddings,
+        metadatas: Optional[List[dict]] = None,
+        ids: Optional[List[str]] = None,
+        **kwargs: Any,
+    ) -> FAISS:
+        """Construct FAISS wrapper from raw documents.
+
+        This is a user friendly interface that:
+            1. Embeds documents.
+            2. Creates an in memory docstore
+            3. Initializes the FAISS database
+
+        This is intended to be a quick way to get started.
+
+        Example:
+            .. code-block:: python
+
+                from langchain import FAISS
+                from langchain.embeddings import OpenAIEmbeddings
+                embeddings = OpenAIEmbeddings()
+                faiss = FAISS.from_texts(texts, embeddings)
+        """
+        # embeddings = embedding.embed_documents(texts)
+        final_texts, final_metadatas = [], []
+        embeddings = []
+        for i, text in tqdm(enumerate(texts), total=len(texts)):
+            try:
+                embeddings.append(embedding._embedding_func(text))
+                final_texts.append(text)
+                if len(metadatas) > 0:
+                    final_metadatas.append(metadatas[i])
+            except Exception as e:
+                text_splitter = RecursiveCharacterTextSplitter(chunk_size=4096, chunk_overlap=128)
+                splitted_texts = text_splitter.split_text(text)
+                embeddings.extend(embedding.embed_documents(splitted_texts))
+                final_texts.extend(splitted_texts)
+                final_metadatas.extend([metadatas[i]] * len(splitted_texts))
+        
+        return cls.__from(
+            final_texts,
+            embeddings,
+            embedding,
+            metadatas=final_metadatas,
+            # ids=ids,
+            **kwargs,
+        )