Update app.py

from datasets import load_dataset
from datasets import Dataset
from sentence_transformers import SentenceTransformer
import faiss
import time
import json
#import torch
import pandas as pd
from llama_cpp import Llama
#from langchain_community.llms import LlamaCpp
from threading import Thread
from huggingface_hub import Repository, upload_file
import os


HF_TOKEN = os.getenv('HF_Token')
#Log_Path="./Logfolder"
logfile = 'DiabetesChatLog.txt'
historylog = [{
"Prompt": '',
"Output": ''
}]

data = load_dataset("Namitg02/Test", split='train', streaming=False)
#Returns a list of dictionaries, each representing a row in the dataset.
length = len(data)

embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
embedding_dim = embedding_model.get_sentence_embedding_dimension()
# Returns dimensions of embedidng


index = faiss.IndexFlatL2(embedding_dim)
data.add_faiss_index("embeddings", custom_index=index)
# adds an index column for the embeddings

#question = "How can I reverse Diabetes?"

SYS_PROMPT = """You are an assistant for answering questions.
You are given the extracted parts of documents and a question. Provide a conversational answer.
If you don't know the answer, just say "I do not know." Don't make up an answer. Don't repeat the SYS_PROMPT."""
# Provides context of how to answer the question

#llm_model = "TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF", tinyllama-1.1b-chat-v1.0.Q5_K_M.gguf
# TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF , TinyLlama/TinyLlama-1.1B-Chat-v0.6, andrijdavid/TinyLlama-1.1B-Chat-v1.0-GGUF"

model = Llama(
model_path="./llama-2-7b-chat.Q4_K_M.gguf",
# chat_format="llama-2",
n_gpu_layers = 0,
temperature=0.75,
n_ctx = 4096,
max_tokens=500,
top_p=0.95 #,
# eos_tokens=terminators
# callback_manager=callback_manager,
# verbose=True, # Verbose is required to pass to the callback manager
)
#initiate model and tokenizer

def search(query: str, k: int = 2 ):
"""a function that embeds a new query and returns the most probable results"""
embedded_query = embedding_model.encode(query) # create embedding of a new query
scores, retrieved_examples = data.get_nearest_examples( # retrieve results
"embeddings", embedded_query, # compare our new embedded query with the dataset embeddings
k=k # get only top k results
)
return scores, retrieved_examples
# returns scores (List[float]): the retrieval scores from either FAISS (IndexFlatL2 by default) and examples (dict) format
# called by talk function that passes prompt

def format_prompt(prompt,retrieved_documents,k,history,memory_limit=3):
"""using the retrieved documents we will prompt the model to generate our responses"""
PROMPT = f"Question:{prompt}\nContext:"
for idx in range(k) :
PROMPT+= f"{retrieved_documents['0'][idx]}\n"
print("historyinfo")
print(history)
if len(history) == 0:
return PROMPT

if len(history) > memory_limit:
history = history[-memory_limit:]

print("checkwohist")
# PROMPT = PROMPT + f"{history[0][0]} [/INST] {history[0][1]} </s>"
# print("checkwthhist")
# print(PROMPT)
# Handle conversation history
for user_message, bot_message in history[0:]:
PROMPT += f"<s>[INST] {user_message} [/INST] {bot_message} </s>"
print("checkwthhist2")
print(PROMPT)
return PROMPT


# Called by talk function to add retrieved documents to the prompt. Keeps adding text of retrieved documents to string that are retreived

def talk(prompt, history):
k = 2 # number of retrieved documents
scores , retrieved_documents = search(prompt, k) # get retrival scores and examples in dictionary format based on the prompt passed
print(retrieved_documents.keys())
# print("check4")
formatted_prompt = format_prompt(prompt,retrieved_documents,k,history,memory_limit=3) # create a new prompt using the retrieved documents
print("check5")
pd.options.display.max_colwidth = 4000
# print(retrieved_documents['0'])
# print(formatted_prompt)
# formatted_prompt_with_history = add_history(formatted_prompt, history)

# formatted_prompt_with_history = formatted_prompt_with_history[:600] # to avoid memory issue
# print(formatted_prompt_with_history)
messages = [{"role":"system","content":SYS_PROMPT},{"role":"user","content":formatted_prompt}]
print(messages)
# binding the system context and new prompt for LLM
# the chat template structure should be based on text generation model format

# indicates the end of a sequence
stream = model.create_chat_completion(messages = messages, max_tokens=1000, stop=["</s>"], stream=False)
# print(f"{stream}")
print("check 7")
print(stream['choices'][0]['message']['content'])
return(stream['choices'][0]['message']['content'])
# text = ""
# for output in stream:
# text += output['choices'][0]['message']['content']
# print(f"{output}")
# print("check3H")
# print(text)
# yield text



# calling the model to generate response based on message/ input
# do_sample if set to True uses strategies to select the next token from the probability distribution over the entire vocabulary
# temperature controls randomness. more renadomness with higher temperature
# only the tokens comprising the top_p probability mass are considered for responses
# This output is a data structure containing all the information returned by generate(), but that can also be used as tuple or dictionary.



TITLE = "AI Copilot for Diabetes Patients"

DESCRIPTION = "I provide answers to concerns related to Diabetes"

import gradio as gr
# Design chatbot
demo = gr.ChatInterface(
fn=talk,
chatbot=gr.Chatbot(
show_label=True,
show_share_button=True,
show_copy_button=True,
likeable=True,
layout="bubble",
bubble_full_width=False,
),
theme="Soft",
examples=[["what is Diabetes?"]],
title=TITLE,
description=DESCRIPTION,
)
# launch chatbot and calls the talk function which in turn calls other functions
print("check14")
#print(historylog)
#memory_panda = pd.DataFrame(historylog)
#Logfile = Dataset.from_pandas(memory_panda)
#Logfile.push_to_hub("Namitg02/Logfile",token = HF_TOKEN)
demo.launch()

app.py

@@ -1,179 +0,0 @@
-from datasets import load_dataset
-from datasets import Dataset
-from sentence_transformers import SentenceTransformer
-import faiss
-import time
-from datetime import datetime
-import json
-#import torch
-import uuid
-import pandas as pd
-from llama_cpp import Llama
-#from langchain_community.llms import LlamaCpp
-from threading import Thread
-from huggingface_hub import Repository, upload_file
-import os
-
-
-HF_TOKEN = os.getenv('HF_Token')
-logfile = 'DiabetesChatLog.txt'
-
-data = load_dataset("Namitg02/Test", split='train', streaming=False)
-#Returns a list of dictionaries, each representing a row in the dataset.
-length = len(data)
-
-embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
-embedding_dim = embedding_model.get_sentence_embedding_dimension()
-# Returns dimensions of embedidng
-
-
-index =  faiss.IndexFlatL2(embedding_dim)
-data.add_faiss_index("embeddings", custom_index=index) 
-# adds an index column for the embeddings
-
-#question = "How can I reverse Diabetes?"
-
-SYS_PROMPT = """You are an assistant for answering questions like a medical person.
-You are given the extracted parts of document, a question and history of questions and answers . Provide a brief conversational answer.
-If you do not know the answer, just say "I do not know." Do not make up an answer. Don't repeat the SYS_PROMPT or say that you are referring to document or an article."""
-# Provides context of how to answer the question
-
-
-model = Llama(
-    model_path="./meditron-7b-chat.Q4_K_M.gguf",
- #   chat_format="llama-2",
-    n_gpu_layers = 0,
-    temperature=0.75,
-    n_ctx = 4096,
-    top_p=0.95 #,
- #   eos_tokens=terminators
-    # callback_manager=callback_manager,
-    # verbose=True,  # Verbose is required to pass to the callback manager
-    )
-#initiate model and tokenizer
-                        
-def search(query: str, k: int = 2 ):
-    """a function that embeds a new query and returns the most probable results"""
-    embedded_query = embedding_model.encode(query) # create embedding of a new query
-    scores, retrieved_examples = data.get_nearest_examples( # retrieve results
-        "embeddings", embedded_query, # compare our new embedded query with the dataset embeddings
-        k=k # get only top k results
-    )
-    return scores, retrieved_examples
-# returns scores (List[float]): the retrieval scores from either FAISS (IndexFlatL2 by default) and examples (dict) format
-# called by talk function that passes prompt
-
-def format_prompt(prompt,retrieved_documents,k,history,memory_limit=3):
-    """using the retrieved documents we will prompt the model to generate our responses"""   
-    PROMPT = f"Question:{prompt}\nContext:"
-    for idx in range(k) :
-        PROMPT+= f"{retrieved_documents['0'][idx]}\n"
-    print("historyinfo")
-    print(f"{history}")
-    if len(history) == 0:
-        return PROMPT
-    
-    if len(history) > memory_limit:
-        history = history[-memory_limit:]
-    
-    print("checkwohist")
-#    PROMPT = PROMPT + f"{history[0][0]} [/INST] {history[0][1]} </s>"
-# Handle conversation history
-    for user_message, bot_message in history[0:]:
-        PROMPT += f"<s>[INST] {user_message} [/INST] {bot_message} </s>"
-    print("checkwthhist2")
-    
-    return PROMPT
-
-
-# Called by talk function to add retrieved documents to the prompt. Keeps adding text of retrieved documents to string that are retreived
-
-def talk(prompt, history):
-    k = 2 # number of retrieved documents
-    scores , retrieved_documents = search(prompt, k) # get retrival scores and examples in dictionary format based on the prompt passed
-    print(retrieved_documents.keys())
-#    print("check4")
-    formatted_prompt = format_prompt(prompt,retrieved_documents,k,history,memory_limit=3) # create a new prompt using the retrieved documents
-    print("check5")
-    pd.options.display.max_colwidth = 4000
-    messages = [{"role":"system","content":SYS_PROMPT},{"role":"user","content":formatted_prompt}]  
-    print(messages)
-    # binding the system context and new prompt for LLM
-    # the chat template structure should be based on text generation model format
-
-# indicates the end of a sequence
-    stream = model.create_chat_completion(messages = messages,max_tokens =400, stop=["</s>"], stream=False)
-#    print(f"{stream}")
-    print("check 7")
-    print(stream['choices'][0]['message']['content'])
-    response = stream['choices'][0]['message']['content']
-        
-#    for user_message, bot_message in history[0:]:
-#        historylog += f"<s>[INST] {user_message} [/INST] {bot_message} </s>"
-    historylog = ''
-    historylog += f"{prompt} \n {response} "
-    print("history log")
-    print(str(historylog))
-    print("history log string printed")
-
-    try:
-    # write data to file
-        unique_filename = f"file_{uuid.uuid4()}.txt"
-        with open(unique_filename, "a") as data:
-            data.write(historylog)
-
-        print("History log printed:")
-        with open(unique_filename, "r") as data:
-            print(data.read())
-    
-    
-    except IOError as e:
-        print(f"An error occurred: {e}")
-
-#    from huggingface_hub import HfApi
-#    api = HfApi()
-#    api.upload_file(
-#        path_or_fileobj="./file.txt",
-#        path_in_repo="file.txt",
-#        repo_id="Namitg02/Test",
-#        repo_type="space"
-#    )
-    
-    print("upload section passed")    
-    
-    for i in range(len(response)):
-        time.sleep(0.05)
-        yield response[: i+1]
-    
-    # calling the model to generate response based on message/ input
-    # do_sample if set to True uses strategies to select the next token from the probability distribution over the entire vocabulary
-    # temperature controls randomness. more renadomness with higher temperature
-    # only the tokens comprising the top_p probability mass are considered for responses
-    # This output is a data structure containing all the information returned by generate(), but that can also be used as tuple or dictionary.
-
-TITLE = "AI Copilot for Diabetes Patients"
-
-DESCRIPTION = "I provide answers to concerns related to Diabetes"
-
-import gradio as gr
-# Design chatbot
-demo = gr.ChatInterface(
-    fn=talk,
-    chatbot=gr.Chatbot(
-        show_label=True,
-        show_share_button=True,
-        show_copy_button=True,
-        likeable=True,
-        layout="bubble",
-        bubble_full_width=False,
-    ),
-    theme="Soft",
-    examples=[["what is Diabetes?"]],
-    title=TITLE,
-    description=DESCRIPTION,
-)
-# launch chatbot and calls the talk function which in turn calls other functions
-print("check14")
-
-demo.launch()
-#demo.launch(auth=("namit", "wolfmagic"))