app.py

import gradio as gr
#import urllib.request
import requests
import bs4
import lxml
import os
#import subprocess
from huggingface_hub import InferenceClient,HfApi
import random
import json
import datetime
from pypdf import PdfReader
import uuid
#from query import tasks
from agent import (
    PREFIX,
    COMPRESS_DATA_PROMPT,
    COMPRESS_DATA_PROMPT_SMALL,
    LOG_PROMPT,
    LOG_RESPONSE,
)
client = InferenceClient(
    "mistralai/Mixtral-8x7B-Instruct-v0.1"
)
reponame="Omnibus/tmp"
save_data=f'https://huggingface.co/datasets/{reponame}/raw/main/'
token_self = os.environ['HF_TOKEN']
api=HfApi(token=token_self)

def find_all(url):
    return_list=[]
    print (url)
    print (f"trying URL:: {url}")        
    try:
        if url != "" and url != None:    
            out = []
            source = requests.get(url)
            print(source.status_code)
            if source.status_code ==200:
                print('trying')
                soup = bs4.BeautifulSoup(source.content,'lxml')
               
                rawp=(f'RAW TEXT RETURNED: {soup.text}')
                print (rawp)
                cnt=0
                cnt+=len(rawp)
                out.append(rawp)
                out.append("HTML fragments: ")
                q=("a","p","span","content","article")
                for p in soup.find_all("a"):
                    out.append([{"LINK TITLE":p.get('title'),"URL":p.get('href'),"STRING":p.string}])
                c=0
                out = str(out)
                rl = len(out)
                print(f'rl:: {rl}')
                for i in str(out):
                    if i == " " or i=="," or i=="\n" or i=="/" or i=="." or i=="<":
                        c +=1
                print (f'c:: {c}')
                #if c > MAX_HISTORY:
                #print("compressing...")
                #rawp = compress_data(c,purpose,task,out,result)  
                #result += rawp
                rawp=out
                return True, rawp
            else:
                
                return False, f'Status:: {source.status_code}'
        else: 
            print('passing')
            return False, "Enter Valid URL"
    except Exception as e:
        print (e)
        return False, f'Error: {e}'


def read_txt(txt_path):
    text=""
    with open(txt_path,"r") as f:
        text = f.read()
    f.close()
    print (text)
    return text

def read_pdf(pdf_path):
    text=""
    reader = PdfReader(f'{pdf_path}')
    number_of_pages = len(reader.pages)
    for i in range(number_of_pages):
        page = reader.pages[i]
        text = f'{text}\n{page.extract_text()}'
    print (text)
    return text

error_box=[]
def read_pdf_online(url):
    uid=uuid.uuid4()
    print(f"reading {url}")
    response = requests.get(url, stream=True)
    print(response.status_code)
    text=""
#################
    
#####################
    try:
        if response.status_code == 200:
            with open("test.pdf", "wb") as f:
                f.write(response.content)
            #f.close()
            #out = Path("./data.pdf")
            #print (out)
            reader = PdfReader("test.pdf")
            number_of_pages = len(reader.pages)
            print(number_of_pages)
            for i in range(number_of_pages):
                page = reader.pages[i]
                text = f'{text}\n{page.extract_text()}'
                print(f"PDF_TEXT:: {text}")
            return text
        else:
            text = response.status_code
            error_box.append(url)
            print(text)
            return text


    except Exception as e:
        print (e)
        return e


VERBOSE = True
MAX_HISTORY = 100
MAX_DATA = 20000

def format_prompt(message, history):
  prompt = "<s>"
  for user_prompt, bot_response in history:
    prompt += f"[INST] {user_prompt} [/INST]"
    prompt += f" {bot_response}</s> "
  prompt += f"[INST] {message} [/INST]"
  return prompt



def run_gpt(
    prompt_template,
    stop_tokens,
    max_tokens,
    seed,
    **prompt_kwargs,
):
    print(seed)
    timestamp=datetime.datetime.now()
    
    generate_kwargs = dict(
        temperature=0.9,
        max_new_tokens=max_tokens,
        top_p=0.95,
        repetition_penalty=1.0,
        do_sample=True,
        seed=seed,
    )
    
    content = PREFIX.format(
        timestamp=timestamp,
        purpose="Compile the provided data and complete the users task"
    ) + prompt_template.format(**prompt_kwargs)
    if VERBOSE:
        print(LOG_PROMPT.format(content))
    
    
    #formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history)
    #formatted_prompt = format_prompt(f'{content}', history)

    stream = client.text_generation(content, **generate_kwargs, stream=True, details=True, return_full_text=False)
    resp = ""
    for response in stream:
        resp += response.token.text
        #yield resp

    if VERBOSE:
        print(LOG_RESPONSE.format(resp))
    return resp

    
def compress_data(c, instruct, history):
    seed=random.randint(1,1000000000)
    
    print (c)
    #tot=len(purpose)
    #print(tot)
    divr=int(c)/MAX_DATA
    divi=int(divr)+1 if divr != int(divr) else int(divr)
    chunk = int(int(c)/divr)
    print(f'chunk:: {chunk}')
    print(f'divr:: {divr}')
    print (f'divi:: {divi}')
    out = []
    #out=""
    s=0
    e=chunk
    print(f'e:: {e}')
    new_history=""
    #task = f'Compile this data to fulfill the task: {task}, and complete the purpose: {purpose}\n'
    for z in range(divi):
        print(f's:e :: {s}:{e}')
        
        hist = history[s:e]
        
        resp = run_gpt(
            COMPRESS_DATA_PROMPT_SMALL,
            stop_tokens=["observation:", "task:", "action:", "thought:"],
            max_tokens=8192,
            seed=seed,
            direction=instruct,
            knowledge="",
            history=hist,
        )
        out.append(resp)
        #new_history = resp
        print (resp)
        #out+=resp
        e=e+chunk
        s=s+chunk
    return out

    
def compress_data_og(c, instruct, history):
    seed=random.randint(1,1000000000)
    
    print (c)
    #tot=len(purpose)
    #print(tot)
    divr=int(c)/MAX_DATA
    divi=int(divr)+1 if divr != int(divr) else int(divr)
    chunk = int(int(c)/divr)
    print(f'chunk:: {chunk}')
    print(f'divr:: {divr}')
    print (f'divi:: {divi}')
    out = []
    #out=""
    s=0
    e=chunk
    print(f'e:: {e}')
    new_history=""
    #task = f'Compile this data to fulfill the task: {task}, and complete the purpose: {purpose}\n'
    for z in range(divi):
        print(f's:e :: {s}:{e}')
        
        hist = history[s:e]
        
        resp = run_gpt(
            COMPRESS_DATA_PROMPT,
            stop_tokens=["observation:", "task:", "action:", "thought:"],
            max_tokens=8192,
            seed=seed,
            direction=instruct,
            knowledge=new_history,
            history=hist,
        )
        
        new_history = resp
        print (resp)
        out+=resp
        e=e+chunk
        s=s+chunk
    '''
    resp = run_gpt(
        COMPRESS_DATA_PROMPT,
        stop_tokens=["observation:", "task:", "action:", "thought:"],
        max_tokens=8192,
        seed=seed,
        direction=instruct,
        knowledge=new_history,
        history="All data has been recieved.",
    )'''
    print ("final" + resp)
    #history = "observation: {}\n".format(resp)
    return resp



def summarize(inp,history,report_check,sum_check,mem_check,data=None,files=None,url=None,pdf_url=None,pdf_batch=None):
    json_box=[]
    if inp == "":
        inp = "Process this data"
    history.clear()
    history = [(inp,"Working on it...")] 
    yield "",history,error_box,json_box

    if pdf_batch.startswith("http"):
        c=0
        data=""
        for i in str(pdf_batch):
            if i==",":
                c+=1
        print (f'c:: {c}')

        try:
            for i in range(c+1):
                batch_url = pdf_batch.split(",",c)[i]
                bb = read_pdf_online(batch_url)
                data=f'{data}\nFile Name URL ({batch_url}):\n{bb}'
        except Exception as e:
            print(e)
            #data=f'{data}\nError reading URL ({batch_url})'
    if pdf_url.startswith("http"):
        print("PDF_URL")
        out = read_pdf_online(pdf_url)
        data=out
    if url.startswith("http"):
        val, out = find_all(url)
        if not val:
            data="Error"
            rawp = str(out)
        else:
            data=out
    if files:
        for i, file in enumerate(files):
            try: 
                print (file)
                if file.endswith(".pdf"):
                    zz=read_pdf(file)
                    print (zz)
                    data=f'{data}\nFile Name ({file}):\n{zz}'
                elif file.endswith(".txt"):
                    zz=read_txt(file)
                    print (zz)
                    data=f'{data}\nFile Name ({file}):\n{zz}'                
            except Exception as e:
                data=f'{data}\nError opening File Name ({file})'                
                print (e) 
    if data != "Error" and data != "":
        print(inp)
        out = str(data)
        rl = len(out)
        print(f'rl:: {rl}')
        c=1
        for i in str(out):
            if i == " " or i=="," or i=="\n":
                c +=1
        print (f'c:: {c}')
        if mem_check:
            json_out = save_memory(inp,out)
            rawp = "Complete"
        if sum_check:
            json_out = compress_data(c,inp,out)  
    
            out = str(json_out)
            if report_check:
                rl = len(out)
                print(f'rl:: {rl}')
                c=1
                for i in str(out):
                    if i == " " or i=="," or i=="\n":
                        c +=1
                print (f'c2:: {c}')
                rawp = compress_data_og(c,inp,out)
            else:
                rawp = out
    else:
        rawp = "Provide a valid data source"
    history.clear()
    history.append((inp,rawp))
    yield "", history,error_box,json_out
SAVE_MEMORY = """
You are attempting to complete the task
task: {task}
Data:
{history}
Instructions:
Compile and categorize the data above into a JSON dictionary string
Include ALL text, datapoints, titles, descriptions, and source urls indexed into an easy to search JSON format
Required keys:
"keywords":["short", "list", "of", "important", "keywords", "found", "in", "this", "entry"],
"title":"title of entry",
"description":"A sentence summarizing the topic of this entry",
"content":"A brief paragraph summarizing the important datapoints found in this entry",
"url":"https://url.source"
"""

def format_json(inp):
    new_json=[]
    start_json={}
    print("FORMATTING:::")
    for i,line in enumerate(inp):
        line = line.strip()
        if "{" in line:
            print (line)
            start_json={}
        #print(f'test:: {line}')
        if "keywords" in line and ":" in line:
            start_json['keywords']=line.split(":")[1].strip(",")
            print (line)
        if "title" in line and ":" in line:
            start_json['title']=line.split(":")[1].strip(",")
            print (line)
        if "description" in line and ":" in line:
            start_json['description']=line.split(":")[1].strip(",")
            print (line)
        if "content" in line and ":" in line:
            start_json['content']=line.split(":")[1].strip(",")
            print (line)
        if "url" in line and ":" in line:
            start_json['url']=line.split(":")[1].strip(",")
            print (line)

        if "}" in line:
            new_json.append(start_json)
            print (new_json)

        
        
        #if line.startswith(("keywords","title","description","content","url","{","}")):
        #    print (line)
        
        '''
        print (f'NEW LINE:: {line}')
        if line.startswith("keywords"):
            keywords_val = line.split(":")[1]
        if line.startswith("title"):
            title_val = line.split(":")[1]
        if line.startswith("description"):
            description_val = line.split(":")[1]
        if line.startswith("content"):
            content_val = line.split(":")[1]
        if line.startswith("url"):
            url_val = line.split(":")[1]
         
        "keywords": ["texas", "news", "breaking", "houston", "dallas", "shooting"],
        "title": "Breaking News from CBS11 - CBS Texas",
        "description": "The latest news and headlines from CBS Texas.",
        "content": "CBS Texas provides the latest news and headlines. The source url is https://www.cbsnews.com/texas/local-news/",
        "url": "http
        '''

        
        #if line.startswith()


def save_memory(purpose, history):
    uid=uuid.uuid4()
    history=str(history)
    c=1
    inp = str(history)
    rl = len(inp)
    print(f'rl:: {rl}')
    for i in str(inp):
        if i == " " or i=="," or i=="\n" or i=="/" or i=="\\" or i=="." or i=="<":
            c +=1
    print (f'c:: {c}')

    seed=random.randint(1,1000000000)
    
    print (c)
    #tot=len(purpose)
    #print(tot)
    divr=int(c)/MAX_DATA
    divi=int(divr)+1 if divr != int(divr) else int(divr)
    chunk = int(int(c)/divr)
    print(f'chunk:: {chunk}')
    print(f'divr:: {divr}')
    print (f'divi:: {divi}')
    out_box = []
    #out=""
    s=0
    ee=chunk
    print(f'e:: {ee}')
    new_history=""
    task = f'Index this Data\n'
    for z in range(divi):
        print(f's:e :: {s}:{ee}')
        
        hist = inp[s:ee]
        
        resp = run_gpt(
            SAVE_MEMORY,
            stop_tokens=["observation:", "task:", "action:", "thought:"],
            max_tokens=4096,
            seed=seed,
            purpose=purpose,
            task=task,
            history=hist,
        ).strip('\n')
        #new_history = resp
        #print (resp)
        #out+=resp

        #print ("final1" + resp)
        try:
            resp='[{'+resp.split('[{')[1].split('</s>')[0]
            #print ("final2\n" + resp)
            #print(f"keywords:: {resp['keywords']}")
        except Exception as e:
            resp = resp
            print(e)
        timestamp=str(datetime.datetime.now())
        timename=timestamp.replace(" ","--").replace(":","-").replace(".","-")
        json_object=resp
        #json_object = json.dumps(out_box)
        #json_object = json.dumps(out_box,indent=4)
        with open(f"tmp-{uid}.json", "w") as outfile:
            outfile.write(json_object)
            
        outfile.close()
        api.upload_file(
        path_or_fileobj=f"tmp-{uid}.json",
        path_in_repo=f"/mem-test2/{timename}---{s}-{ee}.json",
        repo_id=reponame,
        #repo_id=save_data.split('datasets/',1)[1].split('/raw',1)[0],
        token=token_self,
        repo_type="dataset",
        )
        lines = resp.strip().strip("\n").split("\n")
        format_json(lines)

        r = requests.get(f'{save_data}mem-test2/main.json') 
        print(f'status code main:: {r.status_code}')
        try:
            print(f"KEYWORDS:: {json_object['keywords']}")
        except Exception as e:
            print(f"KEYWORDS:: {e}")
        if r.status_code==200:
            
            lod = json.loads(r.text)
            #lod = eval(lod)
            print (f'lod:: {lod}')
        if not r.status_code==200:
            lod = []
        for i,line in enumerate(lines):
            key_box=[]
            desc=""
            #print(f'LINE:: {line}')
            if ":" in line:
                print(f'line:: {line}')
            
            if "keywords" in line and ":" in line:
                print(f'trying:: {line}')
                keyw=line.split(":")[1]
                print (keyw)
                print (keyw.split("[")[1].split("]")[0])
                keyw=keyw.split("[")[1].split("]")[0]
                for ea in keyw.split(","):
                    s1=""
                    ea=ea.strip().strip("\n")
                    for ev in ea:
                        if ev.isalnum():
                            s1+=ev
                        if ev == " ":
                            s1+=ev
                        #ea=s1
                    print(s1)
                    key_box.append(s1)
            if "description" in line and ":" in line:
                #print(f'trying:: {line}')
                desc=line.split(":")[1]

            if key_box and desc:
                lod.append({"file_name":f"{timename}---{s}-{ee}","keywords":key_box,"description":str(desc),"index":f"{s}:{ee}"})
                key_box = []
                desc=""
        if lod:    
            json_object = json.dumps(lod, indent=4)
            with open(f"tmp2-{uid}.json", "w") as outfile2:
                outfile2.write(json_object)
            outfile2.close()
            api.upload_file(
            path_or_fileobj=f"tmp2-{uid}.json",
            path_in_repo=f"/mem-test2/main.json",
            repo_id=reponame,
            #repo_id=save_data.split('datasets/',1)[1].split('/raw',1)[0],
            token=token_self,
            repo_type="dataset",
            )

        ee=ee+chunk
        s=s+chunk       
        out_box.append(resp)
    return out_box




    
#################################
def clear_fn():
    return "",[(None,None)]

with gr.Blocks() as app:
    gr.HTML("""<center><h1>Mixtral 8x7B TLDR Summarizer + Web</h1><h3>Summarize Data of unlimited length</h3>""")
    chatbot = gr.Chatbot(label="Mixtral 8x7B Chatbot",show_copy_button=True)
    with gr.Row():
        with gr.Column(scale=3):
            prompt=gr.Textbox(label = "Instructions (optional)")
        with gr.Column(scale=1):
            report_check=gr.Checkbox(label="Return Report", value=True)
            sum_check=gr.Checkbox(label="Summarize", value=True)
            mem_check=gr.Checkbox(label="Memory", value=True)
            #sum_mem_check=gr.Radio(label="Output",choices=["Summary","Memory"])
            button=gr.Button()
        
        #models_dd=gr.Dropdown(choices=[m for m in return_list],interactive=True)
    with gr.Row():
        stop_button=gr.Button("Stop")
        clear_btn = gr.Button("Clear")
    with gr.Row():
        with gr.Tab("Text"):
            data=gr.Textbox(label="Input Data (paste text)", lines=6)
        with gr.Tab("File"):
            file=gr.Files(label="Input File(s) (.pdf .txt)")
        with gr.Tab("Raw HTML"):
            url = gr.Textbox(label="URL")
        with gr.Tab("PDF URL"):
            pdf_url = gr.Textbox(label="PDF URL")       
        with gr.Tab("PDF Batch"):
            pdf_batch = gr.Textbox(label="PDF URL Batch (comma separated)")
    json_out=gr.JSON()
    e_box=gr.Textbox()
    #text=gr.JSON()
    #inp_query.change(search_models,inp_query,models_dd)
    clear_btn.click(clear_fn,None,[prompt,chatbot])
    go=button.click(summarize,[prompt,chatbot,report_check,sum_check,mem_check,data,file,url,pdf_url,pdf_batch],[prompt,chatbot,e_box,json_out])
    stop_button.click(None,None,None,cancels=[go])
app.queue(default_concurrency_limit=20).launch(show_api=False)