Create app.py

app.py

@@ -0,0 +1,265 @@
+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,
+    SAVE_MEMORY,
+    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)
+
+
+
+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_mem_check,data=None):
+    json_box=[]
+    if inp == "":
+        inp = "Process this data"
+    history.clear()
+    history = [(inp,"Working on it...")] 
+    yield "",history,error_box,json_box
+
+    if data != "Error" and data != "" and data != None:
+        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 sum_mem_check=="Memory":
+            #save_memory(inp,out)
+            rawp = "Complete"
+        if sum_mem_check=="Summarize":
+            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
+
+    
+#################################
+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_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_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)