Create app.py

app.py

@@ -0,0 +1,201 @@
+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 query import tasks
+from prompts import (
+    COMPRESS_DATA_PROMPT,
+    COMPRESS_DATA_PROMPT_SMALL,
+    LOG_PROMPT,
+    LOG_RESPONSE,
+)
+api=HfApi()
+
+
+
+client = InferenceClient(
+    "mistralai/Mixtral-8x7B-Instruct-v0.1"
+)
+
+def parse_action(string: str):
+    print("PARSING:")
+    print(string)
+    assert string.startswith("action:")
+    idx = string.find("action_input=")
+    print(idx)
+    if idx == -1:
+        print ("idx == -1")
+        print (string[8:])
+        return string[8:], None
+
+    print ("last return:")
+    print (string[8 : idx - 1])
+    print (string[idx + 13 :].strip("'").strip('"'))
+    return string[8 : idx - 1], string[idx + 13 :].strip("'").strip('"')
+
+
+
+VERBOSE = True
+MAX_HISTORY = 100
+MAX_DATA = 1000
+
+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,
+    purpose,
+    **prompt_kwargs,
+):
+    print(seed)
+    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=purpose,
+    ) + 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,purpose, task, 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=2048,
+            seed=seed,
+            purpose=purpose,
+            task=task,
+            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=1024,
+        seed=seed,
+        purpose=purpose,
+        task=task,
+        knowledge=new_history,
+        history="All data has been recieved.",
+    )'''
+    print ("final" + resp)
+    history = "observation: {}\n".format(resp)
+    return history
+
+
+
+def summarize(inp,file=None):
+    out = str(inp)
+    rl = len(out)
+    print(f'rl:: {rl}')
+    for i in str(out):
+        if i == " " or i=="," or i=="\n":
+            c +=1
+    print (f'c:: {c}')
+    if rl > MAX_DATA:
+        print("compressing...")
+        rawp = compress_data(c,purpose,task,out)    
+    print (rawp)
+    print (f'out:: {out}')
+    #history += "observation: the search results are:\n {}\n".format(out)
+    task = "complete?"
+    return history        
+#################################
+
+
+examples =[
+    "what are todays breaking news stories?",
+    "find the most popular model that I can use to generate an image by providing a text prompt",
+    "return the top 10 models that I can use to identify objects in images",
+    "which models have the most likes from each category?"
+]
+
+
+app = gr.ChatInterface(
+    fn=run,
+    chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=True, likeable=True, layout="panel"),
+    title="Mixtral 46.7B Powered <br> Search",
+    examples=examples,
+    concurrency_limit=20,
+)
+
+'''
+with gr.Blocks() as app:
+    with gr.Row():
+        inp_query=gr.Textbox()
+        models_dd=gr.Dropdown(choices=[m for m in return_list],interactive=True)
+    with gr.Row():
+        button=gr.Button()
+        stop_button=gr.Button("Stop")
+    text=gr.JSON()
+    inp_query.change(search_models,inp_query,models_dd)
+    go=button.click(test_fn,None,text)
+    stop_button.click(None,None,None,cancels=[go])
+'''
+app.launch(server_port=7860,show_api=False)