llama/elastic_keywords_llama3_3.py

from html import escape
from lxml import etree
from datetime import datetime
from elasticsearch import Elasticsearch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TextIteratorStreamer
from threading import Thread
import torch
import time
from concurrent.futures import ThreadPoolExecutor
import concurrent
import threading
lock = threading.Lock()
#lock1 = threading.Lock()
#from cleantext import clean
#import re


if torch.cuda.is_available():
    model_id = "PartAI/Dorna-Llama3-8B-Instruct"
    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype=torch.bfloat16)
    tokenizer = AutoTokenizer.from_pretrained(model_id)


# pipe = pipeline(
#     "text-generation",
#     model=model,
#     tokenizer=tokenizer,
#     torch_dtype=torch.float16,
#     device_map="auto",
# )




index_name_i = 'semantic_search-v10'




es = Elasticsearch(
    "http://127.0.0.1:6900",
    # ca_certs="/path/to/http_ca.crt",
    basic_auth=("elastic", "SG*7eGwg+KG2_*-1_mMm")
)

counter = 0
total = 0
remained = 0
id = ''
keywords_count = 15
        
messages = [    {'role': 'user', 'content': 'لطفا فقط فارسی جواب بدهید'},
]
def es_iterate_all_documents(es, index, pagesize=250, scroll_timeout="12m", **kwargs):
    """
    Helper to iterate ALL values from a single index
    Yields all the documents.
    """
    global counter
    global total
    global remained
    is_first = True
    while True:
        # Scroll next
        if is_first: # Initialize scroll
            # result = es.search(index=index, scroll="12m", **kwargs, body={
            #     "size": pagesize
            # })
            result = es.search(index=index, scroll="12m", **kwargs, size=pagesize)
            total = result["hits"]["total"]['value']
            remained = total
            print('total = %d' % total)
            is_first = False
        else:
            # result = es.scroll(body={
            #     "scroll_id": scroll_id,
            #     "scroll": scroll_timeout
            # })
            result = es.scroll( scroll_id = scroll_id, scroll = scroll_timeout )
        scroll_id = result["_scroll_id"]
        hits = result["hits"]["hits"]
        counter += len(hits)
        print("progress -> %.2f %% , counte: %d" % ((counter / total)*100, counter))
        # Stop after no more docs
        if not hits:
            break
        # Yield each entry
        yield from ({"source":hit['_source'], "id":hit['_id']} for hit in hits)

def generateKeywords(text, keywords_count, id):
    global remained
    try:
        lock.acquire()
        if torch.cuda.is_available():
            print('ok')
        messages =  [{"role": "user", "content": 
                    '''از "متن" حداقل {} کلیدواژه مهم و پراهمیت را استخراج کن و در قالب لیست به زبان فارسی چاپ کن. "متن": {}
                    '''.format(keywords_count, text)
        }]
        
        input_ids = tokenizer.apply_chat_template(
            messages,
            add_generation_prompt=True,
            return_tensors="pt"
        )#.to(model.device)

        terminators = [
                    tokenizer.eos_token_id,
                    tokenizer.convert_tokens_to_ids("<|eot_id|>")
                ]
        model.generation_config.pad_token_id = tokenizer.pad_token_id

        
        outputs = model.generate(
            input_ids,
            max_new_tokens=256,
            eos_token_id=terminators,
            do_sample=True,
            temperature=0.6,
            top_p=0.85,
        )
        #lock0.release()
        response = outputs[0][input_ids.shape[-1]:]
        keywords = tokenizer.decode(response, skip_special_tokens=True)
        #lock1.acquire()
        resp = es.update(index=index_name_i, id=id, doc={"content_keywords-llama3-str": str(keywords)})
        remained = remained - 1
        #lock1.release()
        print("update id = {}, remained items = {} ".format(id,remained))
        lock.release()
        return id

    except Exception as inst:
        print(type(inst))    # the exception type
        print(inst.args)     # arguments stored in .args
        print("Exception: " + str(inst))



if __name__ == "__main__":
    start_time = time.time()
    futures = []
    with ThreadPoolExecutor() as executor:
        #t0 = time.time()
        for mentry in es_iterate_all_documents(es, index_name_i):
            try:
                entry = mentry['source']
                id = mentry['id']
                text = entry.get('content','')
                full_path = entry.get('other_info', {'full_path':''})['full_path']
                text_len = len(text)
                if full_path == 'عنوان' or full_path == 'موخره' or full_path == 'امضاء' or text_len == 0:
                    continue

                keywords_count = (text_len / 1000) * 15
                if keywords_count < 0.3: 
                    continue
                keywords_count = int(keywords_count)
                if keywords_count == 0: 
                    keywords_count = 1
                futures.append(executor.submit(generateKeywords, text, keywords_count, id))
            except Exception as inst:
                print(type(inst))    # the exception type
                print(inst.args)     # arguments stored in .args
                print(inst)          # __str__ allows args to be printed directly,
                                    # but may be overridden in exception subclasses
                print("Exception:=> %s -> %.2f " % (id , counter / total))


    print(f"pre end...elapsed time:   {time.time()-start_time}")
    results = []
    for future in concurrent.futures.as_completed(futures):
        results.append(future.result())
        print('-')
    # print('\n'.join(results))

    end_time = time.time()
    print(f"elapsed time:   {end_time-start_time}")