fix bugs

chatbot.py

@@ -11,11 +11,11 @@ from sklearn.metrics.pairwise import cosine_similarity
 import datetime
 import re
 import random
+from fastapi.middleware.cors import CORSMiddleware
 from embedder_sbert_qavanin_285k import PersianVectorAnalyzer
-from normalizer import cleaning
-from fastapi import FastAPI
+#from normalizer import cleaning
+from fastapi import FastAPI ,Header
 from pydantic import BaseModel
-
 # LLM Libs
 from openai import OpenAI
 from langchain_openai import ChatOpenAI  # pip install -U langchain_openai
@@ -24,14 +24,23 @@ import requests
 today = f'{datetime.datetime.now().year}{datetime.datetime.now().month}{datetime.datetime.now().day}'
 
 chatbot = FastAPI()
+origins = ["*"]
+
+chatbot.add_middleware(
+    CORSMiddleware,
+    allow_origins=origins,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
 
 # -------------------
-# مدل‌ها و مسیر داده
-# -------------------
-EMBED_MODEL = "/home/sabr/MODLES/rag_chat/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
-RERANKER_MODEL = "/home/sabr/MODLES/rag_chat/BAAI/bge-reranker-v2-m3"
-FAISS_INDEX_PATH = "./qavanin-faiss/faiss_index_qavanin_285k.index"
-FAISS_METADATA_PATH = "./qavanin-faiss/faiss_index_qavanin_285k_metadata.json"
+# مدل‌ها و مسیر دادهsrc/app/qavanin-faiss/faiss_index_qavanin_285k_metadata.json
+# -------------------/src/app/qavanin-faiss
+EMBED_MODEL = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
+RERANKER_MODEL = "BAAI/bge-reranker-v2-m3"
+FAISS_INDEX_PATH = "/src/app/qavanin-faiss/faiss_index_qavanin_285k.index"
+FAISS_METADATA_PATH = "/src/app/qavanin-faiss/faiss_index_qavanin_285k_metadata.json"
 
 RERANK_BATCH = int(os.environ.get("RERANK_BATCH", 256))
 # print(f'RERANK_BATCH: {RERANK_BATCH}')
@@ -67,6 +76,37 @@ def get_client():
     
     return client
 
+def llm_base_request(query):
+    # model = 'cf.gemma-3-12b-it'
+    model = 'gpt-4o-mini'
+    prompt = f'برای متن {query} زیر، عنوانی کوتاه که بین 3 تا 6 کلمه داشته باشد، انتخاب کن. غیر از عنوان، به هیچ وجه توضیح اضافه ای در قبل یا بعد آن اضافه نکن.'
+    client = get_client()
+    try:
+        messages.append({"role": "user", "content": prompt})
+        response = client.chat.completions.create(
+        messages = messages,
+        model= model)  # "gpt-4o", "gpt-4o-mini", "deepseek-chat" , "gemini-2.0-flash", gemini-2.5-flash-lite
+        # gpt-4o                    : 500
+        # gpt-4o-mini           : 34
+        # deepseek-chat:      : 150
+        # gemini-2.0-flash   : error
+        # cf.gemma-3-12b-it : 1
+        # gemini-2.5-flash-lite : 35 خیلی خوب
+        
+        answer = response.choices[0].message.content
+        # پاسخ را هم به سابقه اضافه می‌کنیم
+        messages.append({"role": "assistant", "content": answer})
+       
+        
+    except Exception as error:
+        with open('./llm-answer/error-in-llm.txt', mode='a+', encoding='utf-8') as file:
+            error_message = f'\n\nquery: {query.strip()}\nerror:{error} \n-------------------------------\n'
+            file.write(error_message)
+            
+        return ''
+   
+    return answer
+
 def llm_request(query, model):
     
     if query == '':
@@ -121,7 +161,8 @@ class HybridRetrieverReranker:
         self.N = len(content_list)
 
         # Dense
-        self.embedder = SentenceTransformer(EMBED_MODEL, device=self.device)
+        self.embedder = SentenceTransformer(EMBED_MODEL,cache_folder='/src/MODELS', device=self.device)
+        #self.embedder = SentenceTransformer(EMBED_MODEL, device=self.device)
 
         # Sparse (مثل قبل برای حفظ خروجی)
         self.vectorizer = TfidfVectorizer(
@@ -132,7 +173,7 @@ class HybridRetrieverReranker:
         self.tfidf_matrix = self.vectorizer.fit_transform(self.content_list)
 
         # Reranker
-        self.tokenizer = AutoTokenizer.from_pretrained(RERANKER_MODEL, use_fast=True)
+        self.tokenizer = AutoTokenizer.from_pretrained(RERANKER_MODEL,cache_dir='/src/MODELS', use_fast=True)
         self.reranker = AutoModelForSequenceClassification.from_pretrained(
             RERANKER_MODEL
         ).to(self.device)
@@ -290,16 +331,17 @@ class HybridRetrieverReranker:
 
 def single_query(query: str):
 
-    query = cleaning(query)
-        
+   # query = cleaning(query)
+    retrived_sections_ids = []
     retrived_sections = pipe.search(query, content_list, topk_dense=30, topk_sparse=30, pre_rerank_k=30, final_k=10)
     final_similars = ''
     for i, row in enumerate(retrived_sections, 1):
         id_value = '{' + str(ids[row['idx']]) + '}'
         result = f"id: {id_value} \n{row['prefix']} {row['content']}\n"
+        retrived_sections_ids.append(ids[row['idx']])
         final_similars += ''.join(result)
     
-    return final_similars, retrived_sections
+    return final_similars, retrived_sections_ids
 
 def find_refrences(llm_answer: str) -> List[str]:
     """
@@ -313,7 +355,12 @@ def find_refrences(llm_answer: str) -> List[str]:
     """
     pattern = r"\{[^\}]+\}"
     refrence_ids = re.findall(pattern, llm_answer)
-    
+    new_refrences_ids = []
+    for itm in refrence_ids:
+        refrence = itm.lstrip('{')
+        refrence = refrence.lstrip('}')
+        new_refrences_ids.append(refrence)
+    # refrence_ids = [item.lstrip('{').rstrip('}') for item in refrence_ids]
     return refrence_ids
 
 def replace_refrences(llm_answer: str, refrences_list:List[str]) -> List[str]:
@@ -330,12 +377,12 @@ def replace_refrences(llm_answer: str, refrences_list:List[str]) -> List[str]:
     for index, ref in enumerate(refrences_list,1):
             # breakpoint()
             llm_answer = llm_answer.replace(ref, f'[{index}]')
-            id = ref.lstrip('{')
-            id = id.rstrip('}')
-            refrences += ''.join(f'[{index}] https://majles.tavasi.ir/entity/detail/view/qsection/{id}\n')
+            # id = ref.lstrip('{')
+            # id = id.rstrip('}')
+            # refrences += ''.join(f'[{index}] https://majles.tavasi.ir/entity/detail/view/qsection/{id}\n')
         
-    llm_answer = f'{llm_answer}\n\nمنابع پاسخ‌:\n{refrences.strip()}'
-    return llm_answer
+    # llm_answer = f'{llm_answer}\n\nمنابع پاسخ‌:\n{refrences.strip()}'
+    return llm_answer.strip()
 
 # load basic items
 content_list, ids, prefix_list, faiss_index = load_faiss_index(FAISS_INDEX_PATH, FAISS_METADATA_PATH)
@@ -347,26 +394,62 @@ messages = [
         {"role": "system", "content": "تو یک دستیار خبره در زمینه حقوق و قوانین مرتبط به آن هستی و می توانی متون حقوقی را به صورت دقیق توضیح بدهی . پاسخ ها باید الزاما به زبان فارسی باشد. پاسخ ها فقط از متون قانونی که در پرامپت وجود دارد استخراج شود."},
 ]
 
-def run_chatbot(query:str, chat_id:str):
+models = ["gemini-2.5-flash-lite", "gpt-4o-mini"]
 
+def save_result(chat_obj: object) -> bool:
+    # index result in elastic
+    pass
+
+def run_chatbot(query:str, chat_id:str):
+    prompt_status = True
+    status_text = 'لطفا متن سوال را وارد نمائید'
     if query == '':
-        return 'لطفا متن سوال را وارد نمائید'
+        prompt_status = False
     
     start_time = (datetime.datetime.now())
 
-    result_passages_text, result_passages_ids = single_query(query)
-    end_retrive = datetime.datetime.now()
-    print('-'*40)
-    retrive_duration = (end_retrive - start_time).total_seconds()
-    print(f'retrive duration: {str(retrive_duration)}')
+    # در صورتی که وضعیت پرامپت معتبر باشد، وارد فرایند شو
+    if prompt_status:
+        result_passages_text, result_passages_ids = single_query(query)
+        end_retrive = datetime.datetime.now()
+        print('-'*40)
+        retrive_duration = (end_retrive - start_time).total_seconds()
+        print(f'retrive duration: {str(retrive_duration)}')
 
-    prompt = f'برای پرسش "{query}" از میان مواد قانونی "{result_passages_text}" .پاسخ مناسب و دقیق را استخراج کن. درصورتی که مطلبی مرتبط با پرسش در متن پیدا نشد، فقط پاسخ بده: "متاسفانه در منابع، پاسخی پیدا نشد!"'
-    try:
-        model = "gemini-2.5-flash-lite"
-        llm_answer = llm_request(prompt, model)
-    except Exception as error:
-        model = "gpt-4o-mini"
-        llm_answer = llm_request(prompt, model)
+        prompt = f'برای پرسش "{query}" از میان مواد قانونی "{result_passages_text}" .پاسخ مناسب و دقیق را استخراج کن. درصورتی که مطلبی مرتبط با پرسش در متن پیدا نشد، فقط پاسخ بده: "متاسفانه در منابع، پاسخی پیدا نشد!"'
+
+        llm_model = ''
+        for model in models:
+            try:
+                llm_model = model
+                llm_answer = llm_request(prompt, model)
+            except Exception as error:
+                    error = f'model: {model} \n{error}\n\n'
+                    prompt_status = False
+                    status_text = 'با عرض پوزش، سرویس موقتا در دسترس نیست. لطفا دقایقی دیگر دوباره تلاش نمائید!'
+                
+    else:
+        chat_obj = {
+        'id' : chat_id, # str
+        'title' : '', # str
+        'user_id' : '',
+        'user_query' : query, # str
+        'model_key' : llm_model, # str
+        'retrived_passage' : result_passages_text, # str
+        'retrived_ref_ids' : result_passages_ids, # list[obj]
+        'prompt_type' : 'question-answer', # str
+        'retrived_duration' : retrive_duration, # str
+        'llm_duration' : '0', # str
+        'full_duration' : '0', # str
+        'time_create' : str(start_time), # str
+        'used_ref_ids' : [], # list[str]
+        'prompt_answer' : '', # str
+        'status_text' : status_text,
+        'status' : prompt_status, # or False # bool
+        }
+        
+        # آبجکت ایجاد شده با بازگردان
+        return chat_obj, status_text
             
     llm_answer_duration = (datetime.datetime.now() - end_retrive).total_seconds()
     print(f'llm answer duration: {str(llm_answer_duration)}')
@@ -378,33 +461,51 @@ def run_chatbot(query:str, chat_id:str):
     print(f'full prompt duration: {full_prompt_duration}')
     print('~'*40)
     
-    chat_obj = {
-        'chat-id' : chat_id, # str
-        'chat-title' : '', # str
-        'user-id' : '',
-        'user-query' : query, # str
-        'model' : model, # str
-        'result-passages' : result_passages_text, # str
-        'retrived-passages-ids' : result_passages_ids, # list[obj]
-        'retrive-duration' : retrive_duration, # str
-        'llm-answer-duration' : llm_answer_duration, # str
-        'full-prompt-duration' : full_prompt_duration, # str
-        'chat-date' : str(start_time), # str
-        'used-refrences-in-answer' : used_refrences_in_answer, # list[str]
-        'llm-answer' : llm_answer, # str
-    }
-    # prev_chat_data = []    
-    # with open('./llm-answer/chat-messages.json', mode='r', encoding='utf-8') as file:
-    #         prev_chat_data = json.load(file)
-    #         prev_chat_data.append(chat_obj)
+    status_text ='پاسخ با موفقیت ایجاد شد'
 
-    # with open('./llm-answer/chat-messages.json', mode='w', encoding='utf-8') as output:
-    #     json.dump(prev_chat_data, output, ensure_ascii=False, indent=2)
+    title = llm_base_request(query)
+    if title == '':
+        title = query[0:15]
+        
+    chat_obj = {
+        'id' : chat_id, # str
+        'title' : title, # str
+        'user_id' : '',
+        'user_query' : query, # str
+        'model_key' : llm_model, # str
+        'retrived_passage' : result_passages_text, # str
+        'retrived_ref_ids' : result_passages_ids, # list[obj]
+        'prompt_type' : 'question-answer', # str
+        'retrived_duration' : retrive_duration, # str
+        'llm_duration' : llm_answer_duration, # str
+        'full_duration' : full_prompt_duration, # str
+        'time_create' : str(start_time), # str
+        'used_ref_ids' : used_refrences_in_answer, # list[str]
+        'prompt_answer' : llm_answer, # str
+        'status_text' : status_text, # str
+        'status' : True, # or False # bool
+    }
+    prev_chat_data = []    
+    with open('./llm-answer/chat-messages.json', mode='r', encoding='utf-8') as file:
+            prev_chat_data = json.load(file)
+            prev_chat_data.append(chat_obj)
+                
+    with open('./llm-answer/chat-messages.json', mode='w', encoding='utf-8') as output:
+        json.dump(prev_chat_data, output, ensure_ascii=False, indent=2)
+    
+    # save_result(chat_obj)
+    
+    # ایجاد آبجکت بازگشتی به فرانت  
+    # chat_obj.pop('retrived_passage')
+    # chat_obj.pop('prompt_type')
     
     return chat_obj
 
 @chatbot.post("/credit_refresh")
 def credit_refresh():
+    """
+    Returns remained credit
+    """
     url = "https://api.avalai.ir/user/credit"
     headers = {
     "Content-Type": "application/json",
@@ -415,21 +516,30 @@ def credit_refresh():
     with open('./llm-answer/credit.txt','w') as file:
         file.write(str(remained_credit.json()['remaining_irt']))
     return str(remained_credit.json()['remaining_irt'])
+
+def create_chat_id():
+    date = str((datetime.datetime.now())).replace(' ','-').replace(':','').replace('.','-')
+    print('date  ', date )
+    chat_id = f'{date}-{random.randint(100000, 999999)}'
+    print('chat_id  ', chat_id )
+    return chat_id
+
+print('#'*19)
+print('-Chatbot is Ready!!!!!-')
+print('#'*19)
+
 # تعریف مدل داده‌ها برای درخواست‌های API
 class Query(BaseModel):
     query: str
-
-date = str((datetime.datetime.now())).replace(' ','-').replace(':','').replace('.','-')
-chat_id = f'{date}-{random.randint(100000, 999999)}'
-print('#'*19)
-print('-Chatbot is Ready!-')
-print('#'*19)
 # مسیر API برای اجرا کردن run_chatbot
 @chatbot.post("/run_chatbot")
-def chat(query: Query):
-    
+def run_chat(query: Query):
+    print('query  ', query )
+    chat_id = create_chat_id()
+    print('query.query  ', query.query )
     answer = run_chatbot(query.query, chat_id)
     credit_refresh()
+
     return {"answer": answer}
 
 # uvicorn src.app:app --reload

dockerfile

@@ -1,21 +1,5 @@
-FROM python:3.10.12
-
-RUN pip install cleantext==1.1.4
-RUN pip install elasticsearch7==7.17.12
-RUN pip install faiss_cpu==1.9.0
-RUN pip install fastapi==0.117.1
-RUN pip install hazm==0.10.0
-RUN pip install langchain_openai==0.3.33
-RUN pip install numpy==1.21.5
-RUN pip install openai==1.108.1
-RUN pip install pandas==2.3.2
-RUN pip install pydantic==2.11.9
-RUN pip install scikit_learn==1.7.2
-RUN pip install sentence_transformers==2.5.1
-RUN pip install torch==2.4.0
-RUN pip install transformers==4.55.1
-#RUN pip install torch==2.1.2
-
+FROM qchat_base:1.0.0
+RUN pip install uvicorn[standard]
 
 WORKDIR /src/app 
 
@@ -23,5 +7,5 @@ COPY . /src/app
 
 EXPOSE 80
 
-CMD [ "uvicorn","chatbot:chatbot","--reload","--port","80" ]
+CMD [ "uvicorn","chatbot:chatbot","--reload","--port","80","--host=0.0.0.0"]
 

embedder_sbert_qavanin_285k.py

@@ -30,7 +30,7 @@ from transformers import AutoTokenizer
 from sklearn.decomposition import PCA
 from sklearn.manifold import TSNE
 from sklearn.metrics.pairwise import cosine_similarity
-from normalizer import cleaning
+#from normalizer import cleaning
 try:
     from elastic_helper import ElasticHelper
 except Exception as error:
@@ -43,8 +43,8 @@ except Exception as error:
 # from plotly.subplots import make_subplots
 
 # Persian text processing
-import hazm
-from hazm import Normalizer, word_tokenize, POSTagger
+# import hazm
+# from hazm import Normalizer, word_tokenize, POSTagger
 
 # Configure logging
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
@@ -67,7 +67,7 @@ class PersianVectorAnalyzer:
         """
         self.model_name = model_name
         self.model = None
-        self.normalizer = Normalizer()
+        #self.normalizer = Normalizer()
         self.stop_words = self._load_persian_stop_words()
         self.key_words = [
             "خدا", "بنده", "جهاد", "ولی", "زکات",
@@ -206,7 +206,7 @@ class PersianVectorAnalyzer:
         """
 
         # Normalize text
-        text = self.normalizer.normalize(text)
+        #text = self.normalizer.normalize(text)
 
         # Remove extra whitespace
         text = re.sub(r'\s+', ' ', text)

llm-answer/credit.txt

@@ -1 +1 @@
-6085.11
+197951.1

llm-answer/error-in-llm.txt

@@ -0,0 +1,5 @@
+
+
+query: برای حمایت از ازدواج جوانان و تشکیل خانواده جهت افزایش جمعیت ، چه مواردی پیش بینی شده است؟
+error:Error code: 400 - {'error': {'message': 'Developer instruction is not enabled for this model. Please use a different model that supports developer instructions. Please contact support at support@avalai.ir and include the request ID 01998c28-4ccb-7bc3-97a7-0403baa6ed35 in your email if you believe this is an error.', 'type': 'invalid_request', 'param': None, 'code': 'invalid_argument', 'request_id': '01998c28-4ccb-7bc3-97a7-0403baa6ed35'}} 
+-------------------------------

normalizer.py

@@ -1,4 +1,4 @@
-import hazm
+#import hazm
 from cleantext import clean
 import re
 
@@ -7,7 +7,7 @@ def cleanhtml(raw_html):
     cleantext = re.sub(cleanr, '', raw_html)
     return cleantext
 
-normalizer = hazm.Normalizer()
+#normalizer = hazm.Normalizer()
 wierd_pattern = re.compile("["
         u"\U0001F600-\U0001F64F"  # emoticons
         u"\U0001F300-\U0001F5FF"  # symbols & pictographs
@@ -64,7 +64,7 @@ def cleaning(text):
     text = cleanhtml(text)
     
     # normalizing
-    text = normalizer.normalize(text)
+    #text = normalizer.normalize(text)
     
     # removing wierd patterns
     text = wierd_pattern.sub(r'', text)

run_docker.bash

@@ -0,0 +1 @@
+sudo docker run --name qachat -p 80:80  -v /home/sabr/rag_qavanin_api/:/src/app/ -v /home/sabr/rag_qavanin_api/qavanin-faiss/:/src/app/qavanin-faiss/ -v /home/sabr/rag_qavanin_api/llm-answer/:/src/app/llm-answer/ -v /home/sabr/MODELS:/src/MODELS -it  --restart unless-stopped qachat:1.0.0