added telegram bot

.env.example

@@ -83,15 +83,15 @@ TOKENIZERS_PARALLELISM=true
 
 # env variables for ingesting source PDF files
 CHUNCK_SIZE=1024
-CHUNK_OVERLAP=512
+CHUNK_OVERLAP=64
 SOURCE_PATH="data/pdfs/smu_lib_html/"
 
 # Index for SMU LibBot PDF files - chunk_size=1024 chunk_overlap=512
 FAISS_INDEX_PATH="data/smu_lib_index/"
 
 # telegram bot
+RUN_TELEGRAM_BOT=true
 TELEGRAM_API_TOKEN=
-CHAT_API_URL=http://localhost:8080/chat_sync
 
 # template for env/tgi.conf
 export PORT=64300

app.py

@@ -11,6 +11,12 @@ from app_modules.init import app_init
 from app_modules.llm_chat_chain import ChatChain
 from app_modules.utils import print_llm_response, remove_extra_spaces
 
+if os.environ.get("RUN_TELEGRAM_BOT") == "true":
+    from telegram_bot import start_telegram_bot
+
+    start_telegram_bot()
+    exit(0)
+
 llm_loader, qa_chain = app_init()
 
 show_param_settings = os.environ.get("SHOW_PARAM_SETTINGS") == "true"

app_modules/init.py

@@ -22,61 +22,69 @@ load_dotenv(found_dotenv, override=False)
 # Constants
 init_settings()
 
+llm_loader = None
+qa_chain = None
 
-def app_init(initQAChain: bool = True):
-    # https://github.com/huggingface/transformers/issues/17611
-    os.environ["CURL_CA_BUNDLE"] = ""
 
-    llm_model_type = os.environ.get("LLM_MODEL_TYPE")
-    n_threds = int(os.environ.get("NUMBER_OF_CPU_CORES") or "4")
+def app_init(initQAChain: bool = True):
+    global llm_loader
+    global qa_chain
+    if llm_loader == None:
+        # https://github.com/huggingface/transformers/issues/17611
+        os.environ["CURL_CA_BUNDLE"] = ""
+
+        llm_model_type = os.environ.get("LLM_MODEL_TYPE")
+        n_threds = int(os.environ.get("NUMBER_OF_CPU_CORES") or "4")
+
+        hf_embeddings_device_type, hf_pipeline_device_type = get_device_types()
+        print(f"hf_embeddings_device_type: {hf_embeddings_device_type}")
+        print(f"hf_pipeline_device_type: {hf_pipeline_device_type}")
+
+        if initQAChain:
+            hf_embeddings_model_name = (
+                os.environ.get("HF_EMBEDDINGS_MODEL_NAME") or "hkunlp/instructor-xl"
+            )
 
-    hf_embeddings_device_type, hf_pipeline_device_type = get_device_types()
-    print(f"hf_embeddings_device_type: {hf_embeddings_device_type}")
-    print(f"hf_pipeline_device_type: {hf_pipeline_device_type}")
+            index_path = os.environ.get("FAISS_INDEX_PATH") or os.environ.get(
+                "CHROMADB_INDEX_PATH"
+            )
+            using_faiss = os.environ.get("FAISS_INDEX_PATH") is not None
 
-    if initQAChain:
-        hf_embeddings_model_name = (
-            os.environ.get("HF_EMBEDDINGS_MODEL_NAME") or "hkunlp/instructor-xl"
-        )
+            start = timer()
+            embeddings = HuggingFaceInstructEmbeddings(
+                model_name=hf_embeddings_model_name,
+                model_kwargs={"device": hf_embeddings_device_type},
+            )
+            end = timer()
 
-        index_path = os.environ.get("FAISS_INDEX_PATH") or os.environ.get(
-            "CHROMADB_INDEX_PATH"
-        )
-        using_faiss = os.environ.get("FAISS_INDEX_PATH") is not None
+            print(f"Completed in {end - start:.3f}s")
 
-        start = timer()
-        embeddings = HuggingFaceInstructEmbeddings(
-            model_name=hf_embeddings_model_name,
-            model_kwargs={"device": hf_embeddings_device_type},
-        )
-        end = timer()
+            start = timer()
 
-        print(f"Completed in {end - start:.3f}s")
+            print(
+                f"Load index from {index_path} with {'FAISS' if using_faiss else 'Chroma'}"
+            )
 
-        start = timer()
+            if not os.path.isdir(index_path):
+                raise ValueError(f"{index_path} does not exist!")
+            elif using_faiss:
+                vectorstore = FAISS.load_local(index_path, embeddings)
+            else:
+                vectorstore = Chroma(
+                    embedding_function=embeddings, persist_directory=index_path
+                )
 
-        print(
-            f"Load index from {index_path} with {'FAISS' if using_faiss else 'Chroma'}"
-        )
+            end = timer()
 
-        if not os.path.isdir(index_path):
-            raise ValueError(f"{index_path} does not exist!")
-        elif using_faiss:
-            vectorstore = FAISS.load_local(index_path, embeddings)
-        else:
-            vectorstore = Chroma(
-                embedding_function=embeddings, persist_directory=index_path
-            )
+            print(f"Completed in {end - start:.3f}s")
 
+        start = timer()
+        llm_loader = LLMLoader(llm_model_type)
+        llm_loader.init(
+            n_threds=n_threds, hf_pipeline_device_type=hf_pipeline_device_type
+        )
+        qa_chain = QAChain(vectorstore, llm_loader) if initQAChain else None
         end = timer()
-
         print(f"Completed in {end - start:.3f}s")
 
-    start = timer()
-    llm_loader = LLMLoader(llm_model_type)
-    llm_loader.init(n_threds=n_threds, hf_pipeline_device_type=hf_pipeline_device_type)
-    qa_chain = QAChain(vectorstore, llm_loader) if initQAChain else None
-    end = timer()
-    print(f"Completed in {end - start:.3f}s")
-
     return llm_loader, qa_chain

data/questions.txt

@@ -1,6 +1,4 @@
-What are the library opening hours?
-I'm an undergrad. How many books can I borrow from libraries?
+How many items can a doctoral student borrow?
 Can you list some of recommended resources on generative AI?
-Hi, is it necessary to book a terminal first before being able to use the bloomberg computer in the library? or can i just show up?
-Hi, I am an alumni of SMU (batch of 2018). I wanted to enquire for SMU Alumni rates for access to library resources (databases, investment studio) etc
 I've overdue fine of $4.00. Could you advise on how I can go about paying the fine?
+Hi, I am an alumni of SMU (batch of 2018). I wanted to enquire for SMU Alumni rates for access to library resources (databases, investment studio) etc

telegram_bot.py

@@ -14,7 +14,9 @@ from telegram.ext import (
     filters,
 )
 
-from app_modules.init import *
+from app_modules.init import app_init
+
+llm_loader, qa_chain = app_init()
 
 ctx = ssl.create_default_context()
 ctx.set_ciphers("DEFAULT")
@@ -32,7 +34,6 @@ if __version_info__ < (20, 0, 0, "alpha", 1):
     )
 
 TOKEN = os.getenv("TELEGRAM_API_TOKEN")
-ENDPOINT = os.getenv("CHAT_API_URL")
 
 
 # Define a few command handlers. These usually take the two arguments update and
@@ -55,16 +56,11 @@ async def chat_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> No
     print(update)
     tic = time.perf_counter()
     try:
-        message = {
-            "question": update.message.text,
-            "chat_id": update.message.chat.id,
-        }
-        print(message)
-        x = requests.post(ENDPOINT, json=message).json()
-        temp = time.perf_counter()
-        print(f"Received response in {temp - tic:0.4f} seconds")
-        print(x)
-        result = x["result"]
+        result = qa_chain.call_chain(
+            {"question": update.message.text, "chat_history": []}, None
+        )
+
+        result = result["answer"]
         print(result)
         await update.message.reply_text(result[0:8192])
         toc = time.perf_counter()
@@ -89,7 +85,3 @@ def start_telegram_bot() -> None:
     )
 
     application.run_polling()
-
-
-if __name__ == "__main__":
-    start_telegram_bot()