changed to mistral model

src/chatbot.py

@@ -12,11 +12,14 @@ import os
 #load_dotenv(find_dotenv())
 
 
+
 embeddings = HuggingFaceEmbeddings(model_name="paraphrase-multilingual-MiniLM-L12-v2")
 llm = HuggingFaceHub(
     # Try different model here
-    # repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
-    repo_id="CohereForAI/c4ai-command-r-v01",
+    repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
+    # repo_id="CohereForAI/c4ai-command-r-v01", # too large 69gb
+    # repo_id="CohereForAI/c4ai-command-r-v01-4bit", # too large 22 gb
+    # repo_id="meta-llama/Meta-Llama-3-8B", # too large 16 gb
     task="text-generation",
     model_kwargs={
         "max_new_tokens": 512,
@@ -24,6 +27,8 @@ llm = HuggingFaceHub(
         "temperature": 0.1,
         "repetition_penalty": 1.03,
         }
+        #,huggingfacehub_api_token
+
 )
 # To Do: Experiment with different templates replying in german or english depending on the input language
 prompt1 = ChatPromptTemplate.from_template("""<s>[INST] 
@@ -55,7 +60,7 @@ db = get_vectorstore(embeddings=embeddings, folder_path=folder_path, index_name=
 
 def chatbot(message, history, db=db, llm=llm, prompt=prompt2):
     raw_response = RAG(llm=llm, prompt=prompt, db=db, question=message)
-    response = raw_response['answer']#.split("Antwort: ")[1]
+    response = raw_response['answer'].split("Antwort: ")[1]
     return response  
 
 # Retrieve speech contents based on keywords