Update app.py

app.py

@@ -7,13 +7,14 @@ from dl_hf_model import dl_hf_model
 
 params = {
         "max_new_tokens":512,
-        "stop":["<end>" ,"<|endoftext|>"],
+        "stop":["<end>" ,"<|endoftext|>","["],
         "temperature":0.7,
         "top_p":0.8,
         "stream":True,
         "batch_size": 8}
 
-url = "https://huggingface.co/Aspik101/trurl-2-13b-GGML/blob/main/trurl-2-13b.ggmlv3.q8_0.bin" 
+
+#url = "https://huggingface.co/Aspik101/trurl-2-7b-GGML/blob/main/trurl-2-7b.ggmlv3.q8_0.bin"
 #model_loc, file_size = dl_hf_model(url)
 llm = AutoModelForCausalLM.from_pretrained("Aspik101/trurl-2-13b-GGML", model_type="llama")
 
@@ -25,11 +26,19 @@ with gr.Blocks() as demo:
     def user(user_message, history):
         return "", history + [[user_message, None]]
 
+    def parse_history(hist):
+        history_ = ""
+        for q, a in hist:
+            history_ += f"<user>: {q } \n"
+            if a:
+                history_ += f"<assistant>: {a} \n"
+        return history_
+
     def bot(history):
-        print(history)
-        stream = llm(prompt =  f"Jesteś AI assystentem. Odpowiadaj po polsku. <user>: {history}. <assistant>:", **params)
-        #stream = llm(prompt =  f"{history}", **params)
-        #stream = llm(prompt =  f"Jesteś AI assystentem. Odpowiadaj po polsku. {history}.", **params)
+        print("history: ",history)
+        prompt =  f"Jesteś AI assystentem. Odpowiadaj po polsku. {parse_history(history)}. <assistant>:"
+        print("prompt: ",prompt)
+        stream = llm(prompt, **params)
         history[-1][1] = ""
         answer_save = ""
         for character in stream:
@@ -38,8 +47,7 @@ with gr.Blocks() as demo:
             time.sleep(0.005)
             yield history
 
-        print(answer_save)
-
+        print("answer_save: ",answer_save)
     msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
         bot, chatbot, chatbot
     )