Update app.py

app.py

@@ -1,12 +1,8 @@
 import gradio as gr
-import time
-import ctypes #to run on C api directly 
-import llama_cpp
+from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
-from huggingface_hub import hf_hub_download #load from huggingfaces 
 
-
-llm = Llama(model_path= hf_hub_download(repo_id="TheBloke/airoboros-l2-13b-gpt4-m2.0-GGML", filename="airoboros-l2-13b-gpt4-m2.0.ggmlv3.q6_K.bin"), n_ctx=2048) #download model from hf/ n_ctx=2048 for high ccontext length
+llm = Llama(model_path=hf_hub_download(repo_id="TheBloke/airoboros-l2-13b-gpt4-m2.0-GGML", filename="airoboros-l2-13b-gpt4-m2.0.ggmlv3.q6_K.bin"), n_ctx=2048)
 
 history = []
 history.append(["Hi there!", "Hello, how can I help you?"])
@@ -19,7 +15,7 @@ def generate_text(input_text, history):
     response = output['choices'][0]['text']
     history.append([input_text, response])
     
-   return "", history
+    return "", history
 
 with gr.Blocks() as demo:
     chatbot = gr.Chatbot()
@@ -30,3 +26,10 @@ with gr.Blocks() as demo:
 
 demo.launch()
 
+
+
+
+
+
+
+