Update app.py

app.py

@@ -4,11 +4,18 @@ from llama_cpp import Llama
 llm = Llama.from_pretrained("bartowski/starcoder2-15b-instruct-v0.1-GGUF", filename="starcoder2-15b-instruct-v0.1-Q6_K.gguf", n_ctx=16384, n_gpu_layers=99)
 
 def code_block(txt, mdn):
-    return txt + (res := llm(
-        prompt=txt, 
-        stop=['###', '///', '\n\n\n\n'],
-    )['choices'][0]['text']), mdn+"\n\n"+txt+res
-# COnvert this to gr.Blocks()
+    txt = ""
+    mdn = mdn + "\n\n```javascript\n"
+    res = llm(
+        prompt=txt,
+        stop=['###', '///', '\n\n\n']
+        stream=True
+    )
+    for chk in res:
+        txt += chk['choices'][0]['text']
+        yield txt, mdn + txt + "\n````\n"
+    return txt, mdn+txt+"\n```\n"
+    
 
 with gr.Blocks() as apps:
     with gr.Row():