Update app.py

app.py

@@ -11,11 +11,7 @@ import spaces
 import time
 import subprocess
 
-subprocess.run(
-    "pip install flash-attn --no-build-isolation",
-    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
-    shell=True,
-)
+
 
 token = os.environ["HF_TOKEN"]
 
@@ -58,7 +54,7 @@ def chat(message, history, temperature, do_sample, max_tokens):
         model_inputs,
         streamer=streamer,
         max_new_tokens=max_tokens,
-        do_sample=True,
+        do_sample=True,r
         temperature=temperature,
         eos_token_id=terminators,
     )