add install.sh

app.py

@@ -76,12 +76,12 @@ def respond(
 
     model_inputs = tokenizer([text], return_tensors="pt", return_token_type_ids=False).to(model.device)
 
-    print(f"max_new_tokens={max_tokens}, temperature={temperature}, top_p={top_p}")
+    print(f"max_new_tokens={max_tokens}")
     model_inputs.update(
         dict(max_new_tokens=max_tokens,
                              streamer = streamer,
-#                             temperature = 0.7,
-#                             top_p = 1,
+                             temperature = 0.7,
+                             top_p = 1,
                              presence_penalty = 1.5,
              )
     )