Update app.py

app.py

@@ -21,9 +21,9 @@ model2 = AutoModelForCausalLM.from_pretrained(model_name)
 
 provider = TransformersProvider(model2, tokenizer, device)
 strategy = CreativeWritingStrategy(provider, 
-                                   top_p_flat = 0.6,
-                                   top_k_threshold_flat = 10,
-                                   eos_penalty = 0.7)
+                                   top_p_flat = 0.65,
+                                   top_k_threshold_flat = 9,
+                                   eos_penalty = 0.8)
 creative_sampler = BacktrackSampler(strategy, provider)
 
 def create_chat_template_messages(history, prompt):
@@ -38,9 +38,9 @@ def create_chat_template_messages(history, prompt):
 @spaces.GPU
 def generate_responses(prompt, history):
     messages = create_chat_template_messages(history, prompt)
-    wrapped_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_special_tokens=False, add_generation_prompt=True)
-
-    inputs = tokenizer.encode(wrapped_prompt, add_special_tokens=True, return_tensors="pt").to("cuda")
+    wrapped_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    #it already has special tokens from wrapped_prompt
+    inputs = tokenizer.encode(wrapped_prompt, add_special_tokens=False, return_tensors="pt").to("cuda")
 
     async def custom_sampler_task():
         generated_list = []