Update app.py

update activation of top_p, and gpu required time

app.py

@@ -63,15 +63,15 @@ LANGUAGES = {
 loaded_models = {}
 loaded_tokenizers = {}
 
+
 @spaces.GPU(duration=60)
 def load_model_and_tokenizer(model_key):
     if model_key not in loaded_models:
         model_info = MODELS[model_key]
-        device = "cuda"
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         model = AutoModelForCausalLM.from_pretrained(
             model_info["model_name"],
-            token=HF_TOKEN,
-            torch_dtype=torch.float16
+            token=HF_TOKEN
         ).to(device)
         loaded_models[model_key] = model
 
@@ -84,26 +84,31 @@ def load_model_and_tokenizer(model_key):
             tokenizer.pad_token = tokenizer.eos_token
         loaded_tokenizers[model_key] = tokenizer
 
-@spaces.GPU(duration=140)
+
+@spaces.GPU(duration=120)
 def generate_text(model_choice, prompt, max_length, temperature, top_p, do_sample):
     load_model_and_tokenizer(model_choice)
 
     model = loaded_models[model_choice]
     tokenizer = loaded_tokenizers[model_choice]
-    device = "cuda"
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
     inputs = tokenizer(prompt, return_tensors="pt", truncation=True, padding=True).to(device)
 
-    outputs = model.generate(
-        input_ids=inputs["input_ids"],
-        attention_mask=inputs["attention_mask"],
-        max_length=max_length,
-        temperature=temperature,
-        top_p=top_p,
-        repetition_penalty=1.2,
-        no_repeat_ngram_size=2,
-        do_sample=do_sample,
-    )
+    generation_kwargs = {
+        "input_ids": inputs["input_ids"],
+        "attention_mask": inputs["attention_mask"],
+        "max_length": max_length,
+        "temperature": temperature,
+        "repetition_penalty": 1.2,
+        "no_repeat_ngram_size": 2,
+        "do_sample": do_sample,
+    }
+
+    if do_sample:
+        generation_kwargs["top_p"] = top_p
+
+    outputs = model.generate(**generation_kwargs)
 
     generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
@@ -136,7 +141,7 @@ def update_language(selected_language):
     )
 
 
-@spaces.GPU(duration=140)
+@spaces.GPU(duration=120)
 def wrapped_generate_text(model_choice, prompt, max_length, temperature, top_p, do_sample):
     return generate_text(model_choice, prompt, max_length, temperature, top_p, do_sample)
 
@@ -215,6 +220,12 @@ with gr.Blocks() as iface:
                  do_sample_checkbox, generate_button, output_text]
     )
 
+    do_sample_checkbox.change(
+        fn=lambda do_sample: gr.update(visible=do_sample),
+        inputs=[do_sample_checkbox],
+        outputs=[top_p_slider]
+    )
+
     generate_button.click(
         fn=wrapped_generate_text,
         inputs=[