ui: bump max_new_tokens slider to 512 (default 128) — fits ZeroGPU 180s budget

app.py

@@ -46,7 +46,8 @@ PARAPHRASE_CEILING_CEN = 0.799
 RHO_DENOM = PARAPHRASE_CEILING_CEN - RANDOM_FLOOR_CEN
 
 MAX_INPUT_TOKENS = 256
-MAX_NEW_TOKENS_DEFAULT = 64
+MAX_NEW_TOKENS_DEFAULT = 128
+MAX_NEW_TOKENS_LIMIT = 512  # ZeroGPU 180s budget: ~512 tok × BoN=16 fits
 
 _state: dict[str, Any] = {}
 
@@ -359,7 +360,7 @@ def build_app() -> gr.Blocks:
                         with gr.Row():
                             n_samples = gr.Slider(1, 16, value=4, step=1,
                                                   label="Best-of-N samples")
-                            max_new = gr.Slider(16, 128, value=MAX_NEW_TOKENS_DEFAULT,
+                            max_new = gr.Slider(16, MAX_NEW_TOKENS_LIMIT, value=MAX_NEW_TOKENS_DEFAULT,
                                                 step=16, label="Max new tokens")
                             temperature = gr.Slider(0.3, 1.5, value=0.9, step=0.05,
                                                     label="Temperature")
@@ -389,7 +390,7 @@ def build_app() -> gr.Blocks:
                         text_b = gr.Textbox(label="Passage B", lines=4)
                         alpha = gr.Slider(0.0, 1.0, value=0.5, step=0.05,
                                           label="α  (0 = A, 1 = B)")
-                        max_new_a = gr.Slider(16, 128, value=MAX_NEW_TOKENS_DEFAULT,
+                        max_new_a = gr.Slider(16, MAX_NEW_TOKENS_LIMIT, value=MAX_NEW_TOKENS_DEFAULT,
                                               step=16, label="Max new tokens")
                         go_a = gr.Button("Verbalize mix", variant="primary")
                     with gr.Column():