fix sampling

app.py

@@ -55,6 +55,12 @@ def inference(lm_model_name: str, cls_model_name: str, prompt: str, fp16: bool =
     caif_sampler = load_sampler(cls_model_name=cls_model_name, lm_tokenizer=lm_tokenizer)
     generator.set_caif_sampler(caif_sampler)
     ordinary_sampler = TopKWithTemperatureSampler()
+    kwargs = {
+        "top_k": 20,
+        "temperature": 1.0,
+        "top_k_classifier": 100,
+        "classifier_weight": 5,
+    }
     generator.set_ordinary_sampler(ordinary_sampler)
     if device == "cpu":
         autocast = torch.cpu.amp.autocast
@@ -67,7 +73,8 @@ def inference(lm_model_name: str, cls_model_name: str, prompt: str, fp16: bool =
             max_length=20,
             caif_period=1,
             caif_tokens_num=100,
-            entropy=3.2
+            entropy=3.2,
+            **kwargs
         )
     return sequences[0]