Update evo_plugin_example.py

evo_plugin_example.py

@@ -12,17 +12,20 @@ class _HFSeq2SeqGenerator:
 
     @torch.no_grad()
     def generate(self, prompt: str, max_new_tokens: int = 200, temperature: float = 0.4) -> str:
-        # Seq2Seq models generate responses that follow instructions better than GPT-2 here.
         inputs = self.tok(prompt, return_tensors="pt").to(self.device)
         out = self.model.generate(
             **inputs,
             max_new_tokens=int(max_new_tokens),
             do_sample=temperature > 0.0,
             temperature=float(max(0.01, temperature)),
-            top_p=0.95,
+            top_p=0.9,
+            num_beams=4,                 # beam search makes it less echo-y
+            early_stopping=True,
+            no_repeat_ngram_size=3,      # avoid repeating phrases
         )
         return self.tok.decode(out[0], skip_special_tokens=True).strip()
 
+
 def load_model():
     # The app calls this to obtain a generator instance.
     return _HFSeq2SeqGenerator()