Set a minimal repetition penalty for TinyMistral-248M-SFT-v4

app.py

@@ -31,7 +31,7 @@ def generate(
     elif model_name == "Felladrin/Llama-160M-Chat-v1":
         outputs = pipe(prompt, max_length=1024, use_cache=True, penalty_alpha=0.5, top_k=4, repetition_penalty=1.01)
     elif model_name == "Felladrin/TinyMistral-248M-SFT-v4":
-        outputs = pipe(prompt, max_length=1024, use_cache=True, penalty_alpha=0.5, top_k=5)
+        outputs = pipe(prompt, max_length=1024, use_cache=True, penalty_alpha=0.5, top_k=5, repetition_penalty=1.001)
     else:
         outputs = pipe(prompt, max_length=1024, do_sample=True, temperature=0.8, top_k=40, top_p=0.95, repetition_penalty=1.1)