repetition penalty tokens

moondream.py

@@ -63,8 +63,8 @@ class Moondream(PreTrainedModel):
         prompt,
         tokenizer,
         eos_text="<END>",
-        max_new_tokens=128,
-        repetition_penalty=1.25,
+        max_new_tokens=256,
+        repetition_penalty=1.15,
         **kwargs,
     ):
         eos_tokens = tokenizer(eos_text, add_special_tokens=False)[0].ids
@@ -92,8 +92,8 @@ class Moondream(PreTrainedModel):
         question,
         tokenizer,
         chat_history="",
-        repetition_penalty=1.25,
-        max_new_tokens=128,
+        repetition_penalty=1.15,
+        max_new_tokens=256,
         result_queue=None,
         **kwargs,
     ):
@@ -120,8 +120,8 @@ class Moondream(PreTrainedModel):
         images,
         prompts,
         tokenizer,
-        repetition_penalty=1.25,
-        max_new_tokens=128,
+        repetition_penalty=1.15,
+        max_new_tokens=256,
         **kwargs,
     ):
         eos_tokens = tokenizer("<END>", add_special_tokens=False)[0].ids