Update

app.py

@@ -18,6 +18,8 @@ def run(
     message: dict,
     history: list[dict],
     max_new_tokens: int = 256,
+    temperature: float = 0.7,
+    repetition_penalty: float = 1.2,
 ) -> Iterator[str]:
     if not history and len(message["files"]) == 0:
         gr.Warning("Please upload an image.")
@@ -53,8 +55,9 @@ def run(
         inputs,
         streamer=streamer,
         max_new_tokens=max_new_tokens,
-        temperature=0,
-        do_sample=False,
+        temperature=temperature,
+        do_sample=temperature > 0.0,
+        repetition_penalty=repetition_penalty,
         stopping_criteria=processor.get_stopping_criteria(["\n###"]),
     )
     t = Thread(target=model.generate, kwargs=generate_kwargs)
@@ -80,7 +83,11 @@ demo = gr.ChatInterface(
     type="messages",
     multimodal=True,
     textbox=gr.MultimodalTextbox(file_types=["image"], file_count="single"),
-    additional_inputs=[gr.Slider(label="Max new tokens", minimum=10, maximum=1024, step=1, value=256)],
+    additional_inputs=[
+        gr.Slider(label="Max new tokens", minimum=10, maximum=1024, step=1, value=512),
+        gr.Slider(label="Temperature", minimum=0.0, maximum=1.0, step=0.1, value=0.7),
+        gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.1, value=1.2),
+    ],
     examples=examples,
     title="sbintuitions/sarashina2-vision-14b",
     cache_examples=False,