fix attention mask warning

i literally cannot test this code but normally it should work

modeling_llamavision.py

@@ -105,8 +105,17 @@ class Llamavision(PreTrainedModel):
 
         with torch.no_grad():
             inputs_embeds = self.input_embeds(prompt, image_embeds, tokenizer)
+            
+            attention_mask = torch.ones(
+                inputs_embeds.shape[:2],
+                dtype=torch.long,
+                device=inputs_embeds.device
+            )
+            
             output_ids = self.text_model.generate(
-                inputs_embeds=inputs_embeds, **generate_config
+                inputs_embeds=inputs_embeds,
+                attention_mask=attention_mask,
+                **generate_config
             )
 
         return tokenizer.batch_decode(output_ids, skip_special_tokens=True)