Drop cache_position kwarg (removed from create_causal_mask in transformers 5.9)

On transformers 5.9, `create_causal_mask` no longer accepts a `cache_position` argument:

TypeError: create_causal_mask() got an unexpected keyword argument 'cache_position'

On 5.8 the parameter is present but documented as 'not used anymore but kept for BC'. Dropping the kwarg (and the now-unused `dummy_cache_position` computation) is compatible with both versions.

modeling.py

@@ -54,16 +54,11 @@ class PPLXQwen3Model(Qwen3Model):
             inputs_embeds = self.embed_tokens(input_ids)
             input_ids = None
 
-        # We construct a dummy tensor imitating initial positions
-        dummy_cache_position = torch.arange(
-            inputs_embeds.shape[1], device=inputs_embeds.device, dtype=torch.long
-        )
         attention_mask = {
             "full_attention": create_causal_mask(
                 config=self.config,
                 inputs_embeds=inputs_embeds,
                 attention_mask=attention_mask,
-                cache_position=dummy_cache_position,
                 past_key_values=None,
                 position_ids=position_ids,
                 or_mask_function=bidirectional_mask_function(attention_mask),