pinning transformer version and added missin parameter

Author: Ubuntu

pyproject.toml

@@ -159,6 +159,9 @@ full = [
     "xformers==0.0.30",
     "stable-fast-pruna==1.0.7",
 ]
+zipar = [
+    "transformers==4.54.0",
+]
 dev = [
     "wget",
     "python-dotenv",

src/pruna/algorithms/zipar/utils.py

@@ -544,6 +544,7 @@ def prepare_input_and_cache(self, input_ids, model_kwargs, generation_config, de
             # batch_size should account for both conditional/unconditional input; hence multiplied by 2
             batch_size=batch_size * 2,
             max_cache_len=self.num_image_tokens + seq_len,
+            device=device,
             model_kwargs=model_kwargs,
         )