Fix gpt_bigcode generation

optimum/intel/generation/modeling.py

@@ -179,13 +179,22 @@ def _reorder_cache(
         """
         if self.config.model_type == "bloom":
             return self._reorder_cache_bloom(past_key_values, beam_idx)
+        elif self.config.model_type == "gpt_bigcode":
+            return self._reorder_cache_gpt_bigcode(past_key_values, beam_idx)
 
         # from transformers.models.gpt2.modeling_gpt2.GPT2LMHeadModel._reorder_cache
         return tuple(
             tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
             for layer_past in past_key_values
         )
 
+    # Copied from transformers.models.gpt_bigcode.modeling_gpt_bigcode.GPTBigCodeForCausalLM._reorder_cache
+    @staticmethod
+    def _reorder_cache_gpt_bigcode(
+        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        return tuple(layer_past.index_select(0, beam_idx.to(layer_past.device)) for layer_past in past_key_values)
+
     # Copied from transformers.models.bloom.modeling_bloom.BloomForCausalLM._reorder_cache
     def _reorder_cache_bloom(
         self, past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor

tests/generation/test_modeling.py

@@ -58,7 +58,7 @@ class ModelingIntegrationTest(unittest.TestCase):
         "mistral",
         "llama",
         "llama2",
-        # "gpt_bigcode",
+        "gpt_bigcode",
     )
 
     GENERATION_LENGTH = 100

tests/ipex/test_inference.py

@@ -65,7 +65,7 @@ class IPEXIntegrationTest(unittest.TestCase):
         "gptj",
         "gpt2",
         "gpt_neo",
-        # "gpt_bigcode",
+        "gpt_bigcode",
         "llama",
         "llama2",
         "opt",